Risk-aware stochastic scheduling of multi-market energy storage systems

Battle Count: 50

The paper is highly relevant to quantitative trading in energy markets. The CVaR-constrained framework directly addresses downside risk management in multi-market arbitrage (DA vs ID), which parallels portfolio risk management. The two-stage stochastic structure with here-and-now commitments and wait-and-see recourse mirrors options-like hedging strategies. The rolling-horizon implementation resembles algorithmic trading with periodic re-optimization. The risk-reward trade-off analysis and market participation ratio optimization are directly transferable to energy trading desk operations. However, the paper focuses on physical storage constraints and energy-specific modeling rather than general financial instruments, and uses a price-taker assumption that limits applicability to large-scale trading strategies.

Implementation requires: (1) LP solver (Gurobi) with large-scale problem handling (164k-245k variables, 180k-359k constraints for yearly horizon); (2) Scenario generation and SAA formulation; (3) CVaR linearization via Rockafellar-Uryasev reformulation with auxiliary variables; (4) Multi-market (DA+ID) dispatch modeling with intertemporal state coupling; (5) Physical process models (electrolyzer, storage, fuel cell, battery SOC/SOH dynamics); (6) Rolling-horizon re-optimization loop for online operation; (7) Sensitivity analysis across risk bounds and observation times. The mathematical formulation is well-defined but the engineering integration of physical constraints with stochastic optimization at scale is non-trivial.

The IHS case study uses publicly available ISO-NE 2024 hourly price data from EIA. The mathematical formulation is fully specified with all equations. However, BESS model parameters (self-discharge, round-trip efficiency, inverter size, degradation constant, max cycles) are proprietary and not disclosed. The BESS price data are industry-acquired predictions not publicly available. The two-stage approximation with single observation time is a simplification that may not capture full multi-stage market dynamics. No code repository is provided.