Online Short-circuit Current Forecast Based On Improved BP Neural Network

Since the 20th century, with a large capacity generator sets and a large number of substation equipment to put into use, increasing power load, as well as wind power nuclear power and other new energy to access, China's power grid has been entered a new period of rapid development. Interconnection between the emergence of the load center of large power plants and large power system, making the whole grid shows large, complex heterogeneous characteristics. At the same time, problems also have been appeared, in which a particularly prominent problem is the short-circuit system irregular changes in the size and distribution of current levels occurred with the development of the grid. Thus, according to the changes of power load in the power system's and natural conditions, it is becoming an increasing urgent to quickly calculate and forecast short-circuit current, which has become one of the important topics for power grid to plan and construct electrified wire netting and take measures to limit the short-circuit current, etc.This paper is mainly about online short circuit current based on improved BP neural network prediction study. According to the conventional method of calculation and prediction power system short-circuit current, the basic theory of BP neural network method is studied intensively. In order to overcome the slow convergence speed and the initial value of the improper selection which lead to falling into local minimum points. Firstly, we adopt adaptive Chaos Particle Swarm to optimize BP neural network layers between the initial weights and thresholds, and then the BP network is trained, resulting in improved BP neural network.Through the analysis of various characteristics of short-circuit current, considering the impact of the power load, the generator output, date and type of weather and other factors on the short circuit current, and take these factors as the input feature amount BP neural network. We dispose the abnormal data of the load value of the node in the input characteristic variables and generate input samples in complying with the active power constraints and uniform distribution. Normalization processing is taken in input samples and quantization processing is taken in temperature, weather conditions and date type. We will calculate the precise methods as output expectations and to establish a short-circuit current online forecasting model based on improved BP neural network.Finally, we predict the short circuit current of one day's 24 hours, comparing the short-circuit current forecasting model based on Improved BP Neural Network and Online short-circuit current forecasting model based on BP neural network, equivalent model based on precise and calculation method to obtain short-circuit current actual value The results show that the established online short-circuit current forecasting model based on improved BP neural network which can improve the prediction accuracy and speed, and get a higher degree of fitting short-circuit current actual value, thus, the prediction performance is better than forecasting model based on BP neural network. It is also shows that the improved BP network training time is relatively short for the 139s. Entire prediction operation time is less than 200s.