Research On LSTM Neural Network Photovoltaic Power Generation Prediction Based On Air Quality

With the massive exploitation of primary energy worldwide,the problems of environmental pollution and energy depletion have become increasingly prominent,and countries have begun to adjust their energy consumption structure,gradually shifting their development focus to clean energy.Solar energy is abundant and convenient to use,but its power generation is affected by light intensity and meteorological environment.It has strong volatility and randomness.It is particularly important to accurately predict photovoltaic power generation.In recent years,the scale of my country ’s industrial enterprises has expanded and environmental pollution has been serious.The frequent occurrence of haze weather has led to a decline in the amount of radiation entering the ground and the temperature of photovoltaic panels,which seriously affects the accuracy of power prediction.Therefore,this paper proposes a short-term prediction method for photovoltaic power generation under the haze weather type.First of all,for the problem that the traditional BP neural network is easy to fall into the defect of local optimal solution,RNN and LSTM neural networks with time series memory function are selected as prediction models.By analyzing the influence of air quality parameters on photovoltaic power generation,PM2.5,PM10 and AQI were determined as the main indicators;combined with temperature and radiation intensity,the power generation power factor was formed for training.This model can effectively predict the power generation of photovoltaic power plants in heavily polluted areas and seasons,which is conducive to the construction planning and dispatching planning of such regions.Secondly,due to the problem of insufficient consideration or weak correlation in the selection of traditional similar days,an LSTM photovoltaic power generation prediction model based on comprehensive similar days is proposed.When selecting similar weather days,both weather type and average temperature are taken into account,mean PM and average AQI parameters are considered,and gray correlation degree analysis method is used to determine similar weather days;missing value interpolation method is used to complete the power generation data and Pearson is used Correlation coefficients are used to find similar days of power generation,and the two constitute a comprehensive similar day to filter out prediction sample data,which improves prediction accuracy and training speed.Furthermore,in order to solve the problems of bad data and missing data in historical data,an LSTM photovoltaic power generation prediction model based on similar power stations and comprehensive similar days is proposed.Select similar power stations as the repair method of abnormal data,introduce the photovoltaic panel potential induced attenuation rate,total installed capacity of each power station and monthly total generated power for fuzzy clustering to select similar power stations;This paper introduces the weighted average distance to the traditional FCM clustering to form the FCM-σ clustering algorithm,which reduces the clustering error.Finally,taking the actual historical power generation data of a photovoltaic power station group in a region in northwest China as an example,different weather types are selected as prediction days,and RNN,LSTM,similar day-LSTM,similar day and similar power station-LSTM models are established one by one to compare the prediction results Yes,the conclusion shows that the photovoltaic generation power prediction model based on similar day and similar power station-LSTM has the characteristics of fast operation speed,high fitting degree and strong practicability.