Short-term Photovoltaic Power Forecasting Based On Similar Period Selection Method And LSTM Neural Network

In today's increasingly severe environmental crisis and energy shortages,the development of renewable energy sources such as solar and wind energy has received widespread attention worldwide.In recent years,with the vigorous support of national policies,China's photovoltaic power generation industry has developed rapidly,and the scale and generation capacity of photovoltaic power stations in the power grid system have also increased.However,photovoltaic power generation is random and intermittent,and its large-scale grid connection has an impact on the stable operation of the power grid.Forecasting the power of photovoltaic power generation can help the power planning department to coordinate the deployment,improve the utilization rate of photovoltaic power generation,help reduce the negative impact of photovoltaic grid connection on the stable operation of the power grid,and improve the stability of the power system operation.Although many forecast methods in the existing research have achieved good results,there are still defects such as lack of timing,complicated data required for model input,and large similar data reference spans,resulting in insufficient forecast effects in actual engineering applications.Aiming at the above problems,this article mainly conducts research work from the following three aspects:1)Analysis of influencing factors of photovoltaic power generation forecasting.First of all,the working principle of photovoltaic panels,the structure of photovoltaic power generation system and the equivalent circuit of photovoltaic power generation are studied to consolidate the theoretical basis for experimental research;secondly,the historical data of a photovoltaic power plant in Guangxi is used to control the power generation power through the control variable method.Various meteorological factors are analyzed one by one;finally,principal component analysis is used to reduce the dimensionality of the multivariate data series to reduce the input data dimension and improve the operational efficiency of the forecast model.2)The similar day selection method and the similar period selection method sample selection research.The gray related degree similarity day selection method is researched,and its principles and methods are analyzed;the screening results of the similar day method are analyzed through the measured data of the power station to find out the advantages and disadvantages of this method;improve and optimize the similar day selection method,propose the similar period selection method;through experiments,the optimized similar period selection method has a stronger adaptability to the changing weather conditions of special weather types.3)Modeling and verification of photovoltaic power forecasting based on long short-term memory(LSTM)neural network.Pre-process the input multivariate data sequence,reduce the interference information in the initial sample,and solve the conversion problem between different dimensions;determine the number of input,output,and hidden layer nodes of the model through principal component analysis;qualitatively analyzed the influence degree of each parameter of the model on forecasting,and selected the most ideal parameter combination with comprehensive performance through multiple experiments;in addition,this thesis uses the classic extreme learning machine(ELM)and error back propagation(BP)neural network forecast models as comparative experiments to verify the effectiveness of the photovoltaic power forecast model by similar period selection method and LSTM-based neural network.The experimental results show that both LSTM and ELM models show strong predictive ability,and the fitting accuracy of the LSTM forecast model is better;the similar period selection method improves the accuracy of similar reference by combining similar time periods,and has better stability when forecasting changing weather conditions;the LSTM forecast model combined with the similar period selection method has strong robustness to the forecast of time series data such as photovoltaic power generation,and its forecast accuracy is more reasonable under complex weather conditions,which can be used for power The dispatch department adjusts the power generation plan to provide a more accurate reference basis.