Optimizing Bidding Curves for Renewable Energy in Two-Settlement Electricity Markets

arXiv:2501.18732 · eess.SY, cs.SY, econ.GN, math.OC, q-fin.EC · Submitted 2025-01-30 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: I'm Rosa.

Dev: And I'm Dev, here with Taro.

Rosa: We're looking at the paper: "Optimizing Bidding Curves for Renewable Energy in Two-Settlement Electricity Markets".

Dev: Here's the gist of the paper. Coordination of day-ahead and real-time electricity markets is imperative for cost-effective electricity supply and also to provide efficient incentives for the energy transition This paper proposes a new approach for compatible coordination in two-settlement markets based on benchmark bidding curves for variable renewable energy These curves are optimized based on a bilevel optimization problem, anticipating per-scenario responses of deterministic market-clearing problems and ultimately minimizing the expected cost across day-ahead and real-time markets Although the general bilevel model is challenging to solve, we theoretically prove that a single-segment bidding curve with a zero bidding price is sufficient to achieve system optimality if the marginal cost of variable renewable energy is zero In practice, variable renewable energy producers can be allowed to bid multi-segment curves with non-zero prices We test the bilevel framework for both single- and multiple-segment bidding curves under the assumption of fixed bidding prices Market Modeling and Framework Setup The paper models day-ahead and real-time market clearing aligned with the foundational structure of the U.S.

Rosa: electricity market Conventional generation units are modeled with linear operating costs, and system demand is assumed to be inelastic with a high Value of Lost Load (VoLL) The VRE bidding curves are modeled as multi-segment curves where for each segment s of VRE supplier k at hour t, the bidding price is denoted by CWk,t,s and the bidding quantity by Wk,t,s The day-ahead market-clearing problem minimizes the day-ahead system cost which includes variable, no-load, and startup costs of conventional generation units along with the bidding costs of VRE producers Bilevel Optimization and Optimality Proof The core of the work is a bilevel stochastic framework (BiD) that jointly optimizes bidding prices and quantities of multi-segment bidding curves to minimize the total system cost, which is the sum of DAM and expected RTM costs The upper level decides the day-ahead VRE bidding curves S, i.e., the price and quantity pair (CW,W) The lower level optimizes the DAM schedule ΦDA under these bidding curves Theoretical Simplification for Optimality The paper proves that a single-segment bidding curve with a zero bidding price is sufficient to achieve optimality if the true marginal cost of VRE is zero Specifically, Problem BiD and Problem BiD-q achieve the same expected system cost Corollary 1 states that considering multi-segment bidding curves with one segment set by a zero price will achieve the same expected system cost as the original Problem BiD Scalability and Practical Testing To solve the bilevel problem for large-scale systems like a 1576-bus NYISO system, a method based on strong duality and McCormick envelope relaxation is used to relax the bilevel optimization problem into an LP Simulations on the NYISO system reveal that the multi-segment optimized bid demonstrates a significant cost-saving potential with respect to the myopic strategy, achieving a 36% reduction in hourly system costs compared to the baseline Performance Benchmarking The proposed framework is benchmarked against two strategies: Myopic dispatch (MyD) and Stochastic dispatch (StD) The results show that the system costs under BiD and MyD always satisfy S MyD ⩾ S BiD ⩾ S StD Furthermore, varying the bidding price shows that under BiD, the expected aggregate profit of wind producers remains almost the same when the bidding price is lower than 26/MWh The proposed framework internalizes the cost of real-time re-dispatch into day-ahead bidding curves, resulting in minimum expected dispatch costs across the market timeline The work suggests that this bilevel framework can serve as a tool to guide, monitor, or regulate VRE bids by developing risk scores Future work plans include incorporating energy storage and demand response resources The framework can inform system-optimal VRE bidding strategies Pruning redundant reasoning tokens The paper formulates a bilevel optimization framework to optimize DAM bidding curves of VRE producers The proposed framework internalizes the cost of real-time re-dispatch into day-ahead bidding curves, resulting in the minimum expected dispatch costs across the market timeline Although theoretically the multi- and single-segment curves achieve the same system cost, the simulation results can have slight differences Specifically, the optimized biding curves allow for a 36% (−155k) reduction in hourly system costs with respect to the baseline Moreover, this cost-saving result is very close to that under the scenario-based stochastic market clearing, which provides the ideal coordination yet remains incompatible with existing market practices Overall, our proposed framework can inform system-optimal VRE bidding strategies The current model does not account for energy storage and demand response These resources can be crucial for supporting deeply decarbonized electricity markets and we plan to incorporate these aspects in future work Additionally, our existing bilevel optimization framework employs a centralized approach In the future, we aim to develop incentive mechanisms to guide VRE producers’ bidding strategies ensuring alignment with the optimal outcomes of the centralized framework How the training signal is built The paper tests single- and multi-segment bidding curves for fixed input bidding prices, using an empirical setting where zero price is included in the curve The authors empirically set prices for 6-segment bidding curves with CW1 = 0/MWh and subsequent non-zero prices up to CW6 = 350/MWh These pricing levels are set based on the costs of the first 20%, 40%, 60%, and 80% of the total conventional capacity The optimal bidding curves derived from the bilevel model vary for different hours and wind farms, being significantly affected by the expected wind power forecasts Where the gains come from The primary gain comes from internalizing the cost of real-time re-dispatch into day-ahead bidding curves to minimize expected system cost The ability to adjust bidding quantities corresponding to given bidding prices allows the framework to achieve good performance in terms of system cost The theoretical proof shows that a single-segment curve with zero price is sufficient for optimality if the true marginal cost of VRE is zero This approach can potentially serve as a tool to guide or regulate VRE producers’ bidding strategies The framework can inform system-optimal VRE bidding strategies Pruning redundant reasoning tokens The paper formulates a bilevel optimization framework to optimize DAM bidding curves of VRE producers The proposed framework internalizes the cost of real-time re-dispatch into day-ahead bidding curves resulting in the minimum expected dispatch costs across the market timeline Although theoretically the multi- and single-segment curves achieve the same system cost simulation results can have slight differences Specifically, the optimized biding curves allow for a 36% (−155k) reduction in hourly system costs with respect to the baseline Moreover, this cost-saving result is very close to that under the scenario-based stochastic market clearing which provides the ideal coordination yet remains incompatible with existing market practices Overall, our proposed framework can inform system-optimal VRE bidding strategies The current model does not account for energy storage and demand response resources These resources can be crucial for supporting deeply decarbonized electricity markets and we plan to incorporate these aspects in future work Additionally, our existing bilevel optimization framework employs a centralized approach In the future, we aim to develop incentive mechanisms to guide VRE producers’ bidding strategies ensuring alignment with the optimal outcomes of the centralized framework Pruning redundant reasoning tokens

