State of the Art in Intelligent Distribution Networks and Definition of Application Scenarios and Research Problems
This deliverable will explore the initial literature study and research directions for WP4 of the ADVANTAGE project. This WP will investigate the creation of a truly intelligent power distribution network in the future. The first topic to studied is state estimation in distribution networks that are enabled with microgrids. The second topic will evaluate advanced management systems for distribution networks. Thirdly, load clustering concepts that can enable demand side management on a large scale will be considered. Finally, consideration will be given to the treatment, use and value of data within realistic smart grids.
Led by Aristides Kiprakis (WP4 Leader)
Achilleas Tsitsimelis (Early Stage Researcher)
Mirsad Cosovic (Early Stage Researcher)
Alexandros Kleidaras (Early Stage Researcher)
Anna Fragkioudaki (Early Stage Researcher)
John Thompson (Project Co-ordinator)
Table of contents:
1. Introduction to Work Package Four: Intelligent Distribution Networks
2. State Estimation in Micro Grid-Enabled Distribution Networks
2.1 Introduction – Electric Power Systems
2.2 State Estimation
2.3 State Estimation and Future Research Lines
2.4 Research Areas
3. Advanced Distribution Management System
3.2 Distributed State Estimation
4. Load Clustering for Large Scale Demand Side Management Optimisation
4.2 Demand Side Management Potential
4.3 Flexible Loads: Interruptible And Deferrable
4.4 Importance Of Clusters
5. Treatment, Use and Value of Smart Grid Data in a Real Environment
5.2 System State-based Methods
5.3 Game Theory-based Methods
5.4 Statistical-based Methods
6. Key Research Objectives for The Work Package
1. Introduction to Work Package Four: Intelligent Distribution Networks
The fourth work package of the ADVANTAGE project contributes advancements towards intelligent distribution network that will enable and maintain efficient distribution management, load clustering, demand side management, distributed micro grid control, and two-way communication with customers and micro grids through Advanced Metering Infrastructure (AMI). The Early Stage Researchers involved in WP4 are Mirsad Cosovic (Schneider Electric DMS), Anna Fragkioudaki (Univ. of Seville), Alexandros Kleidaras (Univ. of Edinburgh) and Achilleas Tsitsimelis (CTTC); the work package is led by Dr Aristides Kiprakis (Univ. of Edinburgh).
Current distribution networks are mainly passive systems that provide only unidirectional energy flow from transmission network to users. In the future smart grid, distribution networks should allow for bi-directional flows both of energy and data in order to achieve supply/demand balance, thus contributing to the grid stability. It is envisaged that this WP will develop advanced state estimation solutions for such networks to monitor their stability, along with new advanced distribution management systems to operate the network. Further, it develops enabling techniques for such networks to handle large volumes of offered information, in order to meet operating challenges related to system responsiveness and scalability. The ESRs are thus collaborating on (1) network monitoring and management solutions and (2) techniques to process large volumes of data for the future smart grid.
The first task of this research is to investigate and report the current state of the art in intelligent distribution networks. We specifically focus in the following four topics.
State estimation in micro grid-enabled distribution networks (Achilleas Tsitsimelis, ESR10). The ultimate objective of the power system is to deliver electrical energy to the consumer safely, reliably, economically, and with good quality. An emerging concept that will help the above objective to be achieved is the micro grid. Optimising the control and management of microgrids, requires a good knowledge of the state of the grid. Section 2 presents the underlying principles and current research on this topic.
Advanced Distribution Management Systems (Mirsad Cosovic, ESR11). Section 3 gives a brief overview of the theoretical and practical methods of power system state estimation. The two main concepts used for power system state estimation are hierarchical and fully distributed state estimation. These are broken down to a number of sub categories such as decentralized state estimation, distributed state estimation and fully distributed state estimation, however there is no clear distinction between these terms. This section presents some of the seminal research in this area.
Load Clustering for Large Scale Demand Side Management Optimisation (Alexandros Kleidaras, ESR12). The smart grid of the near future will include a number of components that are either completely new, or are now operated and managed in a completely different way, allowing the provision of ancillary services to the grid. Highly dispersed micro-generation including renewables, energy storage, Vehicle to Grid (V2G) enabled electric car charging as well as real- and near real-time management of flexible demand are going to become tools for the optimisation of the smart grid operation. Section 4 presents the background and the current state of the art in this field of research.
Treatment, Use and Value of Smart Grid Data in a Real Environment (Anna Fragkioudaki, ESR13). Power losses occurring within the process of energy delivery to the consumer are one of the principal problems affecting the efficiency and security of the power distribution networks. Losses can be distinguished into technical (due to naturally occurring phenomena in the power system) and non-technical losses (due to administrative problems or unauthorized energy consumption). In both cases, efficient management of the information flow from the consumers’ smart meters and other data collection points in the power system, and the system operator, is expected to substantially reduce the uncertainties and help identify the points of power loss. The final section of this report presents the research in this area, categorised in system state, game theoretic and statistical methods.
2. State Estimation in Micro Grid-Enabled Distribution Networks
2.1 Introduction – ELECTRIC POWER SYSTEMS
The electric power grid is divided to three main parts. The generating sector, where energy is converted from one of its basic forms, into electric energy. There follows the transmission grid, that interconnects the system with other power grids and transfers the main amount of energy from the first part to the distribution grid which is the final division of the network. This third part of the system, leads to the loads and customers and is connected to the previous level through the substations. Figure 1 shows the typical structure of a power grid .
Figure 1: Sample Power System structure 
The generating system provides the system with electric energy. The transmission and subtransmission systems are meshed networks. More specifically, there is more than one path from one point to another, thus increasing the reliability of the system. The transmission network is a high-voltage network, 138 to 765 kilovolts (kV), designed to deliver power over long distances, that is from generators to load points. The subtransmission network is a low-voltage network, 34 to 138 kV, whose purpose is to transport power over shorter distances from bulk power substations to distribution substations.
The distribution division includes the part below the subtransmission level and is divided into the primary part (Figure 2), that extends from the distribution substation to the distribution transformers and the secondary part that connects the customers to the distribution transformers. At the primary part, each substation contains a transformer to bring the voltage down, 4 to 34 kV, from the previous level and serves one or more feeders. At the secondary part the usual voltages are 240, 208 or 120 V.
Figure 2: Electric Distribution System sectors 
The primary system is most often radial, constituting a tree and it is quite frequently loop radial. That means that the main feeder loops through the load area and comes back to the substation, where the two ends are connected with two circuit breakers, providing in this way reconfiguration capability and at the same time backup capability.
Depending on the size of their power demand, customers can be connected on each division or subdivision of the system through a meter. The main objective of the power grid is to deliver electrical energy in a safe and reliable manner to the consumers preserving at the same time good quality.
2.2 State Estimation
Computer-based systems, supported by supervisory control and data aqcuisition (SCADA) installed throughout the system, deal with the secure operation of the power grid. Because the system is consisted of thousand of buses and its complexity, due to the interconnection of the systems, has increased, it is impractical to monitor every single point and the control centers need to resort to state estimation techniques.
Fred Schweppe in 1970  defined the state estimator as ‘a data processing algorithm for converting redundant meter readings and other available information into an estimate of the state of an electric power system’. Extracted from these computer-based techniques, an accurate estimate of the state of the system under examination is available for the energy control centers, in order to proceed with further control actions. Typically, the state estimation problem can be solved by the classical least squares algorithm, considering the limitations on the available measurements and innacuracies contained in the acquired data i.e. errors.
