Research On Short-term Photocoltaic Power Forecasting Based On Artificial Intelligence

Increasing environmental contamination and the progressive depletion of energy sources have aroused significant interest in solar energy.As the popularity of photovoltaics continues to rise,power forecasting of photovoltaic power generation is becoming increasingly crucial for ensuring grid stability,attaining optimal allocation,and cost-effective unit dispatch.Due to the frequent fluctuation characteristics of photovoltaic power generation,traditional single forecasting models are incapable of providing precise forecasts of photovoltaic power generation.To this goal,this thesis proposes a survey based on artificial intelligence to forecast hybrid photovoltaic power generation.The main research aspects are as follows:In order to solve the shortcoming of the BP neural network in the forecasting of photovoltaic power generation which is prone to local extrema,this study adopts the particle swarm optimization algorithm to optimize the BP neural network parameters.In terms of algorithmic network construction,the study used grey relation analysis to select similar day data,and the dimension reduction of the influencing factors is carried out by principal component analysis.In addition,the hybrid photovoltaic power forecasting model based on similar day theory and PCA-PSO-BP is also proposed.This model is used for the simulation verification of photovoltaic power generation under different weather conditions and compared to existing forecasting models.The results demonstrate that the PCA-PSO-BP based photovoltaic power generation forecasting model has a significant predictive precision dominance across a variety of weather circumstances.This research offers an adaptive data-based hybrid model based on LSTM to solve the problem that the intermittency and fluctuation of photovoltaic power generation have a negative impact on the precision of prediction.Meanwhile,the study introduces CEEMD and PE to decompose and reconstruct the original data,therefore lowering the non-stationarity of the time series.The Kmeans algorithm is utilized to identify training samples of a similar day of the same kind as the prediction day,and the LSTM is employed as the prediction algorithm to propose a photovoltaic power combination forecasting model based on Kmeans and CEEMD-PE-LSTM.After simulation verification with other forecasting models,the model’s mean absolute error percentage and root-mean-square deviation are lower than those of the LSSVM forecasting model and the LSTM forecasting model under a variety of weather conditions.Consequently,it can be demonstrated that the proposed model is a more effective model for forecasting photovoltaic power generation.In conclusion,the photovoltaic power hybrid forecasting model based on artificial intelligence developed in this thesis successfully enhances the prediction accuracy of a single forecasting model,and aids in optimizing the operation of the energy sector and enhancing grid operation stability.