LoRA‐Finetuned LLM for Multi‐Step Renewable Energy Power Forecasting

ABSTRACT

With the rapid development of renewable energy, the demand for wind power forecasting across various aspects has become increasingly urgent. However, existing predictive approaches primarily focus on modelling numerical wind power data while neglecting incorporation of auxiliary textual information. To address this issue, this letter investigates whether the large language model (LLM) can be adapted to wind power forecasting using Low‐Rank Adaptation (LoRA). Task instructions and historical statistical descriptors are combined with numerical wind‐power patch embeddings and LoRA modules are applied to the query, key and value projections of Llama‐3‐8B‐Instruct. Experiments on real‐world datasets demonstrate the effectiveness and flexibility of the proposed approach. This research not only shows the adaptability of LLM to wind power forecasting but also contributes to paving the way for LLM utilization in the evolving landscape of renewable energy forecasting paradigms.