Robust Resilience Optimisation of Rural Integrated Energy Systems Based on Data‐Mechanism Hybrid‐Driven Prediction

ABSTRACT

High‐impact low‐probability extreme weather events, such as cold waves and snowstorms, may interrupt upstream electricity and natural‐gas supplies to rural integrated energy systems (RIES). Resilient dispatch is further complicated by the limited coverage of photovoltaic (PV) power samples under adverse operating conditions and agriculture‐specific energy requirements. This paper proposes a Hybrid Forecast‐Informed Robust Resilience Optimisation (HF‐RRO) framework. A convolutional neural network–long short‐term memory network (CNN–LSTM) model predicts global, direct, and diffuse irradiance, which are converted into PV power through a physical mechanism chain, while prediction errors are used to calibrate the PV uncertainty set. Greenhouse thermodynamics and biogas‐digester kinetics are incorporated to characterise agricultural production requirements. HF‐RRO coordinates robust equipment‐state decisions with economic continuous dispatch and load shedding. Under adverse operating conditions with limited representative PV power samples, the proposed predictor reduces MAE and RMSE by 56.6% and 43.7%, respectively, relative to direct CNN–LSTM prediction, and by 21.2% and 7.0% relative to a Transformer predictor. Under overlapping upstream electricity and natural‐gas interruptions, greenhouse environmental variables remain within allowable ranges, and the electricity and heat recovery indices reach (0.97, 1.00). Compared with RO, DRO and MRO, total operating costs decrease by 22.2%, 6.7% and 11.5%, respectively.