Abstract
Abstract With the increasing demand for sustainable energy solutions, accurate biomass forecasting has become essential for optimizing renewable energy utilization and supporting system-level predictive maintenance and operational decision-making. However, the complexity and high dimensionality of biomass energy data pose significant challenges for conventional forecasting approaches. To address these challenges, this study proposes a framework integrating Spatio-Temporal Graph Convolutional Networks (STGCN) with metaheuristic optimization techniques. The proposed approach improves forecasting performance by capturing spatio-temporal dependencies while optimizing feature selection and model hyperparameters. Initially, the baseline STGCN model achieved an MSE of 0.0025, RMSE of 0.0500, and $$R^2$$ of 0.8317. Feature selection using the Binary Comment Feedback Optimization Algorithm (bCFOA) reduced redundancy and improved the forecasting results to an MSE of 0.0018, RMSE of 0.04243, and $$R^2$$ of 0.912. Further improvement through CFOA-based hyperparameter optimization produced the best performance, achieving an MSE of $$0.000554 \pm 0.000012$$ , RMSE of $$0.02354 \pm 0.00028$$ , MAE of $$0.00410 \pm 0.00009$$ , and $$R^2$$ of $$0.981 \pm 0.002$$ . The results demonstrate that combining graph-based spatio-temporal learning with evolutionary optimization significantly improves forecasting accuracy and model generalization. Moreover, the proposed CFOA-STGCN framework provides a scalable, interpretable, and computationally efficient solution suitable for biomass energy forecasting and operational decision-support systems.