Study On Hybrid Prediction Method Of Short-term Wind Power Based On Measured Data

Compared with the traditional power generation system,wind power has the characteristics of volatility,intermittent and randomness,and affects the stability of the entire power system operation seriously.Therefore,the accurate prediction of wind power has become a hot issue for many scholars.In this thesis,the empirical mode decomposition(EMD),the genetic algorithm(GA)and wavelet neural network were integrated and a hybrid prediction method of short-term wind power based on measured data was propsed to improve the prediction accuracy of short-term wind power.The characteristic curve of wind power was fitted based on measured data.On the basis of wind speed variation and wind power prediction,the historical measured data of Jiugongshan wind farm were collected and analyzed,and the smoothing function was used to select and delete the bad data.The processed wind farm data were applied to fit the characteristic curve of wind power.The fitting accuracy was improved and a good fitting result of characteristic curve of wind power was obtained.First of all,the prediction method of wind power based on BP neural network and the prediction method of wind power based on wavelet neural network were analyzed and compared.BP neural network and wavelet neural network were discussed,and the prediction method of short-term wind power based on BP neural network and the prediction method of short-term wind power based on wavelet neural network were studied respectively.The prediction performance of two kinds of wind power prediction methods using Jiugongshan wind farm data were analyzed and compared.The simulation results showed that the prediction performance of wind power based on wavelet neural network was better than that of wind power based on BP neural network.Secondly,the hybrid prediction method of short-term wind power based on GA-wavelet neural network was studied.To solve the sensibility of wavelet neural network to the initial parameters,the genetic algorithm was used to optimize the initial value of wavelet neural network,and the optimization results were taken as the initial parameters of the wavelet neural network to train the network.The simulation results showed that the hybrid prediction method based on GA-wavelet neural network could reduce the training error of wavelet neural network,and improve the prediction accuracy of short-term wind power.At the end of this paper,the hybrid prediction method of short-term wind power based on EMD-GA-wavelet neural network was proposed.Considering the non-stationarity of wind power series,the empirical mode decomposition method(EMD)was introduced.The hybrid algorithm based on genetic algorithm and wavelet neural network was used to predict.The simulation results showed that the hybrid prediction method of short-term wind power based on EMD-GA-wavelet neural network could achieve a good prediction result.