Ensemble methods improve photovoltaic forecast accuracy with careful component selection
Ensemble Complexity in Photovoltaic Forecasting
Machine LearningArtificial Intelligence
Summary
Predicting how much solar power will be generated can be improved by combining many prediction methods, but adding too many can slow things down or not help much. The authors tested different combinations of forecasting tools on multiple real solar datasets and found that some combinations work better than others. They also discovered that simply relying on weather data doesn't always boost predictions. Their work shows it’s important to test each part of a forecasting system carefully, especially when using weather information and avoiding repeated tests on the same data.
What this means in practice
- •For energy grid operators: Improve solar power output forecasts using tested ensemble methods that balance accuracy and computation based on solar datasets.
- •For renewable energy software developers: Design solar forecasting software that selects ensemble members carefully and limits weather data dependence to enhance reliability.
Authors
Sun Ze, Zhou Liguo, Xu Yuqing, Yu Lei, Jiang Mingming
Abstract
An ensemble can improve photovoltaic forecasts while adding components that contribute little or increase computation. We assess these effects through matched comparisons and ablations of a fixed heterogeneous predictor bank. Hourly experiments use GEFCom2014 and three additional public datasets, with chronological partitions and three seeds. Under retrospective ERA5 assistance, static fusion reduces scaled mean absolute error against matched boosting by 1.11%, 4.41%, and 1.63% on PVDAQ, OPSD, and Ausgrid; only OPSD remains supported after multiple-comparison correction. Weather gating offers no consistent incremental benefit. Exploratory member removals show group-level dependence alongside individual redundancy. A separate, previously inspected fifteen-minute case replaces one neural member with a tree predictor: normalized error falls by 1.72%, but measured inference is slower. These findings support component-wise evaluation with explicit limits on weather availability and test-set reuse.