Short-Term Wind Speed Forecasting Of Wind Farm Based On Wavelet Transform And Deep Belief Network

In the context of the global energy crisis and fossil fuel pollution,energy strategy gradually focus on the development and application of new energy sources.Wind energy,as a kind of non-polluting and renewable energy sources,has special privileges of abundant reserves and mature technology,which can be developed in a large scale,and it is regarded an ideal choice among many new energy sources.At present,the world has gradually liberalized the wind power development policy,and increased investment in wind energy as well.Wind energy has the characteristic of uncertainty and intermittent and randomness,which seriously affects the stability of power grid.Therefore,more rapid and accurate wind speed forecasting methods are required to improve the stability of the grid integrating wind power and reduce its influence on the whole power system.Consequently,the research of wind speed forecasting method of wind farm is of great significance for the development of wind resources on a large scale.This paper proposed a framework of deep wind speed prediction network framework based on deep belief network mapping.Stacking the binary RBM layer by layer hierarchically and adding a logistic regression to the end of the stacks creates a DBN.The framework utilizes an unsupervised greedy layer-by-layer training method to obtain the initial parameters of the DBN,and then implement supervised fine-tuning the predict network to reaches its error minimization.The framework can effectively exploit and extract the useful features of the time series,and avoid being tracked into the small range optimal solution.Recognizing this challenge,a novel wind speed forecasting method based on wavelet transform and deep belief network(DBN)is proposed.Firstly,the wavelet transform is used to decompose the wind speed sequence into sub sequences with different frequencies,and then the different sub DBN models are established respectively.The input of the sub model DBN corresponds to the sub frequency sequence,and the output of the sub DBN model corresponds to the predicted value of the sub frequency sequence.Finally,the predicted results of the sub DBN model are reconstructed to obtain a complete wind speed prediction results.The proposed method is based on the MATLAB platform to write corresponding program for simulation and analysis.The validity of the proposed method is verified by comparing the three typical methods of time series prediction based on autoregressive moving average(ARMA),BP neural network(BPNN)and Morlet wavelet neural network(MWNN).Moreover,the characteristics of the time series of high order nonlinear and non-stationary features can be learned better based on the proposed method of wavelet transform and deep belief network.The simulation results show that the proposed method is much more accurate and better than the typical shallow prediction methods.Furthermore,the proposed method also has low volatility of the error index in many independent runs and a good performance in the multi-step prediction.