Consider a power network composed of N buses and being monitored at specific points. The collected measurements are typically affected by noise and gather the physical information of the system, such as real and reactive power injections, P and Q, branch power flows and bus voltages. Suppose that these M measurements are stacked in a Mx1 vector z_M that is used in the following non-linear model:
z_M=h_M (x)+e_M (1)
Where x stands for the state vector. This vector includes 2N-1 elements, N bus voltages and N-1 bus angles, where one of the buses is used as reference, putting its phase angle to an arbitrary value, usually 0. e_M denotes a disturbance term capturing measurement error and modelling inaccuracies. Error vectors are assumed to be independent random variables with zero mean and known covariance matrix. Functions h_M (x) are in general non-linear, relating the measurements with the state vector x and depend on the system admittance matrix and the type of measurements. The main complication is that to obtain a SE, one needs to solve a non-convex optimization problem. Non-linear functions are linearized via the Gauss-Newton method, leading to the following computationally linear model:
z_M=H_((M*2N-1)) x_((2N-1))+e_M (2)
where H is the Jacobian matrix with M rows and 2N-1 columns .
Finally the system state is estimated by solving the next LS problem:
min┬x〖f‖z_M-Hx‖^2 〗 (3)
2.3 State Estimation and Future Research Lines
Having already referred to the importance of state estimation (SE) on the control and operation of electric power grids, in the following we clarify the crucial role of SE in up-to-date energy management systems (EMSs), where several operational applications rely on a reliable state estimation of the network.
Taking into concern the continuously increasing presence of renewable energy sources to the distribution grid, one of the major directions where the existing electric power system infrastructures are going to evolve, is the transformation into the so-called ‘smart grids’. At this part of the system, distributed technical innovations will be used, such as distribution automation, up-to-date metering systems and innovation in telecommunications. The EMSs will be in charge to perform critical tasks such as optimal power flow and reconfiguration of the distribution grid, helping the development of smart grid to meet the requirements of the future: environmental compliance, improved dependability, reliability and energy conservation.
Several major future aspects of the electric grid will directly affect the research on SE. The implementation of advance metering technologies, such as Phasor Measurement Units (PMUs), bring us one step closer to real time monitoring. The pressing need for more accurate real-time models, contributing in ancillary services such as bidirectional power flows and demand response. Likewise, communication and data processing infrastructure of the network has to deal with an enormous strain. Time skews, communication failures and infrequent instrument calibration can yield bad data, i.e. corrupted power system readings. Besides, market pricing and deregulation of the power grid has led to the operation and monitoring of different geographical areas by different utility companies, exchanging among them more information and raising at the same time security issues as the power grid deals also with challenges on cybersecurity.
The aforementioned challenges have raised the interest of different research communities such as power and signal processing engineering.
2.4 Research Areas
The classical approach on the state estimation problem, introduced by Schweppe  in the late 1960s, consists the solution of conventional weighted least squares problem (WLS) in a centralized manner. That is, using a single central state estimator by aggregating the available measurements and producing the state of the whole power system. In addition to the traditional model, improvements added mostly from computational cost perspective , such as the fast decoupled state estimator or the implementation of orthogonal/QR factorization that can yield to an accurate solution.
The latest literature on SE refers to the multi-area approach for the solution of the problem in order to overcome the barriers mentioned previously. That is, divide the power system into smaller areas where the SE can be solved locally and thus reduces the associated computational complexity, improves robustness and keeps privacy on the local measurements. Essentially, the idea is to process the information locally. Mathematically speaking, equation (1) is transformed in :
z_Mk=h_Mk (x)+e_Mk (4)
where k=1,2,…K denotes the measurement area, x_k^T=[x_ik,x_bk] is the state vector of area k, with x_ik standing for the local variables and x_bk for the border variables associated to the boundary buses that connect different areas, respectively.
For multi-area approach, there exists two approaches: hierarchical and decentralized. In the former, each area computes a local SE which is then transmitted to a central processor where the global SE is finally computed. In the other case, i.e. the decentralized approach, the areas communicate only with other neighboring areas, exchanging border information, i.e. active power flow of the branch that connects two different areas.
As for the hierarchical approach, Korres in  and  presented a hierarchical multi-area SE where minimum information on the border variables is exchanged with the coordination level. The results show an accuracy similar to that attained by the centralized scenario with a low computational cost. However, important communication issues, such as arrival delays were not addressed.
The author presented also observability and bad data analysis. On the one hand, observability analysis is performed to ensure that the state vector of the area under examination is observable. The methods that have been proposed in the literature are numerical and topological . On the other hand, bad data processing deals with the detection of gross errors on the measurements that may corrupt the total set. The traditional tests for this purpose are statistical tests referring as the most common the chi-square test. .
In , these two functions are accomplished in a distributed manner using a numerical approach for the observability analysis and localizing bad data detection for the internal measurements and for the boundary measurements.
In the hierarchical SE context, Exposito et al propose in  a two-level SE with local measurement pre-processing. At the first level, a largest set of raw measurements are processed by a linear SE, passing the results to the second level of estimation, where a conventional non-linear SE is used. It is presented that this method outperforms the traditional SE, from a computational cost perspective, relieving also the communication channel bandwidth, as a reduced set of measurements has to be transmitted to the management system. The same authors present in , present the same model using at the first level a minimal set of intermediate variables in order to produce a linear model without taking into concern the convergence. The results shown better efficiency in terms of computational cost and convergence compared with the existing SE. The drawback of a potential unobservability of the auxiliary variables that are introduced in the first level is noted.
Abbasi and Seifi in , present a novel approach based on the hierarchical manner. Here, the power system is examined indivisibly, taking into concern the transmission part, assuming balance in the phases and the distribution grid assuming unbalanced phases. That is, the SE problem of the whole system is divided in subproblems that represent the state estimation at the transmission part and many smaller scale state estimations at the distribution network. The results validate the accuracy, convergence and efficiency of this method and it is noted that future studies on the implementation of the SE by intelligent estimators has to been made focusing on the optimized control of the interconnected transmission and distribution grid.
For further information on hierarchical SE, the interested reader is referred to .
In the decentralized SE scenario, Conejo et al in  propose a decentralized state estimation approach where the convergence is ensured by exchanging only border information. In particular, the authors resort to dual decomposition theory and break down the problem into several subproblems. Then areas iteratively exchange information on neighbor border buses to increase the accuracy of their local SE. The effect of redundancy level and bad data presence on estimation accuracy is assessed. The results show that the requirements for convergence are mild and that the procedure is robust.
In a similar context, Kekatos and Giannakis , go one step beyond  and propose a novel algorithm based on the alternating direction method of multipliers (ADMM). ADMM is a powerful algorithm that is used to distributed convex optimization in problems within several different engineering sectors . The method, developed in the 1970s, divides the global problem to small local subproblems that are coordinated in order to find the global solution. In this work, a systematic cooperation between local centers is presented, lowering the information exchange and guaranteeing the convergence regardless of local observability. Treating bad data in the same decentralized manner, the authors propose an optimization problem that is a convex quadratic problem, joining at the same time state estimation and bad data identification. This decentralized scenario is analyzed  along with a proposed Jacobi-like algorithm. The authors present a practical case where the measurements are available from Phasor Measurement Units (PMUs), located at grid buses and affected by noise. The two distributed and scalable algorithms can be implemented locally by local data aggregators or by the individual PMUs, in a peer-to-peer fashion. The simulations showed that the algorithms are extremely effective in canceling the measurement noise and the state estimate is accurate enough.
As in , several technical publications deal with PMUs implementation and their specific placement. Being the key technology toward wide area monitoring, the exact location of the units is an up-to-date important issue for system operators worldwide. PMUs consist a function related directly to observability. An extended overview on PMU placement methodologies can be found in  and a total view of their benefits in .
