基于天气系数随机微分方程的可迁移光伏发电功率预测OA

Aiming at the difficulties of strong dependence on high-resolution meteorological data and high cost of repeated model training in regional PV power plant output prediction,this paper proposes an indirect prediction method of PV power generation based on weather coefficients and stochastic differential equations.Firstly,the real-time PV power is decoupled into the product of the station's intrinsic maximum power under the best weather conditions and the regional general weather coefficients,and the daily curve of the station's intrinsic maximum power is fitted through the time window of the nodal division,which improves the representativeness of the data;secondly,the hourly Jacobi diffusion process is constructed to portray the continuous stochastic fluctuations of the weather coefficients,and the two-way long and short-term memory network optimised by genetic algorithm is used to achieve the hour-by-hour public weather forecast to the stochastic differential equation model parameters of the nonlinear mapping,reduce the dependence on high-resolution meteorological grid data;finally,through the weather coefficient prediction value and the corresponding section of the maximum power generation product to complete the photovoltaic power prediction.The experimental results show that the method maintains high prediction accuracy under multiple weather scenarios,and the RMSE can be reduced by 40.8％compared with other models.It shows good model migration capability between different power stations in the same region,and reduces the time required for model training by 61.5％compared to conventional models,thereby significantly reducing computational cost.