Temperature prediction study of cable joint conductor based on the PSO algorithms of BP neural network

摘要

Summary form only given. In power system, cable line is a quite important equipment while cable joint is one of its weak link, which may cause severe losses once break down . Allowing that the cable joint will be accompanied by a rise in temperature no matter what kind of failure occurs, a principle that real-time monitoring and predicting temperature of cable joint is available to own a comprehensive understanding of the cable line's operation, and discover the hidden trouble in cable and improving the safe operation performance of cable wires . Compared with the low monitoring configuration sensitivity, high maintenance cost and low prediction accuracy of traditional monitoring methods, this paper proposes an algorithm based on the PSO(Particle Swarm Optimization) of BP neural network to predict the temperature of wire cable joint which can make up for the present method's temperature detection insufficiency of the cable joint conductor effectively and improve the faults of low prediction accuracy, high initial weight sensitivity of the single BP neural network purposefully . For the collected temperature monitoring data of cable joint, first of all, we should delete some error data that because of misoperation or measuring error, find out the factors which closely associated with forecast . and then imported in MATLAB for data classification, weight setting, data normalization preprocessing, etc . The last is to build the network . In this paper, it is assigned that the individual values optimized by PSO algorithm to the BP network's initial weight and threshold, through training, simulating and obtaining the prediction results for building the new BP neural network . The simulation results show that the BP neural network whose initial weights and threshold optimized by the PSO algorithm has higher prediction precision of conductor temperature and good local searching ability, and also possess the lower probability of network into a local optimum, compared wi- h the single BP neural network . The simulation results as shown in figure 1 Prediction curve of the BP neural network and the PSO algorithms of BP neural network; Figure 2 Error curve of the BP neural network and the PSO algorithms of BP neural network