Research On Short-term Wind Speed Prediction Model Of Neural Network Based On Multivariate Time Series

In recent years, the development and utilization of new energy has become an important research topic and wind energy is widely used in the world as its great potential. Wind power is the most important form of wind energy utilization. Wind speed determines the output of wind turbines, and the volatility and intermittence of wind speed will increase the difficulty of wind power’s controllability. When large-capacity wind power grid appears, the effects become immeasurable and may even cause the instability of wind power, what will damage the grid power quality and the safety and stability of the grid operation. So accurate prediction of wind speed can greatly reduce the effects of random wind speed on stable operation of wind farm, thus the grid can timely scheduled.This paper focuses on short-term wind speed prediction model in the wind farm under the support of National Natural Science Foundation of China (No.51277127). It considers the effect of wind direction on wind speed, and uses the method of unitary time series combined with neural network to establish wind speed prediction model. But the modeling process is complicated. So this paper puts forwards a new method of neural network based on multivariate time series for short-term wind speed prediction. The main research jobs are summarized as follows: (1) It expatiates the background and significance of wind speed prediction, the domestic and foreign wind farm development currently; summarizes research situation, methods and main problems for wind speed prediction at home and abroad.(2) It learns the characteristics and main factors of wind speed diversification, moreover the paper analyzes the characteristics of wind speed in wind farms and the impacts on the grid.(3) The paper studies modeling theory of neural network, and uses particle swarm optimization algorithm to improve BP neural network. It verifies the effectiveness of adding wind direction by grey correlation method to analyze correlation degree of wind speed and wind direction. Then it determines the larger correlate wind direction with wind speed at next time through correlation coefficients. Then it uses wind direction calculated, historical wind speed and residual which are determined by unitary time series as the input variables of the neural network. It establishes neural network prediction model based on unitary time series of adding wind direction. The simulation results show that it’s effective for adding wind direction to improve wind speed prediction accuracy.(4) When calculating correlation degree and correlation coefficients, it relates to the selection of correlation formula, resolution ratio and the determination of correlation degree. Therefore, the paper learns a new method of neural network based on multivariate time series to improve the accuracy. It uses the number of historical wind speed, wind direction and regression residuals determined by multivariate time series model as the number of input variables of the neural network. The simulation results show that the proposed method is feasible and effective to improve wind speed prediction accuracy.