Short-term Wind Speed Prediction With Multiple Meteorological Variables Based On Morphological Decomposition And Improved Long-short Memory Network

Green development has become an important trend,many countries regard environmental protection as an important pillar of green development,so the development and utilization of renewable energy technology is more and more attention.Wind energy is the energy with the largest reserves,the lowest cost,the most mature utilization technology and the smallest environmental impact in the renewable energy.Therefore,in recent years,countries around the world have been developing wind energy more and more vigorously.However,the wind speed is a non-stationary random process,which leads to the instability of the output of wind power generation,thus affecting the power supply stability and reliability of the power system.Therefore,in order to use wind energy more safely and effectively,it is necessary to forecast wind speed.Accurate wind speed prediction results have very important reference significance for the power system dispatcher to make the dispatching operation plan and the stable and orderly development of power market.This dissertation first introduces the domestic and foreign wind speed prediction methods,and analyzes the advantages and disadvantages of each algorithm,on the basis of the existing algorithm,aiming at its shortcomings to improve.In order to solve the problem of hyperbolic tangent activation function used for network state update in LSTM algorithm,an improved LSTM algorithm is proposed by replacing hyperbolic tangent function with logarithmic activation function which is not saturated with gradient.Experimental results show that,compared with the original algorithm,the convergence speed and prediction accuracy of the improved algorithm are improved.Because wind speed is a non-stationary random process which is easily affected by climate,geomorphology and other factors,the accuracy of wind speed prediction will be reduced if the wind speed is predicted directly without any processing of original wind speed signal.In order to improve the accuracy of wind speed prediction,a time-domain decomposition algorithm based on mathematical morphology is proposed in this dissertation.The original wind speed signal is decomposed into trend baseline signal with relatively stable variation in long-term statistical characteristics and residual perturbation component with strong random fluctuation for research.The experimental results show that the prediction accuracy of the model based on morphology and improved LSTM is higher than that of the model based on improved LSTM,the model based on wavelet decomposition and improved LSTM.In view of the characteristics that wind speed is easily affected by temperature,air pressure and other meteorological factors,in order to further improve the prediction accuracy of the model,this dissertation proposes a multi meteorological variable wind speed prediction model based on morphology and improved long-term and short-term memory network.In this model,temperature,air pressure and other meteorological data are also considered as input variables in model training and wind speed prediction.The experimental results show that the wind speed prediction model with multiple meteorological variables considering temperature and pressure has higher prediction accuracy than the traditional single variable wind speed prediction model.