Research On Short-term Prediction Method Of Photovoltaic Power Based On Learning Mechanism

Achieving "carbon peaking and carbon neutrality" has become an important strategic goal of China.Increasing the proportion of renewable energy power generation represented by photovoltaic power generation is an important way to promote carbon emission reduction.The rapid development of photovoltaic power generation in recent years has brought both opportunities and challenges to the power system.Due to the volatility and intermittency of photovoltaic power generation,improving the accuracy of photovoltaic power forecasting is critical to ensuring the stability of power system operation and improving electrical energy.Quality matters.Most of the existing photovoltaic power prediction methods are limited to using a single prediction model,resulting in limited generalization ability of the model,or using a simple arithmetic average to combine the prediction models,lacking sufficient theoretical support.In view of the above problems,in order to further improve the accuracy of photovoltaic power prediction,this paper studies the short-term prediction of photovoltaic power based on the learning mechanism.The main research contents are as follows:(1)Aiming at the volatility and randomness problems in photovoltaic power prediction,a prediction model based on adaptive noise complete ensemble empirical mode decomposition and whale optimization algorithm is proposed to improve the parameters of bidirectional longterm memory network.The model uses the bidirectional long-term memory network to extract the feature information of the forward and backward data at the same time,and realizes the linkage of the forward and backward data.In addition,in view of the large fluctuation and randomness of photovoltaic power,the model adopts adaptive noise complete set empirical modal decomposition to decompose the photovoltaic power sequence into sub-modal sequences with different frequencies,and uses the whale optimization algorithm to improve the bidirectional long and short time.The parameters of the memory network are used to establish a short-term prediction model of photovoltaic power.The simulation results show that the proposed model has strong generalization ability and high fitting degree.(2)Aiming at the problem that the cyclic structure of bidirectional long-short-term memory network cannot be trained in parallel,a photovoltaic power prediction model based on empirical wavelet transform and whale optimization algorithm is proposed to improve random forest parameters.The model uses the empirical wavelet transform to decompose the photovoltaic power sequence into wavelet components with characteristic differences,uses the whale optimization algorithm to improve the parameter selection of the random forest algorithm,and uses the random forest for prediction.Finally,compared with other photovoltaic power prediction models,it is verified that the proposed photovoltaic power prediction model has higher prediction accuracy.(3)A single model can no longer meet the requirements of prediction accuracy.In order to further improve the prediction accuracy and robustness of the model,a combined photovoltaic power prediction model based on a multi-objective locust optimization algorithm is proposed.Use empirical wavelet transform to decompose the data,establish three prediction models of bidirectional long and short-term memory network,random forest and least squares support vector machine respectively,use the whale optimization algorithm to improve the parameter selection of the three models,and finally use the multi-objective locust optimization algorithm Optimize the weights of a single model.Compared with the single prediction model,the combined model has higher stability and prediction accuracy.