AUKF ultra-short-term photovoltaic power prediction model based on DBN deep learning residual correction
Aiming at the limitation of linear model and time-varying factors of measurement noise reducing the accuracy of photovoltaic power prediction,a hierarchical photovoltaic power prediction model is put forward.The adaptive algorithm is designed to amend the actual-time noise appraisal of unscented Kalman filter,which reduces the influence of irradiance in the state space model of photovoltaic power prediction system and time-varying noise of photovoltaic power measurement on the prediction accuracy,and achieves the preliminary prediction of photovoltaic power generation.In the two-level prediction,the elemental prediction residual is corrected based on the DBN deep learning network to reduce the influence of nonlinear meteorological factors on the prediction accuracy.The simulation results show that the modified model has better prediction accuracy,better generalization ability and engineering application value.
万方数据
维普
