| With the increasing proportion of photovoltaic power generation in the country,the accuracy of photovoltaic power generation forecast will directly affect the strategy and feasibility of photovoltaic power grid connection.However,due to the periodicity and intermittency of photovoltaic power generation,the conventional prediction method is often inconsistent with the actual results,and the photovoltaic cells have the disadvantages of being greatly affected by weather factors and low conversion efficiency.If the photovoltaic power is generated into the power grid without treatment,it is easy to cause a strong impact on the power grid,affecting the stable operation of the power grid.Therefore,accurate prediction of photovoltaic power generation is an important premise to reduce the adverse impact on the power grid,which can improve the predictability of photovoltaic power generation.If the power generation cannot be predicted more accurately,it will not only cause the inefficiency of photovoltaic power generation,but also bring difficulties to the grid-connected operation and affect the feasibility of photovoltaic power grid connection.Based on the comparison between solar energy and traditional energy,this thesis discusses the use of artificial intelligence algorithm to improve the reliability of photovoltaic energy,and compares the neural network with the traditional photovoltaic power generation prediction method.This thesis delves into how these algorithms work and how they have been applied in recent years,exploring the use of artificial intelligence algorithms to improve LSTM neural networks and to construct more accurate prediction models by using the Escape bird algorithm(EBS).In order to deal with the time change of photovoltaic power generation and the uncertainty it brings,we take a series of measures to optimize the LSTM algorithm,set parameters according to the previous references,and construct the escape bird algorithm(EBS)to optimize the Attention mechanism(LSTM)combination algorithm model.The original data of DKA yurala photovoltaic power station in Australia are used,and the data are selected,filled and normalized,so as to effectively reduce the deviation caused by the interference of missing or extreme values.At the same time,Pearson(Pearson correlation coefficient method)is used to analyze the characteristics of factors affecting the photovoltaic power generation.Six factors with strong correlation were selected as the input of the model to improve the accuracy of the model prediction.In this thesis,EBS-Attention-LSTM combined algorithm model is established to predict photovoltaic power generation.Meanwhile,traditional LSTM,EBS-LSTM and Attention-LSTM neural networks are respectively used to carry out simulation verification and comparison experiments.At the same time,EBS-BP algorithm model is established,and the traditional BP(error back propagation)neural network and GA-BP(genetic algorithm-error back propagation)algorithm are respectively used to verify and compare the simulation experiments.The simulation results show that the RMSE of EBS-BP neural network is 4.3and 1.3 lower than that of single BP neural network and GA-BP neural network,which is about 52% and 26% lower than that of EBS-BP neural network.It can be seen that EBS algorithm shows good prediction effect when optimizing BP neural network and effectively improves the prediction accuracy.Compared with GA algorithm,EBS algorithm also shows superior optimization effect and more accurate prediction effect.Compared with EBS-LSTM,Attention-LSTM,the RMSE of EBS-LSTM decreased by 3.2,1.2 and 7.4,respectively,and the RMSE of EBs-LSTM decreased by 55%,32% and 74%.It can be concluded that EBS-Attention-LSTM has higher accuracy and better prediction effect than the above three models in this prediction,which proves the effectiveness of the combination of its algorithm models.According to the analysis results,the research direction of improving the two algorithm models is proposed. |
PDF Full Text Request