Generating EUPHEMIA-compatible bids for flexible demand under imperfect information

Technical Summary: Generating EUPHEMIA-Compatible Bids for Flexible Demand under Imperfect Information

Problem Statement
Electricity procurement represents a major operational cost for large industrial consumers, exposing them to significant financial risks from extreme price spikes. While many consumers possess load-shifting capabilities to mitigate these costs, expressing this multi-period flexibility in the European day-ahead electricity market is challenging. Consumers must submit bids compatible with the Pan-European Hybrid Electricity Market Integration Algorithm (EUPHEMIA), which clears the market for maximum social welfare.

Current literature often fails to bridge the gap between operational constraints and market requirements. Many proposed methods either ignore system operational constraints (such as ramping limits and storage dynamics) or generate single demand time series rather than the required price–quantity bids. Consequently, these approaches cannot fully express consumer flexibility or ensure operational feasibility under price uncertainty. While some existing work addresses EUPHEMIA-compatible bids (e.g., exclusive group bids), their applicability to real-world industrial systems with heterogeneous constraints and alternative bidding formats remains unclear.

Methodology
The authors propose a stochastic optimization framework to generate operationally feasible, EUPHEMIA-compatible bids under price uncertainty. The framework consists of four sequential steps:

1. Price Forecasting: A Prophet-style decomposition model generates base price forecasts, incorporating exogenous regressors like system load and renewable generation. To capture heavy tails and extreme price spikes, forecast residuals are modeled using a Lévy-stable distribution rather than a Gaussian distribution.
2. Scenario Generation: Point forecasts are converted into plausible price trajectories. Random noise is drawn from the fitted Lévy-stable distribution, and temporal correlations (intra-day and inter-day) are imposed via Cholesky factorization. Scenarios are scaled by the forecast horizon and reduced to a representative subset using K-means clustering to ensure computational tractability.
3. Stochastic Optimization: The core problem is formulated as a two-stage stochastic Mixed-Integer Linear Program (MILP).
   • First Stage: Decides market offers (bids) before prices are revealed.
   • Second Stage: Determines operational dispatch and system feasibility after a price scenario is realized.
   • Objective: The model minimizes a risk-averse objective function combining expected costs with a Conditional Value-at-Risk (CVaR) penalty to control downside risk.
4. Rolling Horizon: The strategy is evaluated using a rolling-horizon simulation. A multi-day planning horizon is solved to account for look-ahead constraints (e.g., storage dynamics), but only the first 24 hours are implemented. A redispatch step ensures feasibility if the cleared profile deviates from the optimizer's preferred schedule.

The study compares two distinct bidding strategies:

• Hourly Bids (HBs): Represent flexibility via stepwise demand curves in each hour (price–quantity pairs). This allows for high granularity and marginal price responsiveness but risks inter-temporal discontinuities if partial clearing occurs.
• Exclusive-Group Bids (EBs): Represent flexibility via a set of mutually exclusive, coherent consumption profiles submitted at a single opportunity cost. This ensures operational feasibility by selecting a full-day schedule but reduces the ability to arbitrage isolated high-price hours.

Case Studies
The methodology is evaluated on two industrial cases under two weather-driven market scenarios (a typical year and a severe "dunkelflaute" year with volatile prices):

1. Electrolyzer System: A hypothetical, highly flexible system with high ramping capabilities and weak intertemporal coupling, capable of significant load shifting.
2. Steel Plant: A real-world industrial facility (Tata Steel, Netherlands) with tightly coupled processes, strict ramping limits, and strong intertemporal dependencies.

Key Results
The economic performance of the bidding strategies is highly dependent on the operational characteristics of the load and market volatility:

• Bid Structure: For the highly flexible electrolyzer, HBs utilize fewer price–quantity steps, acting as threshold-based decisions. In contrast, the steel plant requires HBs with more steps to hedge against intertemporal feasibility risks. For EBs, the electrolyzer submits diverse profiles allowing the market to select low-price periods, while the steel plant submits clustered schedules due to tighter constraints.
• Cost Performance:
  • Electrolyzer: Under normal weather, HBs yield lower costs. However, under severe, volatile weather, EBs outperform HBs, achieving results closest to the perfect foresight benchmark. The ability to select a coherent schedule that matches realized price trajectories proves superior for highly flexible loads in volatile markets.
  • Steel Plant: HBs consistently outperform EBs across all weather scenarios. The plant's strong intertemporal constraints and limited flexibility make the granular, hour-by-hour responsiveness of HBs more effective than the bundled nature of EBs.

Significance and Contributions
The primary contribution of this work is the development of a method to translate multi-period operational flexibility into EUPHEMIA-compatible bids under price uncertainty. The paper demonstrates that:

1. There is no universally superior bidding format; the optimal strategy depends on the specific operational flexibility of the consumer and market conditions.
2. Highly flexible systems (e.g., electrolyzers) benefit from the schedule-selection capability of Exclusive-Group Bids in volatile markets.
3. Systems with strong intertemporal constraints (e.g., steel plants) benefit from the granularity of Hourly Bids to manage feasibility risks.

The authors conclude that such algorithms are necessary to address structural issues in current market design, enabling consumers with varying levels of flexibility to participate effectively in the European day-ahead market despite incomplete information.