Alarm processing in electrical power systems through a neuro-fuzzy approach

This work presents a methodology that combines the use of artificial neural networks and fuzzy logic for alarm processing and identification of faulted components in electrical power systems. Fuzzy relations are established and form a database employed to train artificial neural networks. The artificial neural networks inputs are alarm patterns, while each output neuron is responsible for estimating the degree of membership of a specific system component into the class of faulted components. The proposed method allows good interpretation of the results, even in the presence of difficult corrupted alarm patterns. Tests are performed with a test system and with part of a real Brazilian system.

I. Introduction

In Modern control centers, system operators usually have to handle a large number of alarms and messages in real time, and to take decisions on power system operation. These alarms may be related to fault occurrences, protection devices misoperations, etc. In many cases it is very difficult and time-consuming to draw conclusions about what has happened, particularly when protection schemes does not operate properly, communication failures occur, corrupted data are processed, etc. Following fault occurrences or other disturbances, it is crucial to restore system normal operating conditions as soon as possible. Then, alarm processing and diagnosis, including the identification of faulted devices, becomes a very important task to be addressed in real-time.