Contribution type: Lightning talk and Poster
Capacity Withholding in Electricity Markets: A Policy Function Approximation of Equilibrium Problems with Equilibrium Constraints
Presenter: Jan Ot Pinya Urgell
Description: Electricity markets are prone to oligopolistic concentration, giving some participants the ability to move prices away from marginal-cost levels through economic or physical capacity withholding. This kind of strategic behaviour raises clearing prices, displaces efficient generation, and reduces consumer welfare. Representing it typically requires bi-level or equilibrium formulations (MPEC/EPEC), but these are difficult to solve at the scale of realistic, multi-firm, multi-technology portfolios, which has kept most existing studies limited to single plants or technologies. Agent-based modelling offers another way to represent this kind of decentralized, strategic decision-making, but scaling it to systems with many interacting agents raises similar computational and convergence challenges.
This work presents a Policy Function Approximation (PFA) approach for integrating strategic capacity withholding into large-scale Energy System Optimization Models. The methodology first formulates the strategic bidding problem of a single profit-maximizing firm as a Mathematical Program with Equilibrium Constraints (MPEC), then extends it to a multi-firm Equilibrium Problem with Equilibrium Constraints (EPEC). To avoid repeatedly solving the computationally expensive MPEC step, the PFA learns, for each strategic firm, a piecewise-linear mapping from system load to that firm’s strategic dispatch deviation. This mapping is constructed by evaluating the firm’s MPEC over a representative set of load points and interpolating the resulting dispatch-withholding observations; heuristics are added to handle out-of-sample load levels and discontinuities in the underlying bidding response.
Background: This work is carried out at TU Delft within the MuESSLi (Multi-Energy System Smart Linking) project, which builds data-driven proxies of energy system models to enable their coupling across sectors and actors. MuESSLi spans five interlinked PhD topics across TU Delft, DTU and UP Comillas, together with industrial partners RTE, TotalEnergies and NaTran.
Optional links: This work is part of a PhD conducted within CRESYM (Collaborative Research for Energy SYstem Modeling), cresym.eu/muessli . A conference paper describing the method is currently in preparation.