Work

Priority-Weighted Clustering and Prediction Intervals for AI-Driven Biogas Energy Forecasting

Year: 2026

Type: article

Abstract: Biogas plays an increasingly important role in advancing decarbonisation goals, particularly in rural regions where livestock and agricultural waste can be converted into renewable energy. However, predicting annual farm-level biogas electricity output remains challenging due to operational variability, uncertain co-digestion practices, and external constraints such as grid integration and environmental disruptions. Traditional point prediction models often provide limited support for practical decision-making because they do not explicitly represent uncertainty around the estimated output. To address this limitation, this paper proposes a data-driven prediction interval framework for annual farm-level biogas electricity-output estimation. The study does not model future temporal horizons; rather, it estimates annual electricity generation at the farm level and constructs prediction intervals around these estimates. The proposed framework combines Self-Organising Map (SOM)-based clustering with Lower Upper Bound Estimation (LUBE) to account for operational heterogeneity among biogas-producing farms. SOM clustering is performed using pre-prediction operational covariates, specifically cattle population and co-digestion status, while electricity output is used only as the prediction target and for post-hoc interpretation. For each operational cluster, a neural prediction interval model is trained using the LUBE approach. A priority-weighted extension is then incorporated to reflect cluster-level operational or strategic importance in the evaluation of interval performance. Experiments on a real-world dataset from U.S. biogas systems show that the proposed framework can support a trade-off between interval width and coverage performance across standard confidence levels. By combining operational clustering with uncertainty-aware interval estimation, the method improves the interpretability and practical relevance of annual farm-level biogas electricity-output prediction for infrastructure planning.

Source: Sustainability

DOI: 10.3390/su18157865

Authors: Mohammad Anwar Hosen, Nazmus Sakib, Michael Johnstone, Burhan Khan, Douglas Creighton

Institution: Deakin University

Language: English

Topic: Energy Load and Power Forecasting

Subfield: Electrical and Electronic Engineering

Field: Engineering

Domain: Physical Sciences

SDGs: Climate action, Affordable and clean energy, Industry, innovation and infrastructure, Zero hunger

Open Access status: gold

Estimated APC paid: $2,896