Estimadores de estado

RTE-VT workshop, Paris, May 29-30, 2006

1

State Estimation in PowerFactory: Algorithmic Aspects
Jochen Alber DIgSILENT GmbH, Heinrich-Hertz-Str. 7 D-72810 Gomaringen, Germany
Email: jalber@digsilent.de

Abstract— This paper will give an insight into Power System State Estimation as it is implemented in the software PowerFactory. The State Estimator function in question consists ofseveral independent components, namely: Measurement Data Plausibility Checking, Observability Analysis, State Estimation (non-linear optimization), and Bad Data Checking. We will focus on new algorithmic aspects within most of these stages. Besides, we will address the performance issues of the presented algorithms. Our main interest, however, will be on a new approach for the Observability Analysiswhich is based on the close investigation of the system's sensitivities. We draw a one-to-one correspondence of the observability of a network to the rank of a corresponding sensitivity matrix. A new concept of fast pseudo-rank calculation is developed. This method then can be used to calculate the observable parts of the network, as well as to compute a classification of the measurements'redundancies. The performance of the algorithm is tested on a real-world network (with data gained from an underlying ABB MicroSCADA system) and compared to common rank calculation with singular value decompositions on sparse matrices. Index Terms—State Estimation, Observability, Sensitivity, Optimization methods, Software

For a general overview, let us first of all have a brief glance at the completeframework of the State Estimator module as provided by the above mentioned software package. A. Components of the PowerFactory State Estimator The State Estimator in PowerFactory comprises the following individual components: 1) Measurement Data Plausibility Checking 2) Observability Analysis 3) State Estimation (Non-Linear Optimization) Figure 1 illustrates the algorithmic interaction of thedifferent building blocks. In a first phase, the Measurement Data Plausibility Checking is sought to detect and separate out all measurements with some apparent error in order to avoid any heavy distortion of the estimated network state due to completely wrong measurements. Various test criteria can be thought of (such as checking for large measured branch flows on open ended branches; checking forconsistent measured power flow directions at each side of the branch elements; checking for reasonably measured node sums, etc.). In a second phase, the network is checked for its Observability. Roughly speaking, a region of the network is called observable, if the measurements in the system provide enough (nonredundant) information to estimate the state of that part of the network. A necessaryrequirement for an observable system is that the number of available measurements is equal or larger than the number of estimated variables. But it can also happen that only parts of the network are observable and some other parts of the system are not observable even if the total number of measurements is sufficient. Hence, it is not only important that there are enough measurements, but also thatthey are well distributed in the network. The entire network is said to be observable if all states can be estimated based on the given measurements. If a network is not observable, it is still useful to determine the islands in the network that are observable. Three main approaches for observability analysis are distinguished in

I. INTRODUCTION tate estimation is the task to provide consistentload flow results for an entire power system, based on real time measurements, manually entered data and the network model. The problem has been studied in great detail over the last years (see [5], [7], [8] and the references therein for a comprehensive overview). A general input instance for a state estimation algorithm includes a set of measurements located in the network and a selection of...