Event-Driven state estimation in electric distribution systems

Official URL: http://risc01.sabanciuniv.edu/record=b1655284 (Table of Contents)

The nature and shape of the modern power grid is evolving as it is becoming more intelligent, flexible and more reliable. The increasing penetration of renewable energy resources is increasing the complexity of the modern power distribution networks and is posing new challenges in terms of system monitoring and control. State estimation is one of the most critical component of energy management systems employed in monitoring and control centres of transmission systems. Although state estimation algorithms are well developed for transmission networks and have been in operation for the last few decades, in distribution networks these tools are likely to face more challenges comparatively such as network observability due to lack of underlying metering network and frequently changing network topology. Therefore, this dissertation presents novel distribution system state estimation (DSSE) algorithm that incorporates eventdriven topology processor in the presence of topological switching, integrates Artificial Neural-Network based forecasting model into SE formulation which solves the network observability problem, and finally evaluates different classes of static and dynamic estimators for distribution systems and proposes unscented Kalman filter based robust state estimator. Topology processing is an important function of state estimator that determines the operational network configuration prior to carrying out the estimation process. This is very crucial because wrong topology information can result in wrong state estimates which in turn will lead to generation of wrong control sequence. The dissertation presents an event-driven topology processor implementation for electric distribution system. This implementation includes detection of change in system configuration (or topology) of distribution system followed by identification of correct topology out of many critical configurations stored in a model bank. Due to lack of metering infrastructure at the distribution level, this algorithm works with less available real-system measurements. This algorithm is validated for IEEE 6 bus, IEEE 14 bus and a practical Medium Voltage Danish Electrical Distribution network and the results are presented in Chapter 3 of this dissertation. Network observability is second important function that need to be carried out before the execution of DSSE algorithm. Due to lacking metering infrastructure at the distribution system level there are very few or no real-time system measurements, which are insufficient for making the distribution networks observable. In order to come up with system state estimate, it is required that network be observable, which means that state estimation should be able to converge to a unique solution. In this dissertation, network observability of electric distribution networks is solved by using load estimation. An ANN-based state estimation technique is proposed to achieve network observability in distribution systems. This algorithm uses demand forecasted load data for estimating system state and generating the load estimates (or pseudo measurements). The algorithm is validated for IEEE test networks and a real-life medium voltage distribution network for diverse operating conditions, and its effectiveness is justified. Information from topology processor and observability analyzer is used by the SE algorithm for determining the system state. The literature contains different classes of estimator (static and dynamic) applied to the power systems especially distribution systems. This dissertation involves evaluation of both static and dynamic state estimators such as Weighted Least Squares (WLS), extended Kalman filter (EKF), unscented Kalman filter (UKF) for distribution systems. An improvement in the conventional UKF algorithm, namely IUKF, is proposed and validated for distribution systems for cases of high measurement noise. In order to show its effectiveness, it is compared with the conventional UKF under different noise scenarios, and it is shown that IUKF outperforms UKF. IUKF is also compared with other estimators (WLS, EKF). Various statistical performance metrics such as Maximum-absolute-deviation (MAD), Maximum-absolute-percent error (MAPE), Root-mean-square-error (RMSE) and overall-performance-index (J), are used to quantify the comparative analysis under various measurement noise scenarios. The existing and the proposed IUKF algorithms are validated for IEEE test distribution networks such as IEEE 30, 33 and 69 bus systems.