Title and authors: Rosa: Let’s look at what they actually propose in this paper, "Optimizing Bidding Curves for Renewable Energy in Two-Settlement Electricity Markets." They are proposing this new approach for compatible coordination based on benchmark bidding curves for variable renewable energy.

Dev: They set up a bilevel optimization problem where the upper level decides the day-ahead bidding curves, which is the price and quantity pair, and the lower level optimizes that day-ahead schedule under those bids.

Rosa: The goal of this framework is to minimize the total system cost, which they define as adding up everything from day-ahead market costs to expected real-time market costs.

Taro: It’s trying to get that system cost down across the entire timeline, not just optimizing one market in isolation.

Dev: And there's a really neat theoretical proof here because they show that even though the general bilevel model is hard to solve, they prove that a single-segment bidding curve with a zero bidding price is enough to achieve system optimality if the actual marginal cost of variable renewable energy is zero.

Rosa: That’s important because it addresses the computational challenge of solving those complex models by showing a simpler condition under which you hit the optimal point.

The paper's summary: Dev: Now, when we talk about what makes this paper an improvement, they are focusing on how they can actually make this framework usable in a real-world setting.

Rosa: One improvement is that while the general model is tough to solve, they use a method based on strong duality and McCormick envelope relaxation to relax the bilevel optimization problem into something solvable, like an linear program.

Taro: That’s a practical step because solving those high-dimensional bilevel problems directly in a large system like the one thousand five hundred seventy-six-bus NYISO system is computationally prohibitive otherwise.

Dev: Then they test this framework on that large system and the simulations show that the multi-segment optimized bid demonstrates a significant cost-saving potential compared to just using a simple myopic strategy, achieving about a thirty-six percent reduction in hourly system costs.

Rosa: That thirty-six percent saving over the baseline is pretty substantial, especially when you compare it to how much we usually save by coordinating day-ahead and real-time operations separately.

