Fault diagnosis in power systems-substation level-through hybrid artificial neural networks and expert system

This paper presents an application of artificial neural networks (ANNs) and expert system (ES) for offline fault diagnosis in power systems using the information of the operated relays and tripped circuit breakers after they reached their final status. The hybrid system candidates the faulted section(s) even in the case of multiple faults. The developed system also has the ability to identify the most probable faulted section based on the depicted performance indices (K_c & K_f), the calculation of which is based on the relay operation classifications. Relays are classified into four categories, namely relays operated correctly as primary protection, relays operated correctly as backup protection, relays failed to operate and malfunction relays. The proposed ANNs identify the faulted section(s) even if the fault is in the inured zone (zone between circuit breaker and current transformer). Moreover, full analysis of breaker operations such as correct, fail to trip and breakers tripped manually operations can also be obtained. The communication problems in the relay signals either missing or noisy signal, can be detected as well. The proposed ES module used the rule-based technique. Only the rules to classify the relay operations are used. The fault type using the monitored short circuit currents is utilized especially for the solid faults. A comparison overview between ANNs, ESs and the hybrid intelligent system for fault diagnosis is presented as well. Neural networks are trained and tested using Stuttgart Neural Network Simulator (SNNS) under SuSE Linux 6.1 on a personal computer. The neural network simulator has the ability to generate a C/C++ source code that may be compiled separately to be used under DOS/UNIX, which is an important portability feature. While the proposed ES has been developed and implemented using XPCE/Prolog facilities of SWI. Finally, test results are highlighted to show the viability of the proposed hybrid intelligent system