Application of GA-BP in Fault Diagnosis of Power Circuit of SVC

The multi-layer feed-forward neural network is in essence a dynamic system including a large number of interconnected processing elements (neurons) working in unison to solve special problems, and this characteristic makes it suitable for the fault diagnosis. If we regard fault signs as inputs of the network and fault causes as outputs, we can construct a network to map the complicated relationship between inputs and outputs. However, the unavoidable shortcomings of the back-propagation (BP) algorithm typically adopted by feed-forward neural network limits its use. The main problem is that gradient methods employed by classical BP algorithm find only a local optimum, the local optimum found depends on the starting point and the goal function must be smooth. From the viewpoint of mathematics, genetic algorithms (GA) is a kind of technique for searching optimal solutions. In this paper, we optimize the BP network weights by genetic algorithms, and then the GA-BP network is applied to the fault diagnosis of power circuit of the static VAr compensator (SVC) based on digital signal processor (DSP) TMS320F240. The experiment shows the performance of the system is excellent