Photovoltaic Power Generation Power Prediction Based On Improved Particle Swarm Optimization Neural Network

This thesis firstly collates and analyzes all kinds of data about this subject at home and abroad,and deeply explores the field of ultra short term power prediction of photovoltaic power generation.By combining with the original data of the power plant,the time similar to a special forecast day is selected,the power data and meteorological data of similar days are set as the training set of this model,and then the data is normalized,and different optimization algorithms are used to improve and perfect the model,and the simulation application is realized after completion.The following is the research content of this paper:The research focuses on the factors that interfere with photovoltaic power generation,such as temperature,solar radiation,wind speed and humidity.By normalizing all the original data interfering with photovoltaic power generation,the hierarchical clustering method was used to obtain the K value in k-means algorithm,and the temperature and solar radiation were set as input variables in this algorithm.Due to the high linear correlation coefficient between solar radiation,temperature and photovoltaic power,the prediction efficiency could be significantly improved.Then,all the original data were analyzed by clustering,and the best similar day was selected for cloudy,sunny and rainy days,and the similar day was regarded as the training sample of this model.One common problem in traditional BP neural networks is that it is easy to select the local best.For this problem,I present the BP mental network model optimized based on particle swarm optimization algorithm,namely PS0-BP model.Compared with traditional BP neural networks,the simulation results of this model are more accurate.However,when the weather fluctuates greatly,the prediction accuracy of PS0-BP model is low.In this paper,the ultra-short-term prediction model of photovoltaic power generation built based on random forest algorithm can effectively improve the prediction accuracy.On the basis of the existing PS0-BP model,the correlation matrix between the random forest prediction model and the single model can be obtained by using the grey correlation degree,and the correlation degree matrix is regarded as the objective of training artificial neural network,and the most suitable weight matrix for the single model is obtained by calculation.Based on this,the RF-PSO-BP combined prediction model is built.This model can significantly improve the accuracy of prediction,but the accuracy still falls short of expectations.This paper uses Elman neural network to improve,the network is based on BP neural network,and is more ideal in stability and dynamics,through reasonable adjustment network to establish the Elman prediction model,using Matlab software through simulation can fully verify that the Elman neural network after the improvement and optimization of the application effect is more ideal.Aiming at the improvement of PSO,a new adaptive particle swarm optimization algorithm(IPSO)is proposed.Then,the improved particle swarm optimization algorithm is used to optimize the threshold and initial weight of Elman network,re-establish the grey correlation weight coefficient,and further establish the RF-IPSO-Elman model.Three different weather conditions are selected as the input data of the model.The output result is a power data point.Then Matlab software was used for simulation processing,and three forecast days with different seasons and weather types were selected and input into RF-IPSO-Elman model for power prediction.It has been proved that RF-IPSO-Elman model,as a new way,can effectively improve the accuracy of short-term photovoltaic power prediction,so that photovoltaic power generation can be controllable,measurable and adjustable.