Research On Power Prediction Of Photovoltaic Power Generation Based On Neural Network Algorithm

Global traditional fossil energy is gradually depleted,environmental pollution is increasingly serious,countries are urgently looking for new alternative energy.Photovoltaic new energy has the advantages of clean,pollution-free and sustainability,which has been widely concerned and studied all over the world.In recent years,the installed capacity of photovoltaic power generation is constantly increasing,but the problems in the operation of photovoltaic power generation are becoming more and more prominent.The process of photovoltaic power generation is subject to the influence of weather and geographical environment,showing the volatility and random multi-interference,and its output power is easy to change with the change of external factors,so the prediction of output power generation is crucial to optimize the operation of photovoltaic power grid and reduce the impact of uncertainty.This paper is devoted to optimizing the neural network model and exploring ways to improve the prediction accuracy of photovoltaic power generation.The main research contents are as follows:(1)Explore various influencing factors of photovoltaic power generation and build the early theoretical basis for the research of the algorithm prediction model.This paper mainly discussed the principle of photovoltaic power generation technology and the composition characteristics of photovoltaic system,and comprehensively analyzes the influence of various meteorological factors on the output power of photovoltaic power generation.(2)Data preprocessing was carried out on the historical data of output power of photovoltaic power generation.In this paper,the data of State Power Rixin PV power prediction big data competition is used as the original data.The original data were processed with outlier processing,correlation degree analysis of feature variables and other data processing operations,and the new data set was divided.(3)An algorithm model of LSTM-CNN was proposed and established,and the advantages of the model were verified by experiments.Firstly,according to the principles of convolutional neural network(CNN)and LSTM and their advantages in extracting different types of features,a LSTM-CNN algorithm model is established to extract the temporal feature information first.Then photovoltaic testing LSTM-CNN algorithm model,under the different types of weather and season with CNN,LSTM two single space model and the neural network algorithm to extract characteristic information of CNN-LSTMhybrid neural network algorithm model do contrast analysis,including LSTM-CNN algorithm model under the weather and the seasonal types of mean absolute error(MAPE)error index was 0.03793 and 0.03903,respectively,are the lowest.(4)Based on LSTM-CNN hybrid neural network algorithm model,GA-LSTM-CNN hybrid algorithm model was established by introducing genetic algorithm(GA)optimization model theory.At the same time,the experiment is compared with the LSTM-CNN algorithm model to verify that the optimized model has better prediction effect.The experimental test results are compared with the prediction accuracy of LSTM-CNN algorithm model.GA-LSTM-CNN algorithm model has lower error indexes than LSTM-CNN algorithm model in terms of weather type and seasonal type.Among them,the MAPE index value of GA-LSTM-CNN algorithm model in weather type and season type decreased by 0.594% and 0.266% respectively compared with that of LSTM-CNN algorithm model.Among the various neural network algorithm models,the hybrid neural network algorithm model is more accurate than the single neural network algorithm model.The LSTM-CNN algorithm model which firstly extracts temporal feature information has higher precision than the CNN-LSTM algorithm model which firstly extracts spatial feature information.Finally,the GA-LSTM-CNN model optimized by GA algorithm further improves the prediction accuracy.Meanwhile,the prediction effect is generally better under the weather type than under the seasonal type.