Working on decentralized SE and based on , Kekatos and Giannakis in , optimized PMU deployment based on estimation-theoretic criteria. The problem of state estimation is posed under the framework of optimal experimental design problem, casting PMU placement as a variation of the problem, minimizing the estimator error through its covariance matrix and using for demonstration the IEEE test cases.
Phadke et al  work on the analysis of placement techniques for PMUs and their interaction with observability taking also into account the communication constraints. The tree search placement technique is illustrated as a topological analysis and its theoretical formulation. The limited communication infrastructure have been also taken into account, posing the constraint problem in the framework of Simulated Annealing (SA). The novelty in this work is the systematic approach of installing PMUs around the network. The same methods are used in  where the criterion is the minimization of the meters while full observability is guaranteed.
Due to the advent of PMUs, providing the option for the construction of a real time state estimator able to forecast state trajectory, Forecasting-Aided SE (FASE) has been revisited. FASE was a first approach on a more real-time state estimation of electric power systems. The concept of FASE was to track the changes that are presented during system operation, trying to provide a real time SE. The main advantage is the addition of a forecasting feature, exploited in order to replace missed measurements via the predicted state.
An extended overview on the topic can be found from Couto Filho and Stacchini de Soouza in . This context is also presented in  where technologies for Smart grid control and operation are presented extensively. Surveys on SE, referring also in FASE can be found in .
3. Advanced Distribution Management System
The motivation for advanced distribution management is primarily computational, even though additional merits of coordination across adjacent control areas were also recognized . In the literature, there is no clear distinction between decentralized state estimation, distributed state estimation and fully distributed state estimation; these terms are related to, more or less, the same concepts. The main two used concepts in the environment of power system state estimation are hierarchical and fully distributed state estimation, which are shown on Figure 3.
Hierarchical and fully distributed state estimation 
3.2 Distributed State Estimation
The authors in  are proposing a fully distributed method to state estimation in large multi-area power systems. The algorithm is implemented for AC and DC state estimation and the proposed algorithm is defined in five steps, but currently bad data detection is not included in the algorithm.
Unlike to the method proposed in , the authors in , as the major issue addressed to bad data detection. They present a method in a decentralized environment for the detection and identification of bad data in power system, divided into two categories. The developed method is Decentralized Adaptive Re-weight State Estimation specifically for wide-area networks.
A very interesting approach is presented in  where the system is formulated as a stochastic model with graphical representation and it conducts statistical inference using an efficient signal processing algorithm. This new approach can be extended in decentralized distribution environment. The similar approach is presented in  with fully distributed algorithm associated with the Message-Passing (MP) algorithms for electric power system state estimation.
A distributed algorithm which does not require either local observability or a central coordinator is proposed in . Also, proposed algorithm does not require an extensive matrix inversion computational burden. This algorithm is implemented in a fully distributed in multi-area interconnected networks.
In the decentralized power system a very important role is played by Phasor Measurement Units (PMU). Phasor Measurement Units have many useful features. They can be used to see dynamic system effects and to help engineers measure the system’s response to disturbances so as to correctly calculate dynamic parameters. PMUs require that the measurements are taken virtually simultaneously so that a GPS signal is required at the substation of the measurement . The authors in  take advantage of PMUs to define distributed state estimation because PMUs can synchronize the measurements between different areas which are perfectly suited for distributed state estimation. Korres, G.N in  developed a distributed multi-area state estimation with PMUs with observability and bad data analysis. This approach is the closest one to the concept of a fully distributed state estimation method.
The state estimation problem in a power system is based on the solving a weighted least-squares function. The problem is nonconvex because of nonlinearity of the equations which describe the power system. The main challenge is how to find a solution which is as near as possible to the global optimal solution with an efficient algorithm in a distributed environment. Zhu, H. and Giannakis, G. B. proposed distributed semidefinite programming for reducing computational complexity through distributed implementations . More details about convex relaxation methods can be found in .
4. Load Clustering for Large Scale Demand Side Management Optimisation
Generation from RES (Renewable Energy Sources) such as photovoltaic cells (PVs) and wind turbines (WTs) in the power grid is increasing in the UK , with similar trends in Europe and the USA, setting targets of RES integration by 2020, while China has doubled wind power production every year since 2004 -. However, this generation from RES is random, intermittent in nature and usually distributed, causing some problems in power systems, e.g., power mismatch due to forecasting errors, distribution voltage rise and frequency fluctuation. With high penetration rates of RES in the Power Systems, these fluctuations become substantial, making it difficult to maintain balance between power supply and demand in the grid. At the same time, overall power consumption is increasing over the years; in particular, peak electric demand is rapidly growing.-
Energy storage systems (ESSs) have been proposed as an effective solution to this problem. It is preferable to keep installed capacity of ESSs as small as possible, since ESSs are generally expensive and especially for installation on a large scale. But recent research is focusing on more feasible methods, namely controllable loads -.
4.2 Demand Side Management Potential
The primary object in power systems is matching precisely the power demand and supply every moment while operating the grid within optimal limits. The main concept behind DSM derives from the potential of some loads as controllable loads, thus making use of already existing components of the grid. Demand-side load management (DSM) services can be procured by electricity system operators through monitoring, aggregation and control to maintain reliability of electric power systems. As described by S. Kawachi et al. , an ideal controllable load has: a) minor loss of convenience by control of power consumption, b) power consumption large enough and c) response speed to signals fast enough to compensate power fluctuation. However, there isn’t any ideal controllable electrical load, thus, some loads satisfy partially these requirements and are treated as controllable loads in practice.
Controllable loads provide various DSM services, depending on the specific characteristics of each type. These services can be load shaping (RES integration or price following) -, -, frequency control -, , voltage control , overload relief (transmission and distribution) , grid reliability , -, peak load reduction , , reserve (in the form of positive or negative regulation)  and more. Based on the service that is provided, different loads or groups of loads are utilized. For instance, T. Masuta & A. Yokoyama  simulate frequency control with WHs (Water Heaters) and EVs (Electric Vehicles), which can be switched on/off for short intervals (in case of high frequency fluctuation) without affecting the quality of service. In , I. Hernando-Gil et al., make use of wet loads for DSM, but in this case shifting the load’s operating time to achieve peak demand reduction. Thus, it is important to identify which services can be provided, by which controllable loads, when and in what volume.
One way of categorizing the services, and hence loads, is by looking the nature of the services and methods used. Direct Load Control (DLC) , is a central or automated control, such as in ; it is used for cases of faults, lost generation, significant forecast errors or RES fluctuation etc., where an immediate response within a couple of seconds is needed in some cases . Services provided through direct control are mainly for grid reliability such as frequency response, reserve , etc. The other type, is Indirect Load Control (also known as dynamic pricing), where costumers are prompted to alter their consumption (demand profile) through the use of dynamic tariffs . Dominant tariffs for this method are Time of use (TOU), Real-time pricing (RTP), Critical peak pricing (CPP) and Peak time rebate (PTR) -. These can be used for peak shifting/shaving, valley filling or Renewable Energy Resources (RES) following methods. A percentage of costumers is expected to alter the starting time of their appliances (or an automated system such as in . The exact response (number of costumers) to dynamic tariffs and their effect on load shaping depends to the prices themselves and human behaviour.
4.3 Flexible Loads: Interruptible And Deferrable
Controllable loads, also known as flexible loads, fall mainly into two categories, the interruptible loads and the deferrable loads (also found in the literature under the term load shifting). Interruptible loads are those whose operation can be altered, usually switched off, for a short amount of time without affecting the quality of service , -. For instance, an EV or WH (usually operate for a few hours) can be switched off (or reduce their consumption) for a few minutes in case of an emergency (e.g. lost generation), as long as the battery gets fully charged or the water temperature is within the thermostat’s limits . Interruptible loads are suited for DLC, thus in case of an emergency a central point can coordinate the interruptible loads properly (power volume, ramp up/down rates, duration) , .
