Research On Prediction Model Of Photovoltaic Generation Based On Deep Learning

At present,solar energy resources are widely exploited and utilized worldwide,but the actual utilization is far from meeting human energy needs.With the increase in the use of renewable energy,forecasting technology is particularly important for the promotion of photovoltaic power generation.However,the efficiency of photovoltaic power generation depends on the weather conditions,and it is easy to generate large fluctuations under different weather conditions,which is characterized by randomness,volatility and intermittentness.Due to the instability of photovoltaic power generation,it is more difficult to make practical planning based on regional solar energy utilization and development.Therefore,accurate power generation prediction is critical to ensure grid stability and economic dispatch.In order to improve the prediction accuracy of photovoltaic power generation and reduce the impact of photovoltaic power generation instability,this paper proposes an end-to-end photovoltaic power generation prediction model based on deep learning.The main work includes:(1)Photovoltaic data preprocessing methodData preprocessing based on acquired real-time power generation raw data of three small distributed photovoltaic power plants in a distributed photovoltaic power plant,including LOF outliers,numerical normalization,and missing values based on the Markov Monte Carlo method The padding is divided according to the original data by 5,15,30 min time precision.(2)A sequence prediction model based on deep learning is proposed.According to different model comparison studies,this paper proposes a photovoltaic power generation prediction model using weather forecast meteorological data.This study turns the prediction problem into a structured sequence output prediction problem while predicting multiple outputs.The model uses Bi-LSTM(bidirectional long-term and short-term memory)as the basic unit,and considers the dependence between consecutive times of the day according to the Bi-LSTM network structure.The Seq2Seq(sequence-to-sequence)model was constructed using the powerful learning capabilities of the deep learning model and the Bi-LSTM features to process photovoltaic power generation data.(3)Comparative analysis model prediction resultsThis paper uses deep learning to process the deep neural network structure of timeseries prediction,namely the recurrent neural network(RNN),by comparing the traditional prediction model least squares support vector machine(LSSVM)model and gradient boosting decison tree(GBDT)in different generator sets and different time precision.Thereby verifying the validity of the model.In this paper,the meteorological data at a fixed time is selected as the input data of the model.The prediction results are compared under the time precision of 5min,15 min and 30 min,and the advantages of the model in the prediction of photovoltaic power generation under different time precision are analyzed.The results show that the Seq2 Seq model based on Bi-LSTM can effectively improve the prediction accuracy of photovoltaic power generation.