Optimal Trading of a Charging-Station Company in Auction Markets for Electricity

TL;DR

The paper tackles profit optimization for a charging-station operator (Chargco) participating in day-ahead and intraday electricity auctions under price uncertainty. It builds a two-stage stochastic MIQP to model DA/ID trading, then linearizes it into a MILP using SOS2-based piecewise linearization for $w_{1t}=(0.5(l_t+p_t))^2$ and a max-affine approach for $w_{2t}=(0.5(l_t-p_t))^2$, enabling tractable optimization. Scenario uncertainty is captured with GAN-based clustering of price paths, complemented by a surrogate model that links electricity/hydrogen loads to prices via random forest and linear regression. The solution framework, ILSD, decomposes the problem into Master and Sub-Problems, enhanced by warm starts, infeasibility handling, valid inequalities, and multi-cut disaggregation, and demonstrates strong computational performance compared with standard methods. Overall, the approach provides profitable, scalable decision tools for charging-station operators operating in volatile electricity markets, with clear pathways to handle time-coupled constraints in future work.

Abstract

This paper addresses a charging-station company (Chargco) for electric and hydrogen vehicles. The optimal trading of the Chargco in day-ahead and intraday auction markets for electricity is modeled as a stochastic Mixed-Integer Quadratic Program (MIQP). We propose a series of linearization and reformulation techniques to reformulate the stochastic MIQP as a mixed-integer linear program (MILP). To model stochasticity, we utilize generative adversarial networks to cluster electricity market price scenarios. Additionally, a combination of random forests and linear regression is employed to model the relationship between Chargco electricity and hydrogen loads and their selling prices. Finally, we propose an Improved L-Shaped Decomposition (ILSD) algorithm to solve our stochastic MILP. Our ILSD algorithm not only addresses infeasibilities through an innovative approach but also incorporates warm starts, valid inequalities and multiple generation cuts, thereby reducing computational complexity. Numerical experiments illustrate the Chargco trading using our proposed stochastic MILP and its solution algorithm.

Figures (13)

• Figure 1: The Chargco owning ECSs and HCSs participating in markets
• Figure 2: The sequence of events in the DA and ID markets.
• Figure 3: Two-stage stochastic optimization problem.
• Figure 4: Schematic diagram of the random forest algorithm.
• Figure 5: Generative adversarial network structure.

Theorems & Definitions (4)

• Lemma 1
• proof
• Lemma 2
• proof