Deferrable loads (load shifting) are mainly described by their flexibility of shifting their operational time (start time usually) , , -. For instance, a washing machine or a dishwasher can be programmed to postpone (or advance) its start time to a more favourable time (i.e. lower price due to excess RES generation or off-peak use) , . Deferrable loads are suitable for indirect load control (dynamic pricing) and is a form of decentralized DSM. Though because of its nature, human behaviour (even when assisted by automated systems ) plays a big part. This can be seen in , with mixed responses from users regarding the acceptable start delay.
4.4 Importance of Clusters
There are three main consumer sectors in power systems: the industrial sector, the commercial sector and the residential sector. Currently, the focus of DSM is primarily on the industrial sector due to the inherently large loads, the existing metering infrastructure (sensors and metering technologies available) and the existing staff with expertise on power systems. Also, the commercial sector, even though it tends to have a more distributed load consumption (smaller loads), facilities with enough flexibility have the ability to participate (sometimes groups of them). Residential loads are gaining more attention lately, but have not been largely used since the loads are small, distributed, and not automated , , , -.
The major challenge lies in the domestic sector (being about 1/3 of a National Grid’s total electricity consumption ) and the small commercial consumers. Mainly because a lot of small consumption units need to be grouped and controlled simultaneously to achieve the same results with large commercial or industrial units, (which have hundreds to thousands times higher consumption and can participate in DSM on their own) . In addition, problems arise from deviation in load profiles, limited knowledge of their load composition (how many interruptible/deferrable loads operate and at which times/conditions) and limited knowledge of their potential for DSM (including the end users’ awareness and thus willingness to participate).
Therefore, knowledge of the composition of the residential sector is essential, in order to know in real time which loads can be used, when they are available and in which volume for each specific DSM service. This effectively means analysing the loads and their potential for DSM, their total volume (aggregated power), how much of it can be utilized and at which times during the day, week, season (thus essentially main driving factors). For example, electric heating as previously mentioned (-, -, ) can be used for DSM, but its availability depends on weather conditions (during cold weather mostly) and human behaviour. Also the constraints of DSM, proper modelling (which should include constrains and main driving factors) and management methods to avoid negative effects/constraints. One such, known as the rebound effect, is the oscillation created when interrupted loads are switched on again, such as in  where TCLs (Thermostatically Controlled Loads) are utilized. Even though in this case it gets reduced over time, because of random factors who affect TCL units (human behaviour, weather variation, differences between TCL unit types), it is still substantial initially, recreating similar fluctuations as the ones trying to correct . Another important part is separating the loads based on their characteristics to uncontrollable, flexible interruptible and flexible deferrable (some of which fall in both categories).
Because of the large number of electricity consumers diversely distributed in the system, it is difficult to grasp the overall aggregated behaviour of the consumers. Analysing the data of Smart Meters is essential and various methods have been used to study load profiles -, but few specifically for DSM, and even less taking into consideration the need to separate loads in appropriate categories (uncontrollable, interruptible and deferrable), the main factors that affect them and thus extract information of the total available power for each DSM service. In  and , the focus is on Smart Grids but on the medium voltage and not on the low voltage where residential users are connected. R. Li et al.  take a step further with load clustering on the substation level, to examine the mixture of residential, commercial and industrial users. Another important use of clustering is shown in , where Y. Kim et al. investigate the use of smart meters’ readings to forecast similar consumers without smart meters. The next step in load analysis is to cluster not only whole usage profiles but cluster their controllable components (flexible interruptible and deferrable), thus extracting information on the appliances utilized and which of those can be used for DSM.
Approaches for the identification of appliances have been made, such as in , where H. Niska uses load clustering to extra info whether electrical heating (a controllable load) is installed, by checking users’ load profiles. A different method is proposed by Y. Lin & M. Tsai , using appliance signature recognition, where specific electrical signatures can be spotted from a user’s profile, which match certain appliances. Additional info though is needed, knowledge of all controllable loads’ profile (volume and time), affecting factors and potential for DSM; thus being able to determine the aggregated available power and periods of residential loads, for each DSM service separately.
5. Treatment, Use and Value of Smart Grid Data in a Real Environment
One of the principal problems which impacts the efficiency and security of the power distribution networks are the power losses occurring within the process of energy delivery to the consumer. These losses can be distinguished into: i) Technical losses -i.e. losses due to naturally occurring phenomena in the power system, such as power dissipation on the transmission lines and transformers- and ii) Non-Technical losses (NTL) -due to administrative problems or unauthorized energy consumption- .
The NTL are in fact unbilled energy that is not scheduled or expected by the utility, thus it can severely affect the power system’s operation . Critical operational problems that may arise include overloads of generation units and the stressing of network equipment due to over-voltages. These result from the fact that the utility cannot schedule sufficient active and reactive power due to dynamic and insufficient load flow information. Furthermore, these over-voltages can impact on the equipment of honest consumers. In extreme cases of excess unplanned load, blackouts and brownouts may also occur. Concerning the distribution utilities, apart from the directly incurred economic losses as a consequence of purchasing energy that is not billed for, maintenance costs also increase due to the aforementioned stressing of the equipment. Last but not least, the environmental impact of NTL is also considerable due to the increase in CO2 emissions. A 10% reduction in NTL in India -which is estimated at about 83000 GWh- would result in a 9.2 million tons CO2 decrease annually .
Electricity theft accounts for a large portion of the power system losses worldwide. Utility companies worldwide are evaluated to incur more than $20 bn. annually due to electricity theft, while the power losses of any distribution system should not exceed 5% . In the Netherlands, 23% of the total losses in power distribution -approximately 1200 GWh/year- is attributed to electricity theft, and is estimated at an approximate €114 m every year . In the United States, electricity theft costs $6 bn. per year . Finally, electricity theft accounts for approximately $4.5 bn. and $5 bn. losses in India and Brazil respectively . ENEL –the Italian electricity utility- was motivated to initiate a large scale roll-out of Smart Meter-based infrastructure in order to minimize NTL of their distribution network. After the installation of smart meters (SM) at the consumers’ premises, the theft hit-rate raised from 5% to 50% . This massive deployment of SM is now extended to Spain, being facilitated by Endesa -one of the Spanish distribution utilities- with 3.5million SM units having been installed by the end of April 2013 . The data provided by the SM devices give a new perspective and unveil numerous possibilities to develop efficient and effective theft detection methods. Research in this respect has recently shown significant progress.
There are various ways that the data retrieved from SMs have been analyzed and exploited in order to detect NTL. An attempt to categorize and group these methods was made, hence in the following, the: a) System State-based, b) Game Theory-based, and c) Statistical-based methods are presented.
5.2 System State-based Methods
State Estimation (SE) has been used by Chen et al. in  and by Huang et al. in  and . In general, weighted least squares MV SE using network topology and redundant measurements by SM was exploited in these works in order to calculate the voltage and the loading of the MV/LV distribution transformer. In the first case, it is supported that by applying SE in each phase of the MV network and taking advantage of the cabability of SE to identify bad measurement data, the theft point can be identified by considering that a tampered SM communicates bad measurements. In the second case, a three-phase SE was used to determine whether there is an anomaly at a MV/LV transformer. If an anomaly is detected, Analysis-of-Variance (ANOVA) is employed at the LV feeder level to compare SM historical consumption data with current measurements in order to determine the problematic SMs. Niemira et al. in  implement a SE model in order to detect malicious data attacks. The attacker prepares an attack measurement vector supposing that the utility uses a DC SE. If these measurements are used with an AC SE the residuals will be increased since the DC models makes simplifications regarding the losses. An analysis on the sensitivity of active & reactive power flows and active & reactive generation power injections residuals of a nonlinear SE to the attack measurement is presented. These are compared with baseline residuals and it is found that the active power generation injection residuals are more representative to detect the attack.
