Research On Short-term Photovoltaic Output Power Prediction Based On BA-LSTM

With the increasing demand for energy in human production and life,and the emphasis on environmental protection,photovoltaic power generation has been developing rapidly as the most promising form of new energy generation.As photovoltaic power generation is subject to strong intermittency and random fluctuations by the external environment,which in turn makes the power supply side of the power system uncertain and presents a challenge to the safe and stable operation of the system.Accurate and effective prediction of the output power of PV plants can therefore ensure safe and stable operation of the power system and enable reliable use of PV energy.The main research contents and achievements are as follows:(1)For the situation of missing data or abnormal data caused by accidental failure and communication interruption of the transmission system of the PV power monitoring platform,a similar day and GRA-BP data repair model is established.Firstly,the power data is tested to identify the days to be repaired with missing or abnormal power,and the grey correlation analysis is used to analyse the similar days,determine the similar day data of the days to be repaired,and establish a BP neural network data repair model to realise the repair of abnormal data.The reliability of the data repair model is verified through the simulation test data,and the interference caused by the abnormal power data to further PV power prediction is avoided.(2)To address the problem that the volatility of PV output power leads to poor prediction accuracy and that neural networks tend to fall into local optima in the prediction process,a short-term PV power output point prediction model based on the Bat Algorithm(BA)optimized Long Short Term Memory Neural Network(LSTM)is proposed.The analysis of the prediction results for different seasons and weather types shows that the BA-LSTM algorithm gives the best prediction results,with root mean square error(RSME)and mean absolute percentage error(MAPE)of 15.45%and 6.42% respectively,which are less than those of the LSTM and radial basis neural network(RBF)algorithms in terms of MAPE and RMSE by 3.65% and 4.6%;10.69%and 6.74% respectively.Therefore the BA-LSTM prediction model has some accuracy in the field of PV power prediction.(3)To address the shortcomings that traditional deterministic PV power point prediction only obtains a single output power prediction value and cannot assess the reliability of the prediction results,interval prediction is applied to short-term PV power prediction,and an interval prediction model based on LSTM is proposed,which estimates the upper and lower bounds of the interval at a given confidence level.To obtain the optimal upper and lower interval limits,an objective function that integrates the prediction interval coverage probability(PICP)and prediction interval normalized average width(PINAW)is constructed,and an improved model for BA to optimize the LSTM network parameters is proposed.The BA-LSTM short-term PV power interval prediction model has 92.78% and 12.42% PICP and PINAW respectively,compared with the single LSTM interval prediction model,which has9.38% higher PICP and 6.31% lower PINAW,indicating that the interval prediction model proposed in this paper has better This indicates that the interval prediction model proposed in this paper has better prediction effect.The interval prediction model proposed in this paper is compared with ANN,SVM and PSO-ELM from the perspective of different seasons and different weather types to verify the superiority of the interval prediction model.