The accelerated transition to renewable energy requires reliable forecasting and optimisation to support stable, efficient grid operations. In recent years, foundation models, including large language models (LLMs) and time-series foundation models, have been investigated for renewable energy forecasting tasks that integrate diverse inputs, such as sensor telemetry, maintenance logs, weather narratives, and geospatial descriptors. This paper presents a structured review of recent studies published from 2020 to 2025 on the application of language-focused foundation models and related transformer-based foundations to wind, solar, and load demand forecasting. Instead of asserting universal performance advantages, we synthesise evidence showing where studies report improvements compared with conventional machine learning (ML) baselines, most often in contexts involving multimodal fusion, longer forecasting horizons, or text-enriched decision support. We also highlight situations in which the reported benefits are limited or strongly influenced by the dataset and evaluation protocol. In addition, we examined practical constraints affecting deployment, including integration with Supervisory Control and Data Acquisition (SCADA) and Internet of Things (IoT) systems, latency, reliability, and governance. We summarise available interpretability approaches, including Shapley Additive Explanations (SHAP), attention visualisation, and post hoc explanations, and identify key open challenges such as computational cost, hallucination risk, and uncertainty quantification in safety-critical forecasting.
Aleassa et al. (Tue,) studied this question.