Weckx et al. in  propose a linearized load flow algorithmic approach using SM data for electricity theft detection via illegal connections, when line lengths are unknown or uncertain. At the same time, basic information of the topology can be extracted and the phase of consumers can be identified in an automatic way. Using (1) with historical measurements from SMs without any fraud taking place then α_(h,h ̃ ) and b_(h,h ̃ ) can be considered as the unknowns and an ordinary least squares problem is defined.
V_(h,k)=V_k^0+∑_(h ̃=1)^N▒〖α_(h,h ̃ ) P_(h ̃,k)+∑_(h ̃=1)^N▒〖b_(h,h ̃ ) Q_(h ̃,k) 〗〗 (5)
where k is the timestep, h the house, N the number of houses, V_k^0 is the voltage magnitude at the MV/LV distribution transformer, P_(h ̃,k) and Q_(h ̃,k) are the active and reactive power of the consumer h ̃ respectively, α_(h,h ̃ ) and b_(h,h ̃ ) are the influence factors of the active and reactive power respectively of consumer h ̃ on the voltage magnitude of consumer h. After the influence factors have been determined, the voltage at each consumer’s premise can be calculated from (1), using new measurements from SMs that possibly entail electricity theft and can then be compared with the voltage measurement of the SM. The parameters α_(h,h ̃ ) and b_(h,h ̃ ) are also indicators of the relative location of the SM and the phase they are connected to. If the SM h is connected at the same phase as the h ̃ the parameter α_(h,h ̃ ) will be negative since the active power has created a voltage drop. If it is connected to another phase, then the parameter will have a low positive value.
Salinas et al. in , taking into account customers’ privacy preserving, propose three distributed algorithms based on peer-to-peer computing in order to calculate customers’ “honesty coefficients”. The distributed LU and QR decompositions of the consumption measurement matrix from SM are employed to solve a linear system of equations (LSE) while preserving each node’s information. For a small network LU decomposition (LUD) can localize the thieves: unfortunately, the same methodology can prove to be unstable for large networks. For the latter ones, LUD with partial pivoting (LUDP) is implemented, as well as QR decomposition (QRD) is also proposed and implemented for this case. The aforementioned methods are applied in cases with constant fraud. In addition to these, adaptive LUD, LUDP and QUD for scenarios with variable theft activity are also presented. Finally, a comparison of the computational load required for each method is given.
Han et al. in  propose a NTL Fraud Detection (NFD) method based solely on data obtained from SMs. The criterion used to identify dishonest customers is the difference between the billed energy and the realized consumption. Assuming that there is a SM at the distribution transformer recording the overall energy supplied to n customers in a neighborhood and that the technical losses are estimated and extracted from E_j, where E_j denotes the energy measured at the distribution transformer, and let E_(i,j) be the energy measured at the i-th SM, and x_(i,j) the electricity reported to the utility by the i-th SM, energy balance gives E_j=∑_(i=1)^n▒E_(i,j) . If the consumer is honest Ei,j/xi,j≈1, for a dishonest customer |E_(i,j)⁄x_(i,j) -1| will be very large. For each SM an accuracy coefficient is defined as a_(i,j)=E_(i,j)/x_(i,j). While the reported energy is available, the measured values are not. There is a function for each SM such that f_i (x_(i,j) )=E_(i,j) ,j=1,2,…,m, based on Taylor approximation f_i (x)=∑_(k=m)^0▒〖a_(k,i) x^k 〗. By replacing the previous two equations in the energy balance equation, (2) is derived, which with m samples of x^k for each SM and E_j known, the accuracy coefficients can be estimated:
E_j=∑_(i=1)^n▒∑_(k=m)^0▒〖a_(k,i) x_(i,j)^k 〗 (6)
5.3 Game Theory-based Methods
The research in electricity theft detection using game-theory methods is not extensive and has yet to reach maturity. Nevertheless, they provide a new perspective on how to deal with the matter of electricity fraud. These methods are mainly used to provide the utility with cost-effective solutions regarding the level of investment in theft detection, as well as tariff and fine regulation. On the down-side they do not deal with theft localization issues.
Cardenas, Amin et al. in  and in  present a model to examine the network observability choices that the utility has when a known part of customers commits fraud. The model finds optimal solutions for both the attacker and the defender. Monopolistic and deregulated environments are considered, and in both cases the game is a “leader-follower” one. The utility is the leader who selects first and the consumers make their move after they are informed about the electricity tariffs and the investment of the utility in electricity fraud detection. The consumers incur different costs, since only the honest customers pay the full bill. The fraudulent customer chooses two amounts of consumption, the one for which he/she will be billed and the one that will be stolen. This latter one is chosen based on the detection likelihood and on the volume of the fine that should be paid if detected. The detection likelihood is determined by the amount of stolen energy and the investment of the utility in theft detection. The utility selects the degree of investment in theft detection and the tariffs for energy consumption. The operating point of the detection test on ROC (ROC expresses the relationship between the detection and the false-alarm likelihood) is based upon the decision on which level of investment should be chosen, and the customers’ equilibrium consumption levels.
The results suggested that in monopolistic environments, the customers are more likely to steal electricity to a greater extent, since the utility’s effort to monitor the network is low. Additionally, the practice of electricity theft is increased when the price is higher or the fines are low. However, there is a marginal cost of theft monitoring and a fixed fraction of dishonest consumers, for which the utility’s profit increases with the investment level. The proposed model is useful to determine the optimal investment level in theft detection for different regulatory environments.
5.4 Statistical-based Methods
Statistical-based theft detection techniques are the most popular ones, since they were available-to-use before the deployment of SMs, while now, they can further advance and improve within the framework of SMs. These methods usually, within the context of data mining, refer to the classification of the consumers’ load profile. The aim is to determine irregular patterns in the electricity consumption over time.
Nagi et al. in  approach the electricity theft detection problem by developing an artificial intelligence technique, specifically a Support Vector Machine (SVM). In this method historical consumption data and additional consumers’ attributes are used to identify irregular consumption profiles that are highly correlated with NTL. The consumers are classified either as “normal” or “fraud” by the SVM model. The consumers’ consumption patterns are determined by employing data-mining and statistical analysis tools trying to identify sudden changes in the consumption profiles. The features that were eventually chosen include: a) 24 (derived from monthly values for two years) daily average consumption values for each customer, which correspond to their load profile, and b) the credit worthiness information CWR (this is produced by the utility’s billing system automatically for customers that do not pay their bills) for each customer. After collaboration and on-site inspection with Tenaga Nasional Berhad, it was found out that the expected hitrate increased from 3% to 60%. In  the previous work was extended, introducing a fuzzy inference system (FIS) in the form of IF-THEN rules. It is a post-processing technique to apply an extra intelligent decision filtering on the customers and distinguish the ones with higher probability of fraud. The previously hitrate raised from 60% to 72%.
Depuru et al. in  implement a SVM model representing several possible forms of theft. Consumption patterns are categorized based on economic zone (agriculture, residential, commercial), seasonal, and geographical criteria. The consumers are grouped according to their residence size and their contracted capacity. The dataset used in this work consisted of 135 average patterns with 10 customers each and measurements of 15-minute intervals (96 attributes for each day). In , the authors enhanced their model with a robust and quick data encoding technique. In this case the encoded data are used as an input to the SVM model, reducing the total number of inputs to five. The encoding technique traces irregularities in the customers’ instantaneous energy consumption. The accuracy of the model is 92%.
