Photovoltaic Output Prediction Model Based On WD-PSO-LSTM

Recent years,renewable energy sources such as wind,solar,hydro,and biomass have gradually replaced non-renewable energy sources.Having benefit-ed from the policies of our governments and international institutions,the ad-vantages of photovoltaic technology and the accessibility of solar energy,a large amount of photovoltaic has been grid-connected in the power system.Howev-er,the uncertainty of Photovoltaic(PV)power generation can bring challenges to the security and stability of the grid system.Therefore,accurate PV output prediction is essential for PV power generation to connect to the grid and ensure safe and stable grid operation.Therefore,accurate PV power output prediction is helpful to connect PV power to the grid and ensure the operation of the grid safe and stable.In this thesis,a short-term PV output prediction method based on wavelet decomposition and particle swarm optimization is proposed.At first,the method classifies the PV output data into different weather types of sunny,cloudy,over-cast and rainy PV data according to the historical weather types,and uses wavelets to decompose these PV output data in order to reduce the difficulty of neural network modeling and improve the accuracy of neural network predic-tion,and then decomposes each weather data into four subseries,three layers of wavelets to obtain one low frequency containing trend and three high frequency containing detail information.The four subseries are then modeled and predicted using a long and short-term memory network(LSTM).To solve the problem of setting the parameters of the LSTM,a particle swarm optimization algorithm is used to find the optimal number of neurons,the number of neural network iterations,and the learning rate of the neural network,so as to find a set of parameters to make the network prediction error as low as possible.Finally,by superimposing the predicted values of the four subsequences,the PV output prediction results of the current corresponding weather are obtained.In this thesis,we have modeled and predicted the PV output data under sunny,cloudy,overcast and rainy days,selected ARIMA model,LSTM network and PSO-LSTM model for simulation experiments,and analyzed the RMSE,MAE,R~2and other related evaluation indexes to verify that the proposed method has higher accuracy and stronger generalization ability than the above methods.