Work

Adaptive energy systems planning: A rolling horizon framework with AI-enabled scenario generation

Year: 2026

Type: article

Abstract: Energy systems face unprecedented challenges driven by deep uncertainty arising from evolving demand patterns, variable renewable integration, changing technology costs, and structural shifts. Building adaptive and flexible planning frameworks is becoming increasingly essential to hedge against these uncertainties, as traditional static optimization approaches that fix investment decisions in advance often struggle to adapt to rapidly evolving conditions. This paper presents an adaptive optimization framework that integrates Bayesian Neural Networks with Markov Chain Monte Carlo sampling (BNN-MCMC) for technology-specific probabilistic generation forecasting within a rolling-horizon stochastic programming formulation. Unlike conventional approaches that rely on static forecasts, the proposed methodology continuously updates scenario representations as new information becomes available, enabling investment decisions that respond more effectively to evolving system conditions. We formalize this integration through the α -BASIS framework, an adaptive extension of Bayesian Augmented Scenario Integration for Stochastic modeling, which transforms predictive distributions into probability-weighted scenarios and embeds them within a rolling-horizon stochastic programming formulation. Through an illustrative case study of the Texas energy network, we compare three planning strategies utilizing historical generation profiles of gas power plants, solar and wind farms: conventional stochastic optimization, adaptive rolling-horizon stochastic optimization, and deterministic optimization with perfect information. Results demonstrate that the adaptive rolling-horizon approach achieves total system costs only 9.7% above the perfect-information benchmark. The rolling-horizon strategy shows particular strength in incorporating updated scenario information, time-varying technology-cost trajectories, and investment timing. Overall, this work provides a practical methodology for energy planners facing deep uncertainty by bridging data-driven probabilistic forecasting with adaptive decision-making under uncertainty.

Source: Applied Energy

DOI: 10.1016/j.apenergy.2026.128500

Authors: Funda Iseri, Halil Iseri, Mahmoud El-Halwagi, Eleftherios Iakovou, Efstratios N. Pistikopoulos

Institution: Texas A&M University

Language: English

Topic: Integrated Energy Systems Optimization

Subfield: Electrical and Electronic Engineering

Field: Engineering

Domain: Physical Sciences

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

Open Access status: hybrid

Estimated APC paid: $4,430