Nonlinear Signal Forecasting Based On Wavelet Decomposition And Neural Network

Prediction of nonlinear signals has always been an important branch of signal processing.Establishing a predictive model for nonlinear signals can better reveal the nature and laws of things,and has important research significance and practical value for guiding human life.Wind speed prediction is of great significance for wind farm management and plays an important role in wind power grid connection.The convective thermal bubble triggered by the heating of the underlying surface of the sun causes the turbulent flow characteristics of the near-surface layer.So the wind speed has great fluctuations and disturbances.It is difficult to predict with a single method.In this paper,the relationship between the predicted wind speed and the historical wind speed and the historical wind speed of the surrounding stations is studied.A hybrid multi-step wind speed prediction model is established and compared with the existing wind speed prediction model.This study is located in eastern China.Wind speed data of two wind farms in Shandong.The main work includes:1.Experiment with BP neural network:(1)predicting multi-step future wind speed by using historical wind speed.(2)predicting wind speed of experimental site by using historical wind speed of surrounding sites and experimental sites.(3)utilizing surrounding sites The historical wind speed is ahead of the multi-step prediction of the wind speed of the experimental site.2.Improve single hidden layer feedforward neural network(SLFN).The wavelet analysis and neural network are used to establish combined prediction model for short-term wind speed prediction.The cuckoo search optimization algorithm(CS)is used to replace gradient descent method in network.The experimental results show that these improved models can improve the wind speed prediction accuracy.In particular,the CS algorithm can quickly and effectively find out the various parameters in the neural network,and achieve global optimization,so that the prediction results are stable and high precision.3.Three multi-step prediction models(CS-WD-ANN,CS-WNN,CS-WD-WNN)based on wavelet decomposition(WD),wavelet neural network(WNN)and cuckoo search algorithm(CS)are proposed.Comparison with the existing wind speed prediction model shows that CSWD-WNN performs best and the prediction results have the smallest statistical error.