Battery Storage Co-Optimization in Day-Ahead and Real-Time Markets with Bayesian Optimization

Battle Count: 50

Highly relevant to quantitative trading in electricity markets. The paper addresses the core problem of co-optimizing static (DA) and dynamic (RT) trading decisions under uncertainty, which is analogous to multi-timescale trading strategies. The Bayesian optimization framework for expensive black-box objectives is broadly applicable to algorithmic trading. The DART spread modeling and energy arbitrage strategies are directly relevant to power trading desks. However, the specific application to battery storage dispatch narrows the immediate applicability compared to general financial trading. The methodology (BO + stochastic control) is transferable to other trading contexts.

High implementation complexity due to the multi-layer architecture: (1) GP-based Bayesian Optimization with UCB acquisition over constrained polytopes, (2) SHADOw-GP actor-critic Regression Monte Carlo solver for the inner RT stochastic control problem with piecewise-linear dynamics and state-dependent constraints, (3) adaptive refinement logic with scoring rules, pruning, and warm-starting, (4) MILP solver for DA-only baselines, (5) Monte Carlo simulation for value estimation. Requires expertise in stochastic control, Bayesian optimization, Gaussian processes, and energy market mechanics. The paper references BoTorch for BO and builds on prior SHADOw-GP implementation.

The paper provides detailed parameter settings (Table 1), algorithm pseudocode (Algorithms 1 and 2), and specifies the use of BoTorch for BO and SHADOw-GP for RT solving. However, no code repository is explicitly linked. The CAISO SP-15 price data is sourced from Grid Status (2026), a public Python API. The RT price model parameters (κ=0.2, σ=1.0, λ=0.05) and BESS characteristics are fully specified. Reproduction would require implementing the SHADOw-GP solver from the authors' prior work (Aung & Ludkovski, 2024, 2025).