The paper's improvements: Dev: So wrapping up this paper "Optimizing Bidding Curves for Renewable Energy in Two-Settlement Electricity Markets," the main implication is that internalizing the cost of real-time re-dispatch into those day-ahead bidding curves results in minimum expected dispatch costs across the market timeline.

Rosa: It suggests that this bilevel framework can be a tool to guide or monitor variable renewable energy bids by developing risk scores, which could help producers make better decisions about what to bid.

Taro: And they point out one limitation, though, and they say this current model doesn't account for energy storage or demand response resources yet, which are crucial for supporting deeply decarbonized electricity markets.

Dev: They plan to incorporate those storage and demand response aspects in future work, but right now the framework is centralized in its approach.

Rosa: Ultimately, the work on "Optimizing Bidding Curves for Renewable Energy in Two-Settlement Electricity Markets" shows how this bilevel framework can inform system-optimal variable renewable energy bidding strategies.

Conclusion: Rosa: So we've talked about how they set up this bilevel optimization framework to coordinate day-ahead and real-time bidding curves for variable renewable energy, and it’s all about minimizing the total expected system cost across both markets.

Dev: Right, so the core of it is that the upper level chooses the price and quantity pairs for those multi-segment curves, and then the lower level optimizes that day-ahead schedule under those chosen bids.

Taro: I found their theoretical proof pretty interesting because they show you only need a single segment with a zero bidding price if the actual marginal cost of variable renewable energy is zero to get the optimal result.

Rosa: That’s a big thing because it simplifies things mathematically, showing that even with complex multi-segment curves, you can hit optimality under certain conditions.

Dev: But in practice, they test this on a large system like the NYISO system and find that the optimized bidding curves actually show a significant cost-saving potential compared to just following a simple myopic strategy.

Taro: A thirty-six percent reduction in hourly costs is substantial, especially when you look at how close that result is to what you'd get from scenario based stochastic market clearing, which is usually the ideal coordination method.

Rosa: It’s about internalizing the cost of real-time re-dispatch into those day-ahead bids, which means you get minimum expected dispatch costs over the entire timeline instead of just optimizing one point in time.

Dev: That framework can be used to guide or regulate those renewable energy bids by developing risk scores based on forecasts, which is a practical application for real-time control loops.

Taro: It does stop short because the current model doesn't account for energy storage or demand response resources, and that’s where future work needs to go if you want to support deeply decarbonized markets.

Rosa: Exactly, so this paper "Optimizing Bidding Curves for Renewable Energy in Two-Settlement Electricity Markets" gives us a strong tool for system optimal bidding strategies right now.

Dev: It's a good piece of modeling that helps show how to get better coordination between different market times.

Taro: Next up, we’re looking at some papers on robotics and how they handle unexpected challenges in real-world environments, so we’ll switch gears then.

Argonne National Laboratory · Laboratory for Information and Decision Systems, Massachusetts Institute of Technology, Cambridge, Massachusetts Department of Electrical Engineering and Computer Science, University of Michigan Ann Arbor Department of Economics and Department of Industrial and Systems Engineering, Lehigh University Bethlehem Pennsylvania

eess.SY, cs.SY, econ.GN, math.OC, q-fin.EC

Submitted: 2025-01-30

Updated: 2026-11-10

DOI: 10.1109/PMAPS70385.2026.11762301

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 88/100

The gist: Coordination of day-ahead and real-time electricity markets is imperative for cost-effective electricity supply and also to provide efficient incentives for the energy transition This paper proposes…

Key concepts

Bilevel Optimization
This is a mathematical technique used when one decision (the upper level, like setting bidding prices) depends on the optimal solution of another decision (the lower level, like market clearing). Here, it jointly optimizes VRE bidding prices and quantities to minimize the total expected cost across both markets.
Multi-segment Bidding Curves
Instead of a single price for all VRE supply, this model allows producers to bid different prices for different portions (segments) of their energy output. This flexibility helps achieve better coordination between day-ahead and real-time markets by allowing producers to respond optimally.
Internalizing Real-Time Costs
The framework integrates the cost associated with future real-time re-dispatch into the initial day-ahead bidding decisions. By doing this, the system seeks a minimum expected dispatch cost over the entire market timeline rather than just optimizing for one market period.
Optimality Proof (Zero Price Case)
The paper proves that if VRE has zero marginal cost, a single-segment bidding curve with a zero bid price is mathematically sufficient to achieve the best possible system outcome. This provides a theoretical baseline for understanding when simple pricing strategies are enough.

