A comprehensive intelligent framework for fault analysis and diagnosis in power transmission networks

Abstract

This thesis presents an algorithm for fault analysis and diagnosis in the power transmission network of New Zealand. When a fault occurs in a transmission network, it must be identified and eliminated as soon as possible. Since control centres are flooded with hundreds of alarm messages during a fault, fault diagnosis, which involves the analysis of alarm messages, is a time consuming task.

To develop a fault diagnostician, we identified the essential components that are necessary for successful development of a diagnostician. Using the essential components, a cooperative algorithm (CoHAF AD) was developed. CoHAF AD has three main components: the analytic and the diagnostic algorithm, the learning algorithm and the other components. The analytic and the diagnostic algorithm, HiMoBF AD, uses four hierarchies of structural and functional knowledge and incorporates both model-based and heuristic knowledge.

The fault analysis is also done in four hierarchies with different levels of abstraction. Level I performs analysis at 'componentry' level and level 2 analyses the switching operations at switching group level. Level 3 of this analysis includes a topology-based dynamic network of neural-nets which analyse the faults at 'cluster' level. The final level, level 4, abstracts all the outputs from other levels and provides analysis at complete power network level. The diagnostic algorithm of HiMoBF AD does the final diagnosis and addresses several sub-problems of power network fault diagnosis such as 'multiple-fault detection' and 'handling missing and inaccurate fault information', and provides solutions to them.

CoHAFAD also includes a learning mechanism, HiMoB-Learn, that learns by 'remoulding' the existing knowledge of HiMoBFAD. The learnt rules are purely based on recognising an inference process for a fault-symptom relation, generalising the solution, and then saving the solution for future use. The learning algorithm is based on Explanation Based Generalisation (EBG) algorithm but incorporates a different generalisation mechanism that is developed to handle HiMoBFAD's unique knowledge representation. The mechanism is a schema-based algorithm that can generalise model-based knowledge of HiMoBFAD.

All levels of HiMoBFAD, HiMoB-Learn, and other components such as the topology processor and user interface, have been developed as individual components that work under a co-operative environment. The complete algorithm, CoHAFAD, is thus a cooperative algorithm. The working details, their method and task decomposition details are also described.

The results of this thesis are being used in practical implementation of the algorithm for New Zealand's power transmission network.