Babu et al. in  use fuzzy C-Means clustering to categorize consumers based on their consumption patterns. The difference of clustering to classification is mainly that clustering refers to the grouping of observations into classes of similar objects without having any pre-classified observations. In fuzzy clustering, an observation can belong to more than one class, with a different degree-of-membership. The fraud identification relies on the fuzzy membership function and the normalized Euclidean distances of cluster centers ordered by unitary index score. The highest score represents fraudulent consumers. The method uses five attributes that are considered to describe a consumption pattern. These attributes include: a) the average consumption, b) the maximum consumption, c) the standard deviation of consumption, d) the sum of inspection comments during the last six months, and e) the average consumption of the neighborhood. Data of another twelve months are required for the clustering process.
Finally, Faria et al. in  utilize the consumer baseline load curves that have been developed for demand response applications. For each period of the historical data, the expected consumption is estimated, then statistical parameters of the expected and the realized consumption are compared and if there is considerable difference the consumer is characterized as a possibly fraudulent one. The baseline types that were used are the following: a) type-I, in which load historical data are used and may include other data such as weather, and b) type-II which is used for aggregated loads.
To conclude, NTL detection is a challenging issue for the power distribution networks. With the onset of AMI, and massive deployment of SM within the context of smart grids, electricity-theft became even more complicated. At the same time, more means to address this problem have emerged. After presenting the main idea of each individual approach, it is pointed out that each method addresses only a few aspects of the multidimensional problem of electricity theft. Hence we believe that energy-theft detection is an active and challenging research area, where the need for progress and future advancements lies within the immediate future.
6. Key Research Objectives for The Work Package
The research objectives for WP4 in the ADVANTAGE project can be summarized as follows:
State Estimation and Bad Data Detection on Distribution Grids taking into account issues of privacy protection, communications overheads and algorithm convergence time.
Evaluation of the perfomance of Real Time Control Algorithms for Distribution networks based on State Estimation using large sets of measurements.
Study on the Observability of MV/LV Networks and the directly related function of Phasor Measurement Units placement and implementation.
Fully distributed algorithm with observability and bad data analysis generally.
Efficiency algorithm which is able to find a solution near to global optimum with guaranteed convergence and with small computational complexity.
The state estimation in distribution grid poses a challenge, especially integrated the state estimation in distribution management system, applied on the real system with all specifics of the distribution grid (underdetermined system, weakly meshed networks, ampere measurements are dominant).
How flexible can each type of controlable loads be?
Services that can be provided through diffrent types of flexible loads.
What is the potential of the Domestic sector for DSM, thus the available volume of flexible loads during specific times of the day.
Energy Balance at MV level – what is the losses threshold to characterize a feeder as suspicious
MV & LV state estimation for bad-data/fraudulent data detection
Investigation of SM-concentrator diaphony and topology inaccuracies
Statistical analysis of historical customer data
Zia A. Yamayee and Juan L. Bala, Jr, ‘Electromechanical Energy Devices and Power Systems‘, John Wiley and Sons, Inc, 1994.
John J. Grainger, William D. Stevenson, Jr, ‘Power Systems Analysis’, McGraw-Hill International Editions, 1994.
F. C. Schweppe and D. B. Rom, ‘Power System Static-Static State Estimation, Part II: Approximate Model’, IEEE Transactions on Power Apparatus and Systems, No. 1, January 1970.
A. G. Exposito, Ali Abur, A. de la Villa Jaen and C. Gomez-Quiles, ‘A Multilevel State Estimation Paradigm for Smart Grids’, Proceedings of the IEEE, Vol. 99, No. 6, June 2011.
G. N. Korres, ‘A Distributed Multi-Area State Estimation’, IEEE Transactions on Power Systems, Vol. 26, No. 1, February 2011.
G. N. Korres and G. C. Contaxis, ‘Application of a reduced model to a distributed state estimator’, IEEE Power Engineering Society, Winter Meeting 2(2000) 999-1004.
G. B. Giannakis, V. Kekatos, N. Gatsis, S. J. Kim, H. Zhu and B. F. Wollenberg, ‘Monitoring and Optimization for Power Grids’, IEEE Signal Processing Magazine, September 2013, 107-128.
A. G.Exposito and A. de la Villa Jaen, ‘Two-Level State Estimation with Local Measurements Pre-Processing’, IEEE Transactions on Power Systems, Vol. 24, No. 2, May 2009.
A. G.Exposito, A. de la Villa Jaen and Q. G. Quiles, ‘A Factorized Approach to WLS State Estimation’, IEEE Transactions on Power Systems, Vol. 26, No. 3, August 2011.
Ali Reza Abbasi and Ali Reza Seifi, ‘A New Coordinated Approach to State Estimation in Integrated Power Systems’, Elsevier, Electric Power Systems Research 45,152-158, 2013.
A. G. Exposito, A. de la Villa Jaen, C. Gomez-Quiles, P. Rousseaux and Thierry Van Cutsem, ‘A Taxonomy of multiarea state stimation methods’, Elsevier, Electric Power Systems Research 81, 1060-1069, 2011.
A. J. Conejo, S. de la Torre and M. Canas, ‘An Optimization Approach to Multi-Area State Estimation’, IEEE Transactions on Power Systems, Vol. 22, No. 1, February 2007.
V. Kekatos and G. B. Giannakis, ‘Distributed Robust Power System State Estimation’, IEEE Transactions on Power Systems, Vol. 28, No. 2, May 2013.
S. Bolognani, R. Carli and M. Todescato, ‘State Estimation in Power Distribution Networks with Poorly Synchronized Measurements’, IEEE Conference on Decision and Control, CDC ‘14, 2014.
N. M. Manousakis, G. N. Korres and P. S. Georgilakis, ‘Taxonomy of PMU Placement Methodologies’, IEEE Transactions on Power Delivery, Vol. 27, No. 2, May 2012.
A. Abur, A. G. Exposito, A. de la Villa Jaen, C. Gomez-Quiles and P. Rousseaux, ‘On the Use of PMUs in Power System State Estimation’, 17th Power Systems Computation Conference, Stockholm, Sweden , August 22-26, 2011.
S. Boyd, N. Parikh, E. Chu, B. Peleato and J. Eckstein, ‘Distributed Optimization and Statistical Learning via the Alternating Direction Method of Multipliers’, Foundations and Trends in Machine Learning, Vol. 3, No 1, (2010), 1-122.
S. Joshi and S. Boyd, ‘Sensor Selection via Convex Optimization’, IEEE Transactions on Signal Processing, Vol. 57, No. 2, February 2009.
V. Kekatos, G. B. Giannakis and B. Wollenberg, ‘Optimal Placement of Phasor Measurement Units via Convex Optimization’, IEEE Transactions on Power Systems, Vol. 27, No. 3, August 2012.
R. F. Nuqui and A. G. Phadke, ‘Phasor Measurement Unit Placement Techniques for Complete and Incomplete Observability’, IEEE Transactions on Power Delivery, Vol. 20, No. 4, October 2005.
A. Y. Abdelaziz, A.M. Ibrahim, R. H. Salem, ‘Power System Observability with Minimum Measurement Units Placement’, International Journal of Engineering, Science and Technology, Vol. 5, No. 3, 2013, pp. 1-18.
M. B. D. Coutto Filho and J. C. Stacchni de Souza, ‘Forecasting-Aided State Estimation – Part I: Panorama‘, IEEE Transactions on Power Systems, Vol. 24, No. 4, November 2009.