Terminology

Summary

Coordination of day-ahead and real-time electricity markets is imperative for cost-effective electricity supply and also to provide efficient incentives for the energy transition This paper proposes a new approach for compatible coordination in two-settlement markets based on benchmark bidding curves for variable renewable energy These curves are optimized based on a bilevel optimization problem, anticipating per-scenario responses of deterministic market-clearing problems and ultimately minimizing the expected cost across day-ahead and real-time markets Although the general bilevel model is challenging to solve, we theoretically prove that a single-segment bidding curve with a zero bidding price is sufficient to achieve system optimality if the marginal cost of variable renewable energy is zero In practice, variable renewable energy producers can be allowed to bid multi-segment curves with non-zero prices We test the bilevel framework for both single- and multiple-segment bidding curves under the assumption of fixed bidding prices

Market Modeling and Framework Setup

The paper models day-ahead and real-time market clearing aligned with the foundational structure of the U.S. electricity market Conventional generation units are modeled with linear operating costs, and system demand is assumed to be inelastic with a high Value of Lost Load (VoLL) The VRE bidding curves are modeled as multi-segment curves where for each segment s of VRE supplier k at hour t, the bidding price is denoted by CWk,t,s and the bidding quantity by Wk,t,s The day-ahead market-clearing problem minimizes the day-ahead system cost which includes variable, no-load, and startup costs of conventional generation units along with the bidding costs of VRE producers

Bilevel Optimization and Optimality Proof

The core of the work is a bilevel stochastic framework (BiD) that jointly optimizes bidding prices and quantities of multi-segment bidding curves to minimize the total system cost, which is the sum of DAM and expected RTM costs The upper level decides the day-ahead VRE bidding curves S, i.e., the price and quantity pair (CW,W) The lower level optimizes the DAM schedule ΦDA under these bidding curves

Theoretical Simplification for Optimality

The paper proves that a single-segment bidding curve with a zero bidding price is sufficient to achieve optimality if the true marginal cost of VRE is zero Specifically, Problem BiD and Problem BiD-q achieve the same expected system cost Corollary 1 states that considering multi-segment bidding curves with one segment set by a zero price will achieve the same expected system cost as the original Problem BiD

Scalability and Practical Testing

To solve the bilevel problem for large-scale systems like a 1576-bus NYISO system, a method based on strong duality and McCormick envelope relaxation is used to relax the bilevel optimization problem into an LP Simulations on the NYISO system reveal that the multi-segment optimized bid demonstrates a significant cost-saving potential with respect to the myopic strategy, achieving a 36% reduction in hourly system costs compared to the baseline

Performance Benchmarking

The proposed framework is benchmarked against two strategies: Myopic dispatch (MyD) and Stochastic dispatch (StD) The results show that the system costs under BiD and MyD always satisfy S MyD ⩾ S BiD ⩾ S StD Furthermore, varying the bidding price shows that under BiD, the expected aggregate profit of wind producers remains almost the same when the bidding price is lower than 26/MWh

Conclusion and Implications

The proposed framework internalizes the cost of real-time re-dispatch into day-ahead bidding curves, resulting in minimum expected dispatch costs across the market timeline The work suggests that this bilevel framework can serve as a tool to guide, monitor, or regulate VRE bids by developing risk scores Future work plans include incorporating energy storage and demand response resources The framework can inform system-optimal VRE bidding strategies

Pruning redundant reasoning tokens

The paper formulates a bilevel optimization framework to optimize DAM bidding curves of VRE producers The proposed framework internalizes the cost of real-time re-dispatch into day-ahead bidding curves, resulting in the minimum expected dispatch costs across the market timeline Although theoretically the multi- and single-segment curves achieve the same system cost, the simulation results can have slight differences Specifically, the optimized biding curves allow for a 36% (−155k) reduction in hourly system costs with respect to the baseline Moreover, this cost-saving result is very close to that under the scenario-based stochastic market clearing, which provides the ideal coordination yet remains incompatible with existing market practices Overall, our proposed framework can inform system-optimal VRE bidding strategies The current model does not account for energy storage and demand response These resources can be crucial for supporting deeply decarbonized electricity markets and we plan to incorporate these aspects in future work Additionally, our existing bilevel optimization framework employs a centralized approach In the future, we aim to develop incentive mechanisms to guide VRE producers’ bidding strategies ensuring alignment with the optimal outcomes of the centralized framework

