Short Term Photovoltaic Power Prediction Model Based On CEEMDAN-ARO-LSTM Model

Energy is the material foundation for human survival and development,occupying an important strategic position in the national economy.With the improvement of modernization,fossil fuels are gradually unable to meet human needs.Solar energy,as a pollution-free,noiseless,and easily accessible new energy source,is widely exploited and utilized.The most common use of solar energy is for power generation.However,due to the instability of natural weather,photovoltaic power generation has characteristics such as volatility and instability.If photovoltaic power generation with unstable output is directly integrated into the power grid system,it will inevitably cause a significant impact on the power grid system and disrupt its stability.Predicting the power of photovoltaic power generation based on past historical data can effectively reduce the impact on the power grid system during the grid connection process.Based on the characteristics of photovoltaic power generation,this article adopts a combination of theoretical analysis and empirical testing to conduct research:(1)Using CEEMDAN(Complete Ensemble Empirical Model Decomposition With Adaptive Noise)decomposition to denoise the original data sequence,the original data sequence is decomposed into several IMF components.Compared with the original signal,the decomposed IMF components have less noise,And there is a global stop standard at each stage of decomposition,greatly improving the efficiency of decomposition.Based on the size of sample entropy,the IMF component was reconstructed to obtain three components related to the original data sequence,namely high-frequency component,low-frequency component,and trend term component.By using CEEMDAN decomposition,non-stationary raw data was smoothed while retaining most of the original signals;(2)Aiming at the problem that LSTM(Long Short Term Memory)neural network parameters are difficult to select and debug,the mean square error(RMSE)is used as the fitness function of the ARO(Artificial Rabbits Optimization)optimization algorithm,and the three hyperparameter of the LSTM model,namely,the number of neurons,the minimum number of batch processes,and the number of training times,are used as the optimization objectives of the ARO algorithm.The root mean square error(RMSE)Average absolute error(MAE)and Goodness of fit(R2)are used as evaluation indicators of the model to find the optimal hyperparameter of LSTM model;(3)Combining the noise reduction ability of CEEMDAN decomposition,the long time series prediction ability of LSTM model,and the optimization ability of ARO algorithm,a CEEMDAN-ARO-LSTM model is constructed.Firstly,denoise the original data through CEEMDAN decomposition;Then,the ARO algorithm is used to optimize the hyperparameter of the three component LSTM model;Finally,the optimal LSTM model is used to predict each component,and the predicted results are aggregated to obtain the final predicted result.To verify the effectiveness of the proposed model,simulation experiments were conducted using 2020 power generation data from a photovoltaic power plant in Ningxia.The CEEMDAN-ARO-LSTM model proposed in this paper was used for prediction.By comparing and analyzing the prediction results with LSTM model,CEEMDAN-LSTM model,and AROLSTM model,it was concluded that the prediction accuracy of the CEEMDAN-ARO-LSTM model is higher than that of the other three models.Overall,the CEEMDAN-ARO-LSTM model integrates data preprocessing,parameter optimization,and time series data prediction,providing certain reference value for reducing the potential impact of photovoltaic power generation on the power grid system during grid connection.