Study On Short-term Wind Power Prediction Method Based On IPSO-GA Neural Network

The wind power has the disadvantages of intermittence and randomicity, with theincreasing proportion of wind connected to power grid, it will bring challenge to thesafety and stabilization of power grid and power quality, and then restrict the scale ofwind power development to some extent. It is an effective approach for the aboveproblem to carry out short-term wind power prediction, but in China, the relatedstudies just developed in recent years and there are still many shortages in the forecastprecision，the reliability and the adaptability of different wind farm. The powerprediction model based on the historical data and wind farm power short-termprediction method based on IPSO-GA neural networks has been studied, thus theprediction accuracy is improved. In this paper, the following aspects of work areconducted:Firstly, wind field measurement data are pre-processed, a relatively completedatabase is established, and the pre-processing principle are studied. Because of theexcellent characteristics of neural network, artificial neural network is usedin the prediction of wind farm power is proposed. Through comparative study betweenRBF network and BP network, the RBF network was considered to be more suitablefor wind power prediction than BP network, wind power predicion model based onRBF neural network is eastablished, which has clear physical meanings and highprediction precision. As the high correlation of the samples can cause the lowprediction accuracy of RBF neural networks, then principal component analysismethod is used to define a linear combination of original varibles as new varibles, so asto eliminate the correlation between samples, thus the samples are optimized to someextent.Secondly, the parameters of the RBF neural network are more difficult to decide,and these parameters directly determine the network performance. As particle swarmoptimization algorithm has advantage of simple principle and high convergence rate,PSO is considered to optimize the parameters. But the refresh scheme of particle isrelatively single, crossover operator and mutation operator of genetic algorithm isintroduced, and then a hybrid optimization algorithms based on improved particleswarm optimization and genetic algorithm(IPSO-GA) is developed to improve theconvergence performance and optimization capabilities of the algorithm. The proposed algorithm is adopted to handle three nonlinear functions, the results show that it ismore suitable for the nonlinear objective function optimization and has higherconvergence and accuracy.Finally, IPSO-GA is used to realize full structure optimization of RBF network,hierarchical encoding is treated as the parameters encoding of the RBF network, whichis composed by two parts of the control genes and parameter genes. Then a wind farmpower short-term prediction method based on IPSO-GA neural network is proposed.The short-term wind power prediction result of a wind farm in Guizhou shows thatRBF neural network has faster convergence speed and better generalizationperformance than BP neural network, and optimized RBF network has greaterprediction accuracy.