Research On Short-Term Photovoltaic Power Prediction Model Based On Multi-Time Scale Data

In recent years,the global energy crisis and environmental problems have forced the world ’s energy changes.Looking at the global energy change,it is not difficult to find that solar energy with renewable,clean and pollution-free characteristics has gradually become a major protagonist of this change.In order to practice the concept of carbon saving and emission reduction,countries around the world are constantly increasing the proportion of renewable energy power generation.Under the call of global carbon saving and emission reduction,China ’s photovoltaic power generation has been in full swing,but it also brings a series of challenges to the existing power system.The impact of haze caused by air pollution on photovoltaic power generation can not be ignored.At the same time,only considering a single time scale data as input data will affect the accuracy of photovoltaic prediction.In view of the above problems,this thesis studies the short-term photovoltaic power prediction model based on deep learning.The main research contents are as follows:Firstly,aiming at the situation that there are a large number of abnormal data and missing data in the historical power generation data set and the influencing factor data set collected by the photovoltaic power station,the data preprocessing and data normalization are carried out on the collected data set.On the basis of the influencing factors of traditional photovoltaic power generation,the influence of haze concentration on the output power of photovoltaic power generation is considered,and the influence of photovoltaic module temperature on the photoelectric conversion efficiency is considered.Pearson correlation coefficient method is used to analyze the correlation between the influencing factors of photovoltaic power generation output power,and the dimension reduction of the input data of the model is realized.Secondly,aiming at the adverse effects of the intermittency and volatility of photovoltaic power generation on the prediction accuracy,a short-term power generation prediction model based on convolutional neural network-bidirectional long short-term memory neural network is proposed based on deep learning.The combined model has the advantages of the above two neural networks.It can not only fully extract the potential characteristics of the data,but also deeply mine the time series relationship of the data.Compared with other models,the model has better prediction effect.Thirdly,based on the convolutional neural network-bidirectional long short-term memory neural network model,considering that different time scale data contain different feature information,a convolutional neural network-bidirectional long short-term memory neural network prediction model based on multi-time scale data is established.This method divides the input data into multiple time scales,and extracts the deep features implied by the influencing factor data from different time scales.Compared with the convolutional neural network-bidirectional long short-term memory neural network model,the combined prediction model based on multi-time scale data has higher prediction accuracy.Finally,based on the short-term prediction model based on multi-time scale data,the whale optimization algorithm is used for optimization.In view of the fact that different parameters of the combined prediction model based on multi-time scale data have a great influence on the prediction effect,considering that the whale optimization algorithm has the advantage of strong ability to jump out of local optimum,a whale optimization convolutional neural network-bidirectional long short-term memory neural network model based on multi-time scale data is established.Through the comparative analysis of simulation and data,the feasibility of the model optimized by the whale optimization algorithm is proved,and the prediction accuracy is further improved.