MODEL-BASED LEARNING FOR FAULT-DIAGNOSIS IN POWER TRANSMISSION NETWORKS

This paper presents a hybrid algorithm for fault diagnosis in power tr ansmission networks. The prime objective of a power transmission netwo rk is to supply power to the customers and to meet the lend demands. W hen a fault occurs in a transmission network, it must be identified an d eliminated as soon as possible. Since control centres are flooded wi th hundreds of alarm messages during a fault, fault diagnosis, which i nvolves the analysis of alarm messages, is a time consuming task. Faul ts in a power network are caused by various environmental conditions w hich change with time and their diagnosis involves real time decision making which cannot be modelled by using traditional mathematical mode lling techniques. Earlier research suggested that decision-support sys tems can aid system controllers during emergency situations. Artificia l intelligence techniques such as expert systems, neural networks and model based reasoning have been used in the past but these techniques lack efficiency when applied to large systems such as power networks. As part of our an going research towards the development of an intelli gent architecture for fault diagnosis, we attempt to develop an effici ent model by combining two reasoning techniques (model-based and heuri stic) and incorporating a machine learning algorithm called Explanatio n Eased Generalisation (EBG) for automated knowledge acquisition. The machine learning algorithm learns shallow knowledge from the reasoning process of the combined (or hybrid) model and stores it for future pr oblem solving. In this paper, the developed hybrid fault diagnosis sys tem is discussed and its performance is compared with the traditional model based reasoning approach.