Application Research Of Neural Network And Particle Swarm Optimization In Fault Diagnosis Of High Voltage Electric Power Metering System

With the popularization of electric power and the sustained growth of electricity demand, electric grids are being more widely built and the power metering system is developing and more complex structured, which result in the increase of fault probability. Whether the internal fault of electronic components or Man-made Fault produced by artificial changes of wiring, will cause inaccurate measurement or even power outages, thus seriously affect the nation-building and people’s daily life. Therefore, fault diagnosis technology of high voltage electric power metering system came into being and has received widespread attention of the power sector and researchers.According to the structure of high voltage electric power metering system and characteristics of its faults, this paper will apply digital filtering, neural network and particle swarm optimization(PSO) to signal extraction and fault diagnosis and make the following analysis.Firstly, the structure of high voltage electric power metering system will be introduced. This paper will give a brief introduction of the structure and working principle of Current Transformer(CT) and Potential Transformer(PT). According to the classification of elements fault, the common faults of CT and PT in the metering system are analyzed and fault detection methods are put forward, especially for hidden failure like short-circuit fault of CT. The faults will be analyzed theoretically and a test formula will be deduced and verified through the actual signal.Then, an overall design for fault detection circuit will be proposed. According to the structure of the metering system and the characteristics of its faults, a fault detection circuit will be designed and tested by building up verification platform. An equiripple linear-phase FIR digital filter will be designed, with which the collected fault signals will be extracted for a simple analysis in different load conditions.Finally, the BP neural network and PSO algorithm are applied to the fault diagnosis of high voltage electric power metering system, the common fault types will be identified. Fault features are extracted from the system operating parameters and data samples are normalized. Then, the appropriate network structure and parameters are selected and BP neural network model is constructed. The collected sample data are used to train and test the network. At last, particle swarm optimization was used to optimize the weights of BP neural network and a fault diagnosis system is designed based on particle swarm optimized neural network. Finally, by comparing the fault diagnosis system based on PSO-BP and the diagnostic tests of BP network, the results indicate that the fault diagnosis methods for high voltage electric power metering system adopted in this paper are feasible and effective.