Solar Forecasting with Causality: A Graph-Transformer Approach to Spatiotemporal Dependencies

2025-09-18

AI Summary

This study addresses the problem of high-accuracy short-term solar irradiance forecasting using only ground-based global horizontal irradiance (GHI) measurements from the target site X and nearby meteorological stations S—without requiring sky imagery or satellite data. To this end, we propose a lightweight causal spatiotemporal modeling framework. Methodologically, it innovatively integrates observable synchronous variables, latent regional climatic factors, and dynamic time-lagged effects within a unified architecture comprising embedding modules, a spatiotemporal graph neural network, and a gated Transformer—explicitly capturing inter-site causal relationships and nonlinear temporal shifts. Empirically evaluated across diverse geographical regions, our approach significantly outperforms state-of-the-art univariate, multivariate, and multimodal baselines; notably, it reduces prediction error by 25.9% relative to the commercial forecasting system Solcast. The framework thus provides a low-cost, easily deployable technical solution for renewable energy management.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Machine Learning: Causal Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Accurate solar forecasting underpins effective renewable energy management. We present SolarCAST, a causally informed model predicting future global horizontal irradiance (GHI) at a target site using only historical GHI from site X and nearby stations S - unlike prior work that relies on sky-camera or satellite imagery requiring specialized hardware and heavy preprocessing. To deliver high accuracy with only public sensor data, SolarCAST models three classes of confounding factors behind X-S correlations using scalable neural components: (i) observable synchronous variables (e.g., time of day, station identity), handled via an embedding module; (ii) latent synchronous factors (e.g., regional weather patterns), captured by a spatio-temporal graph neural network; and (iii) time-lagged influences (e.g., cloud movement across stations), modeled with a gated transformer that learns temporal shifts. It outperforms leading time-series and multimodal baselines across diverse geographical conditions, and achieves a 25.9% error reduction over the top commercial forecaster, Solcast. SolarCAST offers a lightweight, practical, and generalizable solution for localized solar forecasting.

Problem

• Predicting solar irradiance using historical data from multiple stations
• Modeling spatiotemporal dependencies with causal graph-transformer approach
• Reducing reliance on specialized hardware and preprocessing requirements

Innovation

• Graph-Transformer for spatiotemporal dependencies
• Causally informed model using only public sensor data
• Gated transformer learns temporal lagged influences