Ultra Short-term Forecast Of Photovoltaic Generation Based On GRU Of Deep Learning Algorithm

In the contradiction between the continuous growth of energy demand and the deterioration of ecological environment,it is an inevitable trend to seek efficient,clean and renewable energy to change the existing energy system and promote the sustainable development of energy.China has made it clear that carbon dioxide emissions will peak by2030 and strive to achieve carbon neutrality by 2060.Therefore,to achieve these two goals at the same time,the development and utilization of new energy is one of the most effective energy conservation and emission reduction measures.Solar energy has the advantages of large resources and wide distribution.PV generation technology is an effective way to develop solar energy to achieve emission reduction.However,PV generation does not have the continuous adjustable controllable characteristics of traditional power generation.Therefore,while using solar energy,the fluctuation and uncertainty of solar energy also bring great challenges to the real-time scheduling and safe operation of power grid.In order to make full use of solar energy resources,ensure the real-time balance of power generation,transmission and consumption when PV generation is connected to the grid,and accurately predict the PV generation is of great significance for the operation and dispatching of power system.This paper presents an ultra short-term PV generation prediction method based on deep learning network of gated recurrent unit.Firstly,the influence of meteorological factors on PV generation is analyzed,and the operation data and numerical weather prediction data of photovoltaic plant are preprocessed,including abnormal value processing,missing value supplement and normalization processing.Then,Spearman correlation coefficient method is used to analyze the preprocessed data to find out the main factors affecting the PV generation,which are used as the input variables of the prediction model.Then,setting the prediction model parameters,and the training ability of the network model is optimized through the experimental method.Finally,by comparing the prediction results of GRU with RNN,LSTM,BP and ANN,it is verified that the deep learning network is more suitable for dealing with the nonlinear time series problem of PV generation,and the advantages and high prediction accuracy of GRU network are also verified.Considering that the PV generation is affected by different meteorological factors at different times,a single model can not consider the impact of meteorological factors at different times on the PV generation.Based on GRU network algorithm,this paper proposes a ultra short-term power prediction method for time-division gated recurrent unit network.Firstly,the correlation coefficient is used to analyze the correlation variables at each time,and the strong correlation time data is determined as the input of the model.Then,according to different input data and different data dimensions,the neural network parameters of the network model are determined.A GRU model is established for each time to predict the PV generation at that time,so as to realize the prediction of PV generation.Finally,compared with the single GRU model,the proposed time-division gated recurrent unit model has higher prediction accuracy than the single GRU model.