Photovoltaic Power Prediction Based On Ground-based Cloud Classification And Stacking Model

Photovoltaic power generation is affected by the irregular motion of clouds and its fluctuation characteristics are closely related to the weather type,which affects the prediction accuracy of short-term photovoltaic power generation prediction.Therefore,this paper proposes a research method of photovoltaic power generation prediction based on ground-based cloud map classification and Stacking model by monitoring the motion patterns of clouds under different weather types.First,based on Numerical Weather Prediction(NWP),we construct weather fractal factors and weather fractal vectors,classify weather types into sunny weather,cloudy weather,cloudy weather,rain showers and all-rainy weather,and classify PV power generation under various weather types into clear-sky smoothing process data and fluctuating process data by variational modal decomposition.process data.Next,the ground-based cloud map is corrected for distortion,the gray value of each pixel point in the cloud map is extracted,and the gray threshold of each cloud map is calculated to determine the cloud type.The characteristic movement of clouds in the ground-based cloud map is monitored by the accelerated robust features,and the speed and direction of the clouds’ movement are calculated.Then,the speed and direction of the cloud mass movement are distinguished under different cloud map types,and the classification of the ground-based cloud map is realized.Then,the fluctuation process data of PV power generation presents three fluctuation characteristic parameters of moving average,standard deviation and peak ratio to classify the fluctuation data under different ground-based cloud maps.According to the PV power generation and historical cloud map data of a PV power plant in Jiangsu from January 1,2017 to December 31,2017,a Convolutional Neural Network(CNN)combined with Long Short-Term Memory(LSTM)model is constructed for the Various scenarios are simulated to analyze the advantages of weather fractal and ground-based cloud map classification in PV power prediction.Finally,to improve the prediction accuracy of PV power generation,a prediction model with multi-model fusion Stacking is proposed by combining the adaptation conditions of machine learning algorithms.The CNN-LSTM model and the XGBoost(e Xtreme Gradient Boosting)model are used as the base model of the Stacking model,and the Extreme Learning Machine(ELM)is selected as the meta-model of the Stacking model,and the 3-cross-training method is used for various scenarios The training method is used to predict the PV power under various scenarios.By comparing the prediction errors of CNN-LSTM model,XGBoost model,and Stacking model,the results show that the Stacking model significantly improves the prediction accuracy of PV power generation.