GPU-Accelerated Dynamic Programming for Multistage Stochastic Energy Storage Arbitrage

TL;DR

This paper introduces a GPU-accelerated dynamic programming approach for energy storage arbitrage that significantly speeds up computations while maintaining high accuracy, enabling practical real-time decision-making in electricity markets.

Contribution

The paper presents a novel tensor-based GPU implementation of dynamic programming for multistage stochastic energy storage valuation, overcoming computational limitations of previous models.

Findings

Achieves up to 100x speedup over CPU-based methods.

Attains 8,000x faster computation than commercial MILP solvers.

Maintains sub-0.3% optimality gap compared to exact solutions.

Abstract

We develop a GPU-accelerated dynamic programming (DP) method for valuing, operating, and bidding energy storage under multistage stochastic electricity prices. Motivated by computational limitations in existing models, we formulate DP backward induction entirely in tensor-based algebraic operations that map naturally onto massively parallel GPU hardware. Our method accommodates general, potentially non-concave payoff structures, by combining a discretized DP formulation with a convexification procedure that produces market-feasible, monotonic price-quantity bid curves. Numerical experiments using ISO-NE real-time prices demonstrate up to a 100x speedup by the proposed GPU-based DP method relative to CPU computation, and an 8,000x speedup compared to a commercial MILP solver, while retaining sub-0.3% optimality gaps compared to exact benchmarks.