Source-linked AI summary
Consensus-based approach to peer-to-peer electricity markets with product differentiation
Etienne Sorin, Lucien Bobo, Pierre Pinson
TL;DR
The paper addresses the mismatch between increasingly distributed, flexible power systems and conventional electricity-market structures. It proposes a peer-to-peer MBED market with product differentiation and solves it using decentralized RCI optimization. The framework supports preference-sensitive market outcomes and decentralized clearing with small optimality gaps and limited information exchange.
Problem
Electricity markets have not yet embraced power systems’ increasingly distributed structure and consumers’ more proactive role.
Method
The paper introduces a peer-to-peer Multi-Bilateral Economic Dispatch market with product differentiation and a decentralized Relaxed Consensus+Innovation solution approach.
Results
The RCI implementation achieves decentralized MBED clearing with a 0.03% relative cumulative optimality gap, 4.2% maximum relative gap, and 298 average iterations.
Takeaways & Limitations
Product differentiation can support more proactive consumer behavior, including favoring local or clean generation, within a peer-to-peer market framework.
Takeaways & Limitations
The proposed forward market does not address renewable-driven variability and uncertainty, which remain for reserve and balancing mechanisms whose adaptation to peer-to-peer markets is unclear.
Abstract
from arXiv · showhide
With the sustained deployment of distributed generation capacities and the more proactive role of consumers, power systems and their operation are drifting away from a conventional top-down hierarchical structure. Electricity market structures, however, have not yet embraced that evolution. Respecting the high-dimensional, distributed and dynamic nature of modern power systems would translate to designing peer-to-peer markets or, at least, to using such an underlying decentralized structure to enable a bottom-up approach to future electricity markets. A peer-to-peer market structure based on a Multi-Bilateral Economic Dispatch (MBED) formulation is introduced, allowing for multi-bilateral trading with product differentiation, for instance based on consumer preferences. A Relaxed Consensus+Innovation (RCI) approach is described to solve the MBED in fully decentralized manner. A set of realistic case studies and their analysis allow us showing that such peer-to-peer market structures can effectively yield market outcomes that are different from centralized market structures and optimal in terms of respecting consumers preferences while maximizing social welfare. Additionally, the RCI solving approach allows for a fully decentralized market clearing which converges with a negligible optimality gap, with a limited amount of information being shared.
I. INTRODUCTION
The paper proposes a peer-to-peer electricity market that lets agents negotiate bilaterally over differentiated products and prices without a supervisory agent. It uses MBED and decentralized consensus-based optimization to support consumer-centric participation.
- Consumers, prosumers, and other power-system actors are given a more proactive role in a consumer-centric market design.
- The proposed structure allows agents to express preferences through product differentiation and corresponding price differentiation.
- The MBED framework is solved through consensus-based optimization to enable fully decentralized market clearing.
- Peer-to-peer markets replace supervisory clearing with direct multi-bilateral negotiation among agents over energy and price.
- The market communication graph can connect agents directly, while “neighbors” need not be physically colocated.
- The formulation simplifies derivations by restricting agents to producers or consumers, while prosumers can be represented by separate buyer and seller entities.
B. Product Differentiation
Product differentiation embeds valuations of non-energy attributes into bilateral trading costs, allowing preferences, network charges, taxes, or social criteria to influence dispatch and prices.
- Agent costs combine production cost or consumer willingness to pay with bilateral trading costs.
- Production costs and consumer utilities are modeled as quadratic functions of each agent’s power set-point.
- Convex cost and utility functions with bijective gradients can extend beyond the paper’s quadratic modeling choice.
- Bilateral trading costs are linear in the quantity traded with each neighboring agent.
- Trade characteristics can encode criteria such as emissions, distance, ratings, or services through criterion-specific parameters.
- The framework can represent consumer preferences, dynamic taxes, regulatory incentives, and differentiated network charges through bilateral coefficients.
C. MBED Formulation
MBED maximizes community social welfare subject to agent limits, sign restrictions, and bilateral reciprocity, producing differentiated trade prices through reciprocity-constraint shadow prices.
- MBED maximizes community social welfare under power-injection limits, sign constraints, and bilateral reciprocity constraints.
- Shadow prices λ_nm of reciprocity constraints define prices for individual trades, allowing prices to differ across transactions.
- The convex MBED problem has a single optimum obtainable through centralized or distributed optimization methods.
- The presented model is single-time-step and omits network constraints, reliability, and other operational aspects, though extensions are discussed.
- The formulation is separable across market participants because agents are coupled by reciprocity constraints while quadratic objective terms remain independent.
- Strong duality implies that converged local solutions can achieve the maximum-social-welfare solution without a structural gap versus centralized optimization.