A. P. Sakis Meliopoulos, George Cokkinides, Renke Huang, Evangelos Farantatos, Sungyun Choi, Yonghee and Xuebei Yu, ‘Smart Grid Technologies or Autonomous Operation and Control’, IEEE Transactions on Smart Grid, Vol. 2, No. 1, March 2011.
Y. F. Huang, S. Werner, J. Huang, N. Kashyap and V. Gupta, ‘State Estimation in Electric Power Grids‘, IEEE Signal Processing Magazine, September 2012, 33-43.
Le Xie; Dae-Hyun Choi; Kar, S.; Poor, H.V., “Fully Distributed State Estimation for Wide-Area Monitoring Systems,” Smart Grid, IEEE Transactions on , vol.3, no.3, pp.1154,1169, Sept. 2012
Xiao Li; Scaglione, A., “Robust Decentralized State Estimation and Tracking for Power Systems via Network Gossiping,” Selected Areas in Communications, IEEE Journal on , vol.31, no.7, pp.1184,1194, July 2013
Ying Hu; Kuh, A.; Kavcic, A.; Nakafuji, D., “Real-time state estimation on micro-grids,” Neural Networks (IJCNN), The 2011 International Joint Conference on , vol., no., pp.1378,1385, July 31 2011-Aug. 5 2011
Yunxin Li, “Fully distributed state estimation of smart grids,” Communications (ICC), 2012 IEEE International Conference on , vol., no., pp.6580,6585, 10-15 June 2012
Le Xie; Dae-Hyun Choi; Kar, S., “Cooperative distributed state estimation: Local observability relaxed,” Power and Energy Society General Meeting, 2011 IEEE , vol., no., pp.1,11, 24-29 July 2011
A. Wood and B. Wollenberg, “Power Generation, Operation, and Control”, A Wiley-Interscience publication, Wiley, 1996.
Qinghua Huang; Schulz, N.N.; Srivastava, A.K.; Haupt, T., “Distributed state estimation with PMU using grid computing,” Power & Energy Society General Meeting, 2009. PES ’09. IEEE , vol., no., pp.1,7, 26-30 July 2009
Hao Zhu; Giannakis, G.B., “Power System Nonlinear State Estimation Using Distributed Semidefinite Programming,” Selected Topics in Signal Processing, IEEE Journal of , vol.8, no.6, pp.1039,1050, Dec. 2014
Zhang, B.; Lam, A.Y.S.; Dominguez-Garcia, A.D.; Tse, D., “An Optimal and Distributed Method for Voltage Regulation in Power Distribution Systems,” Power Systems, IEEE Transactions on , vol.PP, no.99, pp.1,13, Feb. 2015
S. Kawachi, H. Hagiwara, J. Baba, K. Furukawa, E. Shimoda & S. Numata, “Modeling and simulation of heat pump air conditioning unit intending energy capacity reduction of energy storage system in microgrid,” Proceedings of the 2011 14th European Conference on Power Electronics and Applications, 2011
M. Kajgaard, J. Mogensen, A. Wittendorff, A. Veress, A. Todor & B. Biegel, “Model predictive control of domestic heat pump,” 2013 American Control Conference, 2013
T. Masuta & A. Yokoyama “Supplementary Load Frequency Control by Use of a Number of Both Electric Vehicles and Heat Pump Water Heaters,” IEEE Transactions on Smart Grid, VOL. 3, 2012
M. Starke, D. Letto, N. Alkadi, R. George, B. Johnson, K. Dowling & S. Khan “Demand-Side Response from Industrial Loads,“ 2013 NSTI Nanotechnology Conference and Expo, VOL. 2, 2013
G. Papaefthymiou, B. Hasche, & C. Nabe, “Potential of heat pumps for demand side management and wind power integration in the German electricity market,” IEEE Transactions on Sustainable Energy, VOL.3, 2012
O. Malík, & P. Havel, “Active Demand-Side Management System to Facilitate Integration of RES in Low-Voltage Distribution Networks”, IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 2, APRIL 2014
N. Lu & D. Chassin “A State-Queueing Model of Thermostatically Controlled Appliances,” IEEE Transactions on Power Systems, VOL. 19, 2004
Y. Kim, J. Kirtley, L. Norford & K. Leslie, ”Variable speed heat pump design for frequency regulation through direct load control,” 2014 IEEE PES T&D Conference and Exposition, 2014
I. Hernando-Gil, B. Hayes, A. Collin & S. Djokic, “Distribution network equivalents for reliability analysis. Part 1: Aggregation methodology,” Innovative Smart Grid Technologies Europe (ISGT EUROPE), 2013 4th IEEE/PES, 6-9 Oct. 2013,
H. Hao, B. Sanandaji, K. Poolaa & T. Vincent, “Aggregate Flexibility of Thermostatically Controlled Loads,” IEEE Transactions on Power Systems, 2014
A. Subramanian, M. Garcia, D. Callaway, K. Poola & P. Varaiya, “Real-time scheduling of distributed resources,” IEEE Transactions on Smart Grid, VOL. 4, 2013
K. Huang & Y Huang, “Integrating direct load control with interruptible load management to provide instantaneous reserves for ancillary services,” IEEE Transactions on Power Systems, VOL. 19, 2004
T. Chang, M. Alizadeh & A. Scaglione, “Real-Time Power Balancing Via Decentralized Coordinated Home Energy Scheduling” IEEE Transactions on Smart Grid, 2013
A. Faruqui, D. Harris & R. Hledik, “Unlocking the €53 billion savings from smart meters in the EU: How increasing the adoption of dynamic tariffs could make or break the EU’s smart grid investment,” Energy Policy, Volume 38, Issue 10, October 2010, pages 6222-6231
P. Du & N. Lu, “Appliance commitment for household load scheduling,” Pes T&D 2012, VOL. 2
Q. Shan-shan, C. Jing & C. Xiao-hai, “Interruptible Load Management in Power Market,” 2014 China International Conference on Electricity Distribution (CICED 2014)
L. Tuan & K. Bhattacharya, “Interruptible load management within secondary reserve ancillary service market,” 2001 IEEE Porto Power Tech Proceedings
X. Pan, Y. Li & L. Wang, “Research on coordinative optimal dispatch of interruptible load on multi time scale,” China International Conference on Electricity Distribution, CICED, 2012
D. Niyato, Q. Dong, P. Want & E. Hossain, “Optimizations of Power Consumption and Supply in the Smart Grid: Analysis of the Impact of Data Communication Reliability,” IEEE Transactions on Smart Grid, VOL. 4, 2013
Z. Y, L. Jia, M. Murphy-Hove, A. Pratt & L. Tuong, “Modeling and stochastic control for home energy management,” IEEE Transactions on Smart Grid, VOL. 4, 2013
R. Stamminger & R. Friedrich-Wilhelms, “Synergy Potential of Smart Appliances,” EIE project: Smart Domestic Appliances in Sustainable Energy Systems (Smart-A), D2.3 of WP 2 from the Smart-A project, 2008
T. Zhu, S. Shaw & S. Leeb, “Electric load transient recognition with a cluster weighted modeling method,” IEEE Transactions on Smart Grid, VOL. 4, 2013
X. Jinfeng, Y. Huakun, Q. Mark, J. Li & Q. Ga, “Application of Wavelet-based clustering approach to load profiling measurements,” 2014 China International Conference on Electricity Distribution (CICED 2014)
S. Park, S. Ryu, Y. Choi & H. Kim, “A Framework for Baseline Load Estimation in Demand Response : Data Mining Approach,” 2014 IEEE International Conference on Smart Grid Communications
G. Chicco, O. -M. Ionel, & R. Porumb, “Electrical Load Pattern Grouping Based on Centroid Model With Ant Colony Clustering,” IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013
