Approximate Dynamic Programming for Degradation-aware Market Participation of Battery Energy Storage Systems: Bridging Market and Degradation Timescales

Battle Count: 50

The paper addresses sequential decision-making under uncertainty in electricity markets, which shares structural similarities with algorithmic trading (stochastic prices, delayed rewards, state-dependent decisions). The ADP/MPC framework and timescale separation techniques are transferable to trading contexts. However, the paper is specifically about physical battery asset management in energy markets, not financial instrument trading. The methodology (value function approximation, backward induction, one-step MPC) has conceptual parallels to optimal execution and market-making strategies, but the domain, constraints, and objectives differ substantially.

High complexity due to: (1) PDAE discretization of physics-based battery models (SPM with SEI growth), (2) nested time-scale formulation with 10s increments within bidding intervals, (3) offline backward induction requiring solving MPC problems at 100 health grid points × SoC grid × 1138 market scenarios, (4) parametric regression for value function and gradient, (5) lifting map construction, (6) Gauss-Radau collocation for DAE-to-discrete-time conversion, (7) KL decomposition for uncertainty modeling, (8) primal-dual interior point solver for NLP. Requires expertise in battery electrochemistry, stochastic control, numerical optimization, and multiscale modeling.

The algorithm (Algorithm 1) is described in detail with explicit mathematical formulations. Battery model parameters are sourced from published literature (Forman et al. 2012, Cao et al. 2020). Market data is from public sources (RTE, ENTSO-E). However, no code or GitHub repository is provided. The uncertainty model construction (KL decomposition, Gauss-Legendre quadrature) is described but implementation details for the full pipeline are not fully specified. The regression basis functions and hyperparameters are given.