III. THE RELAXED CONSENSUS + INNOVATION SOLUTION APPROACH
RCI is a decentralized optimization approach inspired by Consensus + Innovation for solving the MBED problem through iterative local updates.
- RCI is a decentralized optimization method inspired by the Consensus + Innovation method for pool-based markets.
A. Structure of the Method
RCI decomposes the global MBED optimization into local agent problems and updates their variables through consensus-informed gradient steps. The method combines dual-variable updates, projected primal updates, and a flow-charted three-step iteration.
- Core decomposition: RCI splits the global optimization problem into local agent problems whose equilibria satisfy feasibility and the global KKT conditions.Each participant solves a local problem while reacting to other agents’ primal and dual estimates.
- Primal update: The local problem is updated with a gradient step and feasibility projection rather than solved directly.This avoids costly local optimization and mitigates binary, oscillating trade outcomes caused by choosing only the most attractive price.
- Constraint handling: Power boundary constraints use Lagrangian relaxation, while sign constraints are enforced through closest-point projection.The associated boundary dual variables are estimated using complementary slackness.
- Iteration structure: Each RCI iteration updates trading-reciprocity dual variables, power-boundary dual variables, and decision variables in three successive steps.The algorithm’s overall structure is summarized in Figure 2.
B. Dual Updates
RCI updates decentralized price and boundary dual estimates while adjusting primal trade variables toward local optimality. Consensus and innovation terms align bilateral price estimates and enforce reciprocity, yielding uniform perceived prices across effective trades at convergence.
- Trading-price update: The λ-update combines a consensus term for bilateral price agreement with an innovation term enforcing the trading equality constraint.Each participant calculates its own estimate, with convergence requiring λ_nm = λ_mn.
- Parameter tuning: Positive tuning sequences α^k and β^k must provide persistent excitation, and their ratio trades off convergence speed against resilience to setup changes.Adaptive factors are suggested as a possible performance improvement.
- Boundary update: The μ-update adjusts dual variables for power-boundary constraints while accounting for complementary slackness.It uses a persistent positive tuning sequence η^k.
- Primal update: The P-update uses KKT conditions of the relaxed local problem to update each agent’s decision variables.Sign constraints are enforced through a max operator for producers and a min operator for consumers.
- Convergence interpretation: At convergence, perceived prices are uniform across an agent’s effective trades and equate marginal production cost or consumption utility.The averaged optimality condition becomes equivalent to the optimality condition of effective trades.
D. Operational Aspects of the Iterative Process
The decentralized implementation keeps computation local and shares only limited information during iterations. Its main operational boundaries are convergence-monitoring uncertainty and scalability costs that grow with communication connectivity.
- Information exchange: All RCI updates are performed locally by each agent, with only limited information exchanged among agents engaged in multi-bilateral negotiations.This supports privacy and data security because internal data and computation remain local.
- Privacy: Agents need not share internal production, consumption, or criterion parameters to reach optimality.The implementation therefore achieves decentralized clearing without exposing those parameters.
- Stopping criteria: The iterative process terminates after primal, dual, and price-related convergence conditions are satisfied.Three positive tolerances, ϵ_λ, ϵ_P, and ϵ_μ, define the stopping conditions.
- Convergence monitoring: Dual-only convergence monitoring is commonly used, but whether primal and dual convergence are linked here requires further study.Criterion (26) is optional and provides more precise convergence monitoring.
- Scalability: With full peer-to-peer communication, each agent faces a number of decision variables of order N^2, making scalability an open issue.Future work is expected to consider sparsifying the communication matrix.
IV. APPLICATION AND CASE STUDIES
The RCI approach is evaluated in a simplified but realistic proof-of-concept setup. The evaluation includes a single-time-step illustration and simulations over a full year.
- Evaluation design: The evaluation uses a simplified though realistic setup as a proof of concept.The study first describes the system setup before presenting the algorithm’s operation and product-differentiation effects.
- Case studies: A single-time-step case illustrates RCI operation and product differentiation, followed by full-year simulations.The single-step analysis provides intuition before the market evaluation is extended temporally.
A. Simulation Setup
The simulation uses a 12-agent, two-bus setup with differentiated bilateral trading and yearly hourly operation. Results examine RCI convergence and how the criterion value changes inter-bus exchanges.
- Simulation setup: The setup contains 12 agents: six producers and six consumers, including renewable, fossil, household, and industrial participants.Wind and solar production sequences, consumption sequences, and flexibility capacities are drawn from referenced datasets.
