Distributed Photovoltaic Power Prediction Based On Spatio-Temporal Multi-Dimensional Point By Point Information Elements

In recent years,with economic growth and increasing energy demand,it is of great strategic significance to continuously reduce coal power generation,vigorously develop and use photovoltaic,hydropower,and other new energy,realize clean energy instead of thermal power generation,develop a low-carbon economy and restructure the energy system.Distributed photovoltaic power generation,because of its wide distribution,flexible installation and other advantages,has gradually become the most potential new energy.However,due to the great influence of light conditions,there are changes in circadian rhythm and seasonal rhythm.At the same time,it is also affected by the weather type,so distributed photovoltaic power generation has the characteristics of volatility,randomness,and intermittency.To realize the safe and stable operation of the power system and keep the proper balance between supply and demand,the forecast of the output power of distributed photovoltaic generation is a promising solution.Based on the analysis of spatio-temporal multidimensional information elements,this paper makes theoretical research and technological breakthroughs in the improved algorithm of distributed photovoltaic power prediction.The main work is summarized as follows:Firstly,this paper analyzes the multidimensional spatio-temporal influence factors of distributed photovoltaic power,including the following two aspects: 1.Analyze the time and space influence factors;2.Study weather classification methods.By analyzing the spatio-temporal influencing factors,the main influencing factors of photovoltaic power generation can be screened more reasonably,which will prepare for the weather classification research;Through the study of weather classification,the temporal and spatial influencing factors are associated with weather types,which provides the basis for the subsequent prediction research.Secondly,to describe the key influencing factor of cloud occlusion,this paper uses a satellite cloud image to extract the cloud occlusion factor and proposes a photovoltaic power prediction method based on a gray satellite cloud image and optimized LSTM.Firstly,the satellite cloud image over the centralized photovoltaic power station is obtained,and the image processing(homomorphic filtering,standardization,pixel value correction)is used to complete the pre-processing of the satellite cloud image;Then,based on the processed cloud image,the gray co-occurrence matrix was used to extract cloud occlusion elements in different scanning ways,and the best cloud occlusion elements were selected according to the correlation;Finally,the optimal cloud shielding elements are input into the optimized LSTM model to complete the centralized photovoltaic power prediction.Finally,considering the above multi-dimensional influencing factors and centralized photovoltaic power prediction results,an improved KMC-Vine Copula model is proposed to complete the point-by-point probability prediction of distributed photovoltaic power.Based on the Copula correlation theory,the temporal and spatial correlation of centralized PV and distributed PV is cascaded.Because the traditional single Copula cannot reflect complex correlation,a two-dimensional Optimized Copula model is proposed to improve this shortcoming.However,since it cannot reflect the high dimensional correlation,this paper builds a more flexible multidimensional Vine Copula model as the final prediction model.Through calculation examples,compared with the two-dimensional model,the model proposed in this paper improves by 26.87%(interval width),41.6%(reliability),and 39.23%(sharpness)respectively under three evaluation indexes by integrating different weather types(cloudy,sunny and overcast).