Short-term Power Load Forecasting Based On BP Neural Network Optimized By GA-PSO Algorithm

With constantly developing of power market, forecasting electricalload becomes an important work for the power system managementdepartment. Accurately making a prediction for power load can bettermake power network planning and power units of maintenance plan, andmore reasonably arrange operation mode of power network. It is of greatsignificance to improve economic benefit and social benefit of powerenterprises, keep power system safe and stable, and ensure people’s dailylife in order.Firstly, the thesis introduces the research background, research statusat home and abroad, and research significance of power load prediction.This thesis describes the basic theory of power load prediction. Secondly,the key technologies of modern prediction were introduced in detail. Thethesis introduces the basic theory of artificial neural network. This thesisparticularly studies BP neural network, include the structures,corresponding learning algorithm steps, advantage and disadvantage.Meanwhile characteristics and fundamentals of Genetic Algorithm andParticle Swarm Optimization Algorithm are also researched and analyzed.Thirdly, the thesis design forecasting model of BP neural network,include the input layer, output layer, hidden layer and the definition oftransfer function. Finally, genetic algorithm (GA), particle swarm optimization (PSO) and GA-PSO algorithm are respectively applied tooptimize the weights and thresholds of BP neural network. Establish aforecasting model of GA-BP neural network, a forecasting model ofPSO-BP neural network and a forecasting model of GA-PSO-BP neuralnetwork. The historical load data, historical temperature and date in acertain region of European were used for simulation test. Forecast thehourly load of24hours in a day this area. The prediction results wereanalyzed to compare the performance of forecasting models. Thesimulation results showed that the forecasting model of GA-PSO-BPneural network not only speeds up the convergence rate of the neuralnetwork, but also improve the prediction precision of the short-termpower load.