残余物
控制理论(社会学)
光伏系统
不变(物理)
数学
风力发电
计算机科学
功率(物理)
LTI系统理论
应用数学
风速
预测误差
气象学
环境科学
航程(航空)
作者
Yi Wang,Yining Ma,Menggang Kou,Runze Li
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-07-01
卷期号:: 200-204
标识
DOI:10.1109/netps70564.2026.11650500
摘要
Accurate wind and photovoltaic (PV) forecasts are needed for renewable power scheduling, yet deployed forecasts often retain systematic errors across stations, seasons, and lead times. This paper proposes Invariant Causal Residual Correction (ICRC), a post-processing method that corrects existing forecasts rather than replacing operational systems. ICRC treats each station-season pair as an environment, selects features with stable error relationships, combines anchor regression with a gradient-boosting residual model, and uses a validation-based reliability coefficient to limit overcorrection. Experiments on more than one year of 15-minute data from two wind farms and two PV stations in Southwest China cover ultra-short-term and short-term forecasting. Against ridge regression, random forest, and plain histogram gradient boosting correctors, ICRC gives the best reported root mean square error (RMSE), mean absolute error (MAE), and accuracy (ACC). Across 104 station–lead pairs, one shared model improves average ACC from 83.63% to 88.67%, reduces RMSE by 30.8%, and achieves a 92.3% win rate.
科研通智能强力驱动
Strongly Powered by AbleSci AI