S. M. Mousavi, & H. A. Abyaneh, “Effect of Load Models on Probabilistic Characterization of Aggregated Load Patterns,” IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011
G. Chicco, & I.-S. Ilie, “Support Vector Clustering of Electrical Load Pattern Data,” IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009
F. Quilumba, W. Lee, H. Huang, D. Yang, & R. Szabados, “Using Smart Meter Data to Improve the Accuracy of Intraday Load Forecasting Considering Customer Behavior Similarities,” IEEE Transactions on Smart Grid, VOL. 6, 2015
W. Labeeuw, & G. Deconinck, “Residential Electrical Load Model Based on Mixture Model Clustering and Markov Models,” IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 9, NO. 3, AUGUST 2013
S. Ramos, J. Duarte, J. Soasres, Z. Vale & F. Duarte, “Typical load profiles in the smart grid context a clustering methods comparison,” IEEE Power and Energy Society General Meeting, 2012
S. Ramos, j. Duarte, F. Duarte, Z. Vale & P. Faria, “A data mining framework for electric load profiling,” 2013 IEEE PES Conference on Innovative Smart Grid Technologies (ISGT Latin America)
R. Li, C. Gu, F. Li, G. Shaddick & M. Dale, “Development of Low Voltage Network Templates — Part I : Substation Clustering and Classification,” IEEE Transactions on Power Systems, 2014
Y. Kim, J. Ko & S. Choi, “Methods for generating TLPs (Typical Load Profiles) for smart grid-based energy programs,” EEE SSCI 2011 – Symposium Series on Computational Intelligence – CIASG, IEEE Symposium on Computational Intelligence Applications in Smart Grid
H. Niska, “Extracting controllable heating loads from aggregated smart meter data using clustering and predictive modelling,” Proceedings of the 2013 IEEE 8th International Conference on Intelligent Sensors, Sensor Networks and Information Processing: Sensing the Future, ISSNIP 2013
Y. Lin & M. Tsai, “Non-Intrusive Load Monitoring by Novel Neuro-Fuzzy Classification Considering Uncertainties,” IEEE Transactions on Smart Grid, VOL. 5, 2014
G. B. Dabre, A. A. Dutta y A. N. Kadu, «Performance Evaluation of Distribution Network and Reduction in Technical & Non-Technical Losses by Using Energy Efficient Equipment and Cost Benefit Analysis in the Power Sector,» de International Conference on Power, Automation and Communication (INPAC), Amravati, 2014.
T. B. Smith, «Electricity theft: a comparative analysis,» Energy Policy, vol. 32, nº 18, pp. 2067-2076, 2004.
S. S. S. R. Depuru, L. Wang, V. Devabhaktuni y N. Gudi, «Measures and Setbacks for Controlling Electricity Theft,» de North American Power Symposium (NAPS), Arlington, TX, 2010.
P. Kadurek, J. Blom, J. F. Cobben y W. L. Kling, «Theft detection and smart metering practices and expectations in the Netherlands,» de Innovative Smart Grid Technologies Conference Europe (ISGT Europe), IEEE PES, Gothenburg, 2010.
S. Salinas, M. Li y P. Li, «Privacy-Preserving Energy Theft Detection in Smart Grids,» de IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (SECON), Seoul, 2012.
R. Jiang, R. Lu, Y. Wang, J. Luo, C. Shen y X. Shen, «Energy-Theft Detection Issues for Advanced Metering Infrastructure,» Tsinghua Science and Technology, vol. 19, nº 2, pp. 105-120, April 2014.
C.-N. Lu, S.-C. Huang y Y.-L. Lo, «Non-Technical Loss Detection using State Estimation and Analysis of Variance,» IEEE Transactions on Power Systems, vol. 28, nº 3, pp. 2959-2966, August 2013.
Endesa, «Endesa Press,» Endesa, [En línea]. Available: http://www.endesa.com/en/saladeprensa/noticias/Endesa-smart-meters. [Último acceso: 20 May 2013].
L. Chen, X. Xu y C. Wang, «Research on Anti-electricity Stealing Method Base on State Estimation,» de Power Engineering and Automation Conference (PEAM), IEEE, Wuhan, 2011.
Y.-L. Lo, S.-C. Huang y C.-N. Lu, «Non-Technical Loss Detection Using Smart Distribution Network Measurement Data,» de Innovative Smart Grid Technologies – Asia (ISGT Asia), IEEE , Tianjin, 2012.
W. Niemira, R. B. Bobba, P. Sauer y W. H. Sanders, «Malicious Data Detection in State Estimation Leveraging System Losses & Estimation of Perturbed Parameters,» de International Conference on Smart Grid Communications (SmartGridComm), IEEE, Vancouver, BC, 2013.
S. Weckx, C. Gonzalez, J. Tant, T. De Rybel y J. Driesen, «Parameter Identification of Unknown Radial Grids for Theft Detection,» de IEEE PES International Conference and Exhibition on Innovative Smart Grid Technologies (ISGT Europe), Berlin, 2012.
S. Salinas, M. Li y P. Li, «Privacy-Preserving Energy Theft Detection in Smart Grids: A P2P Computing Approach,» IEEE Journal on Selected Areas in Communications, vol. 31, nº 9, pp. 257-267, 2013.
W. Han y Y. Xiao, «NFD: A Practical Scheme to Detect Non-Technical Loss Fraud in Smart Grid,» de IEEE International Conference on Communications (ICC), Sydney, NSW, 2014.
A. A. Cardenas, S. Amin, G. Schwartz, R. Dong y S. Sastry, «A Game Theory Model for Electricity Theft Detection and Privacy-Aware Control in AMI Systems,» de Allerton Conference on Communication, Control, and Computing (Allerton), Monticello, IL, 2012.
S. Amin, G. A. Schwarz, A. A. Cardenas y S. S. Sastry, «Game-Theoretic Models of Electricity Theft Detection in Smart Utility Networks: Providing New Capabilities with Advanced Metering Infrastructure,» Control Systems, IEEE, vol. 35, nº 1, February 2015.
J. Nagi, K. S. Yap, S. K. Tiong, S. K. Ahmed y M. Mohamad, «Nontechnical Loss Detection for Metered Customers in Power Utility Using Support Vector Machines,» IEEE Transactions on Power Delivery, pp. 1162-1171, 13 October 2009.
J. Nagi, K. S. Yap, S. T. Tiong, S. K. Ahmed y F. Nagi, «Improving SVM-Based Nontechnical Loss Detection in Power Utility Using the Fuzzy Inference System,» IEEE Transactions on Power Delivery, pp. 1284-1285, 2011.
S. S. S. R. Depuru, L. Wang y V. Devabhaktumi, «Support Vector Machine Based Data Classification for Detection of Electricity Theft,» de IEEE/PES Power Systems Conference and Exposition (PSCE), Phoenix, AZ, 2011.
S. S. S. R. Depuru, L. Wang y V. Devabhaktuni, «Enhanced Encoding Technique for Identying Abnormal Energy Usage Pattern,» de North American Power Symposium (NAPS), Chanpaign, IL, 2012.
T. V. Babu, T. S. Murthy y B. Sivaiah, «Detecting unusual customer consumption profiles in power distribution systems — APSPDCL,» de IEEE International Conference on Computational Intelligence and Computing Research (ICCIC), Enathi, 2013.
P. Faria, Z. Vale, P. Antunes y A. Souza, «Using Baseline Methods to Identify Non-technical Losses in the Context of Smart Grids,» de IEEE PES Conference on Innovative Smart Grid Technologies Latin America (ISGT LA), Sao Paulo, 2013.