Research On Prediction Method Of Ultra-short-term Wind Speed Of Wind Farm Based On Normal Distribution Noise Neural Network

Wind power generation is subjected to the constraints of intermittent,volatility and uncertainty and other factors,which brings severe challenges to the security and stability of integration of wind power.The ultra-short-term wind power prediction is mainly applied in the adjustment and control of generating units in the short-term to mitigate the adverse effects of wind power generation on grid and effectively increase the capacity of wind power grid.But wind speed is closely related to wind power,accurate wind speed prediction is the basis and prerequisite of power forecasting,which makes the research of the ultra-short-term wind speed prediction great meaningful on the wind farm.This paper,taking the historical wind speed from the wind tower of a certain wind farm in Ningxia as the research object,conducting over the methods research of ultra-short-term wind speed forecasting by optimizing the NDN neural network.1)A Wind Input Matrix(WIM)is established,which makes the wind speed at adjacent time points have a strong temporal correlation,and the adjacent dimension of the matrix has strong spatial correlation.This is illustrated by comparing the traditional wind speed input matrix with the WIM under the ultra-short-term wind speed prediction of the BP neural network model.The normal distribution noise(NDN)neural network is constructed by introducing the normal distribution random noise to BP neural network,and the ultra-short-term wind speed is predicted by use of BP and NDN neural networks based on the WIM.It is found that the wind speed prediction accuracy of the NDN neural network model is higher than that of the BP model,but there is a delay in the prediction time.2)The Particle Swarm Optimization(PSO)algorithm and the Firefly Algorithm(FA)are used to optimize the NDN network weights and thresholds,respectively.The simulation results show that the prediction effect of PSO-NDN model is better than that of the NDN model,but forecasting time delays still exist in between the prediction curve of the PSO-NDN model and the measured curve.The experimental results show that the prediction accuracy of FA-NDN neural network model is poor and the prediction accuracy is not as good as that of NDN model,which is only slightly higher than that of traditional BP network model.3)By introducing wavelet decomposition(Wavelet Decomposition,WD)technology,which is able to combine with the two hybrid optimization models(PSO-NDN FA-NDN),forming the two combined optimization models(WD-PSO-NDN?WD-FA-NDN).The four hours ahead wind forecasting accuracy of the above two combined optimization models are satisfied.Among them,the predictive curve of WD-FA-NDN is in better agreement with the measured curve.It can effectively capture the trend of wind speed and can improve the local time delay greatly.So WD-FA-NDN is an effective method of ultra-short-time wind speed forecasting.