Research On Short-term Wind Speed Prediction Method For Wind Farm

With the global energy crisis and environmental pollution becoming more and more serious,the development of renewable energy has become the key point to protect the environment,save energy,reduce emissions and realize the sustainable development of human society.As a pollution-free renewable energy source,wind energy has been widely used.Due to the inherent randomness and volatility of wind speed,large-scale grid integration of wind power will inevitably bring huge challenges to the stable and safe operation of the power system.Accurate wind speed prediction for wind farms is an effective way to ensure the stability of the wind power system and increase the utilization rate of wind energy.Due to the randomness and instability of wind speed,it is difficult to achieve accurate wind speed prediction.Based on the characteristics of wind speed,this paper establishes single prediction models,combined prediction models based in signal decomposition technology and combined prediction models based on residual correction strategy for short-term wind speed prediction.The short-term wind speed prediction is performed on four wind speed data sets respectively,and the prediction errors are compared and the reasons for the errors are analyzed.The main contents of this paper are as follows:(1)According to the characteristics of the target study area,the characteristics and changing rules of the four original wind speed series are analyzed.Based on the characteristics of wind speed time series,the working principles of time series analysis method,BP neural network,Elman neural network and LSTM neural network are analyzed,and the order determination method of ARIMA model is studied.The mentioned linear and nonlinear single prediction models are constructed,and the relevant parameters of the prediction models are analyzed.The experimental results show that LSTM is suitable for the prediction of non-stationary wind speed series,followed by Elman.(2)Aiming at the shortcomings of the single models with different prediction mechanisms and huge prediction errors,a data preprocessing technology is introduced:signal decomposition technology.The wavelet decomposition,empirical mode decomposition,complementary ensemble empirical mode decomposition and empirical wavelet transform are employed to decompose the original wind speed series.Based on the characteristics and performance of single prediction models,long short term memory neural network is established to predict low-frequency components and the Elman neural network is used to predict high-frequency components to balance the calculation efficiency and prediction accuracy.Training and testing on four actual wind farms to verify the feasibility and superiority of the signal decomposition technology,as well as the advantages of empirical wavelet transform in the decomposition of wind speed signals.(3)Based on the performance of the combined prediction models based on signal decomposition method,the residual series of the prediction results of the empirical wavelet transform are analyzed.The sample entropy,stationarity and autocorrelation of the residual series are studied.Aiming at the high disorder and complexity of the residual series,the deep neural network architecture is applied to establish a deep long short term memory neural network with three hidden layers for residual correction.The results show that residual correction can further improve the accuracy of wind speed prediction,and the deep long short term memory network can improve the prediction accuracy of complex,high-frequency and chaotic residual series.