(234d) Advances in Scenario Generation for Energy Investment Planning:Data-Driven Uncertainty Quantification in Stochastic Optimization

Global energy markets are under rising pressure as they tackle soaring demand, volatile supply chains, and intensifying geopolitical tensions. This growing uncertainty and volatility, driven by unforeseen demand trends, events and supply chain disruptions, have significantly complicated long-term planning and strategic decision-making. To address these challenges, stochastic optimization models are employed to capture a wide range of possible futures, enabling more resilient and informed decision-making [1]. However, a crucial assumption in stochastic programming is that a scenario tree is provided to represent the probability distribution of the underlying stochastic process. The structure and accuracy of this scenario tree play a crucial role in determining the accuracy of the optimization outcomes, where a poor scenario-generation method can significantly compromise the reliability of the model, leading to suboptimal or misleading strategic decisions.

This study presents a novel data-driven framework designed to support strategic energy infrastructure planning under uncertainty. The framework includes: (i) the development of predictive models to explore future possibilities, (ii) the quantification of uncertainty in predicted outcomes, (iii) the generation of probabilistic scenarios based on these predictions and their associated uncertainties, and (iv) the integration of these scenarios into a stochastic, multi-period mixed-integer linear programming (MILP) model. The proposed methodology is applied to the Texas energy market to address uncertainties in energy generation and enhance decision-making for capacity expansion and investment planning. To this end, machine learning (ML) techniques are employed for both forecasting and uncertainty quantification, while statistical methods are used for scenario generation. By explicitly accounting for uncertainty, this approach enables the formulation of more robust, flexible, and adaptive strategies for future investment and operational decisions.

Keywords: stochastic modeling, uncertainty quantification, data-driven scenario generation

Reference

[1] Funda Iseri, Halil Iseri, Harsh Shah, Eleftherios Iakovou, Efstratios N. Pistikopoulos, Planning strategies in the energy sector: Integrating Bayesian neural networks and uncertainty quantification in scenario analysis & optimization, Computers & Chemical Engineering, https://doi.org/10.1016/j.compchemeng.2025.109097, 2025.