- Network and product differentiation: Agents are split across two buses, with three generators and three loads on each bus, while bilateral costs use inter-agent distance.Opposite-bus trades are assigned a fixed 1 km distance so their marginal trading cost is identical.
- Simulation protocol: The system is simulated for one year at an hourly time-step, using persistence-based warm starts to improve performance.The expected persistence error is high because renewable penetration and variability are high.
- RCI behavior: Producer 3’s negotiations show uniform perceived prices on effective trades while other trades satisfy optimality conditions.The comparison includes time step 1 without a warm start and time step 8 with persistence-based initialization.
- Product-differentiation experiment: Without product differentiation, inter-bus exchanges are high; increasing the criterion value shifts trades toward intra-bus exchanges until the buses become autonomous or flexibility is exhausted.Beyond cn ∈[0; 2], further increases reduce intra-bus trades as agents lower production or consumption.
C. Convergence Analysis
The RCI method approaches the centralized MBED optimum with improving objective accuracy and consensus feasibility, while product differentiation reduces line use at a bounded direct-cost increase. Convergence can become slower near zero differentiation and scaling remains challenging.
- RCI convergence: The RCI method is evaluated against a centralized quadratic-program solution using the optimality gap rather than decision-variable output.The authors argue that gradient-based solving makes objective-function comparison more meaningful.
- RCI convergence: The objective-function accuracy improves logarithmically, first approximating the solution and then refining its accuracy.The first phase covers approximately the first 600 iterations in the reported case.
- RCI convergence: 0.03% relative cumulative optimality gap and 4.2% maximum relative optimality gap are reported, with 298 average iterations to convergence.The average time to reach the stopping criteria is 0.1 s in the stated MATLAB and hardware setup.
- Product differentiation: A criterion value of 1 ce.kWh^-1.km^-1 reduces annual energy flows by more than 95% and maximum line power by more than 40%.The comparison is against a case without product differentiation, equivalent to a pool-based model.
- Product differentiation: Almost 50% energy-flow reduction requires less than 0.01% direct-cost increase, while over 90% energy-flow reduction requires less than 2%.The latter condition also corresponds to a 30% reduction of the power peak in the reported setup.
- Convergence limitations: As the criterion value approaches zero, iterations increase rapidly because the MBED has multiple solutions when bilateral trading costs are not valued.The authors suggest adapting the model toward a hybrid multi-bilateral and pool-based dispatch in that case and call for broader sensitivity analysis.
V. CONCLUSIONS AND PERSPECTIVES
The paper proposes a peer-to-peer electricity market based on multi-bilateral trading and product differentiation, solved through decentralized RCI optimization. The framework achieves acceptable optimality gaps with limited information exchange, while its forward-market scope and scaling issues motivate further work.
- Conclusions: The proposed market combines multi-bilateral trading and product differentiation in a peer-to-peer structure for increasingly decentralized power systems.The framework is motivated by increasing consumer and prosumer flexibility.
- Conclusions: RCI implements the MBED without a central agent and with limited information exchange, achieving acceptable optimality gaps.The conclusion describes the implementation as distributed and efficient.
- Perspectives: Product differentiation meaningfully affects power exchanges and can support more proactive behavior such as favoring local or clean generation.Future work includes heterogeneous criterion values, strategic behavior, and free riders.
- Perspectives: Future RCI work will address more constrained economic dispatch models and extend MBED to congestion, reactive-power compensation, and losses.The paper also discusses directly attributing trade-induced costs rather than socializing them as in many pool-based markets.
- Scope boundary: The proposed setup is a forward mechanism that does not address renewable-driven variability, demand uncertainty, reliability, reserves, or real-time balancing.The authors identify adaptation of short-term, balancing, and risk-sharing mechanisms as future work.
APPENDIX
The appendix establishes strong duality for the MBED by showing convexity and feasibility, using the corresponding pool-based model to justify feasibility.
- Implementation: Wind, solar, and household-consumption series provide the generation and demand data, with wind turbines scaled to 100 kW and solar installations to 50 kW.Household series include upward and downward flexibility; the supplied passage states that flexibility costs and additional parameters are specified elsewhere.
- Proof of Strong Duality: Because the MBED is convex and feasible, Slater’s condition holds and strong duality follows.This is the appendix’s stated clarification of the strong-duality claim made in Section II-C.
- Proof of Strong Duality: The MBED is convex because its cost functions have positive quadratic coefficients and its constraints are linear.Convexity is a prerequisite for the strong-duality argument developed in the appendix.
- Proof of Strong Duality: MBED feasibility follows from feasibility of the corresponding pool-based model after replacing local balance constraints with a global balance constraint.A feasible pool-based solution is mapped to a feasible MBED solution through bilateral variables.