Accurate electricity load forecasting underpins decisions on energy procurement, generation scheduling, and supply‑demand balancing, directly affecting the economic efficiency and reliability of power systems. In practice, future demand is shaped by a mixture of web notices, weather alerts, and event schedules; these events interact over time, sometimes reinforcing, sometimes offsetting each other, which hampers conventional models. To tackle this, we introduce ElecLoad, a context‑aware agent designed for multi‑event load forecasting. During training, ElecLoad ingests historical load together with event context, learning each event’s relevance to the target load, the timing of its impact, and potential interactions among events. Guided by this analysis, ElecLoad adaptively decides whether to forecast the whole horizon directly or to partition it into segments where event effects change. When partitioned, sub‑agents independently produce initial forecasts for each segment; the main agent then reviews all sub‑forecasts and provides selective refinement feedback to the sub‑agents, forming a closed‑loop improvement process. Experiments on several public load datasets show that ElecLoad outperforms existing baselines in most scenarios, achieving state‑of‑the‑art performance. The implementation is open‑source at https://github.com/zhjai/ElecLoad.
Review: ElecLoad’s explicit modeling of event effects and their interactions enables a self‑adaptive forecasting workflow, offering a promising direction for load prediction under complex contextual influences.