Research On Short-term Photovoltaic Power Generation Power Interval Prediction Method Based On Enhanced Recurrent Neural Network

With the vigorous development of the photovoltaic power generation industry,countries around the world are paying more and more attention to the photovoltaic power generation industry,and the total amount of newly installed photovoltaic capacity in the world is increasing rapidly.With the strong support of national policies for the photovoltaic power generation industry,the development speed of my country’s photovoltaic power generation industry has been in a leading position in the world in recent years,and the cumulative installed capacity of domestic photovoltaic power generation is far ahead of other countries.By the end of 2020,it has ranked first in the world for six consecutive years.However,due to the intermittent and fluctuating characteristics of solar energy,the photovoltaic output power also exhibits the same characteristics.When the photovoltaic power generation system is directly connected to the power grid,it will bring serious challenges to the safety,stable operation and power quality of the power grid.High-precision prediction of the output power of photovoltaic power generation systems is an effective method to solve this problem.In this paper,according to the power generation characteristics of photovoltaic power plants,the main meteorological factors affecting photovoltaic output power are analyzed for reasons,and the pearson correlation coefficient method is used to study the correlation between photovoltaic output power and various meteorological factors.Finally,according to the correlation analysis results,the meteorological factors that can be used as model input variables are determined to prepare the data for the subsequent modeling.Secondly,based on the Variational Mode Decomposition（VMD）algorithm,the original sequence value of photovoltaic output power is decomposed to obtain sub-modal sequences such as trend component,detail component and random component,and the decomposed sub-sequence is used as the prediction model.The input can effectively weaken the non-stationarity of the original series,thereby improving the robustness and accuracy of the prediction model.Then,in view of the problem that the length of the semantic vector C does not change in the traditional Encoder-Decoder codec used in the long short-term memory network（LSTM）during the model training process,a fusion attention mechanism is proposed to optimize the LSTM encoding-decoding process;combined with VMD decomposition technology,a short-term photovoltaic output power combined single-point prediction model based on VMD and ALSTM is proposed.The experimental results show that the VMD decomposition technology can reduce the non-stationarity of photovoltaic output power data and improve the robustness of the model,and the Attention mechanism also improves the prediction accuracy of the model because it improves the encoding and decoding method of the LSTM network.Therefore,the forecasting ability of the proposed short-term combined forecasting model is effectively improved in many aspects.Finally,due to the intermittent and fluctuating characteristics of photovoltaic output power,the current mainstream deterministic point prediction of photovoltaic output power is actually difficult to fully meet the needs of power system dispatching operation,while interval prediction can not only provide deterministic At the same time,it can also give the fluctuation range of the output power at the moment when the confidence level is satisfied,so as to provide decision makers with more comprehensive prediction information.Therefore,this paper proposes a combined prediction method of photovoltaic output power interval combined with nonparametric kernel density estimation theory（NKDE）.The numerical example analysis shows that the method has good feasibility and potential value.