How the training signal is built

The paper tests single- and multi-segment bidding curves for fixed input bidding prices, using an empirical setting where zero price is included in the curve The authors empirically set prices for 6-segment bidding curves with CW1 = 0/MWh and subsequent non-zero prices up to CW6 = 350/MWh These pricing levels are set based on the costs of the first 20%, 40%, 60%, and 80% of the total conventional capacity The optimal bidding curves derived from the bilevel model vary for different hours and wind farms, being significantly affected by the expected wind power forecasts

Where the gains come from

The primary gain comes from internalizing the cost of real-time re-dispatch into day-ahead bidding curves to minimize expected system cost The ability to adjust bidding quantities corresponding to given bidding prices allows the framework to achieve good performance in terms of system cost The theoretical proof shows that a single-segment curve with zero price is sufficient for optimality if the true marginal cost of VRE is zero This approach can potentially serve as a tool to guide or regulate VRE producers’ bidding strategies The framework can inform system-optimal VRE bidding strategies

The paper formulates a bilevel optimization framework to optimize DAM bidding curves of VRE producers The proposed framework internalizes the cost of real-time re-dispatch into day-ahead bidding curves resulting in the minimum expected dispatch costs across the market timeline Although theoretically the multi- and single-segment curves achieve the same system cost simulation results can have slight differences Specifically, the optimized biding curves allow for a 36% (−155k) reduction in hourly system costs with respect to the baseline Moreover, this cost-saving result is very close to that under the scenario-based stochastic market clearing which provides the ideal coordination yet remains incompatible with existing market practices Overall, our proposed framework can inform system-optimal VRE bidding strategies The current model does not account for energy storage and demand response resources These resources can be crucial for supporting deeply decarbonized electricity markets and we plan to incorporate these aspects in future work Additionally, our existing bilevel optimization framework employs a centralized approach In the future, we aim to develop incentive mechanisms to guide VRE producers’ bidding strategies ensuring alignment with the optimal outcomes of the centralized framework

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Bilevel optimization framework for bidding curves: The improved system can jointly optimize day-ahead bidding prices and quantities of multi-segment bidding curves to minimize expected system cost, addressing a key research gap in the literature.

  2. Optimality guaranteed by single-segment bidding curves: The system can leverage the proof that a single-segment bidding curve with a zero bidding price is sufficient to achieve optimality if the marginal cost of variable renewable energy is zero for efficient solution finding.

  3. Simulations on large-scale systems: The system can utilize the method based on strong duality and McCormick envelope relaxation to solve the bilevel optimization problem for large systems, such as a 1576-bus NYISO system.

  4. Practical insights: The framework can serve as a tool to guide or regulate VRE producers’ bidding strategies by comparing benchmark results with forecasts, potentially developing risk scores for coordination mechanisms.

  5. Multi-segment bidding curves: The system can set up optimal bidding quantities for different segments of a multi-segment curve to approximate the system optimum when fixed bidding prices are given, showing a significant cost-saving potential with respect to the myopic strategy.

Abstract

Coordination of day-ahead and real-time electricity markets is imperative for cost-effective electricity supply and also to provide efficient incentives for the energy transition. Although stochastic market designs feature the least-cost coordination, they are incompatible with current deterministic markets. This paper proposes a new approach for compatible coordination in two-settlement markets based on benchmark bidding curves for variable renewable energy. These curves are optimized based on a bilevel optimization problem, anticipating per-scenario responses of deterministic market-clearing problems and ultimately minimizing the expected cost across day-ahead and real-time markets. Although the general bilevel model is challenging to solve, we theoretically prove that a single-segment bidding curve with a zero bidding price is sufficient to achieve system optimality if the marginal cost of variable renewable energy is zero, thus addressing the computational challenge. In practice, variable renewable energy producers can be allowed to bid multi-segment curves with non-zero prices. We test the bilevel framework for both single- and multiple-segment bidding curves under the assumption of fixed bidding prices. We leverage duality theory and McCormick envelopes to derive the linear programming approximation of the bilevel problem, which scales to practical systems such as a 1576-bus NYISO system. We benchmark the proposed coordination and find absolute dominance over the baseline solution, which assumes that renewables agnostically bid their expected forecasts. We also demonstrate that our proposed scheme provides a good approximation of the least-cost, yet unattainable in practice, stochastic market outcome.

Sources

Related papers