Optimal Solution Analysis and Decentralized Mechanisms for Peer-to-Peer Energy Markets

• ICT, P2P, DER, and ADMM denote information and communication technology, peer to peer, distributed energy resource, and alternating direction method of multipliers.Source: Nomenclature · p. 1
• P_ij, P_i, and P_i,tr denote peer-to-peer traded power, peer i’s vector of traded powers, and peer i’s total traded power; its lower and upper bounds are also named.Source: Nomenclature · p. 1
• G, A, D, and L denote the P2P graph and its adjacency, degree, and Laplacian matrices; 1_n and I_n denote the all-ones vector and identity matrix.The notation also specifies diagonal or block-diagonal matrices, vectors, and real matrices with dimensions n × m.Source: Nomenclature · p. 1

1 Introduction

• P2P systems can reduce energy losses through short-distance exchange and offer localized, decentralized trading that gives prosumers flexibility in managing energy balance and profit.The introduction also cites potential security and privacy benefits from ICT and opportunities for different market models.Source: §1 · p. 2
• Prosumer roles and preferences vary over time, and direct bilateral trading imposes a power-balance constraint on each communicating pair rather than one system-wide balance constraint.The cited passage notes that prosumers’ generation and consumption profiles are time-varying and predicted in advance.Source: §1 · p. 2
• Prior ADMM approaches can require sequential local-variable updates, a platform agent, or global energy-transfer information, while decentralized approaches are motivated by avoiding a central point of failure.The introduction also notes that existing ADMM approaches may require a central coordinator or sequential updates despite parallel prosumer work.Source: §1 · pp. 2–3
• Existing P2P studies commonly assume all peers trade successfully and choose suitable cost parameters, although prosumers may not know the parameters needed to obtain their expected traded power and price.Source: §1 · p. 3
• The paper proposes decentralized ADMM with parallel prosumer updates and decentralized learning to tune cost-function parameters for successful trading at expected total traded-power amounts.Source: §1 · p. 3

2 P2P Electricity Trading Problem

• Each prosumer is a nonstrategic agent that can produce and consume, and is assumed to buy or sell—but not both—at each time step.Source: §2 · p. 3

2.1 Inter-peer Communication Structure

• The communication graph is undirected and bipartite, separating buying and selling prosumers so edges connect only across the two groups.Source: §2.1 · p. 4

2.2 Objective Function

• Each prosumer’s cost combines utility, weighted bilateral trading costs, and a network implementation cost; the assumed cost function is convex.Trade weights can distinguish products, while the implementation cost is paid to the network operator.Source: §2.2 · p. 4

2.3 System Constraints

• Bilateral trades must balance between each pair, while physical power-flow constraints are omitted and represented through the network implementation cost.Source: §2.3 · p. 4

2.4 Overall Optimization Problem

• The optimization is convex and separable by time step; the paper compares its P2P solution with pool-based social-welfare maximization.The study solves one time step at a time, while noting that real-time optimization may be impractical at scale or with communication latencies.Source: §2.4 · p. 5

2.5 Characterization of Optimal Solution

• With no trade weights and a connected graph of successfully traded peers, P2P prices and total traded powers match the pool market; disconnected graphs split into smaller markets with different prices.Source: §2.5 · pp. 5–6
• The classical Lagrangian characterization gives optimal prices and total traded powers, but not the individual power in each transaction.The paper next proposes decentralized methods to derive bilateral powers and prices and tune cost parameters for desired successful trades.Source: §2.5 · p. 6

3.1 Decentralized P2P Negotiation Mechanism

• The decentralized ADMM approach updates P and X in parallel, avoiding the sequential updates required by cited existing P2P ADMM methods.Local variables at each prosumer are also updated in parallel, while each peer solves its own optimization problem; central communication and computation burden is avoided, and privacy can be better guaranteed.Source: §3.1 · pp. 6–7

3.1.1 The Update for Variable X

• The X update is a convex quadratic problem that decomposes by prosumer, with each Xi constrained by the peer’s buyer or seller role.Source: §3.1.1 · p. 7

3.1.2 The Update for Variable P

• The P update is a convex optimization with quadratic cost and bilateral constraints Pij + Pji = 0, so strong duality holds and KKT conditions apply.The paper derives iterative updates for individual traded power and transaction price.Source: §3.1.2 · pp. 7–8
• As k →∞ and ρ →0, optimal total traded power and optimal energy price converge to those stated in Theorem 1.Source: §3.1.2 · p. 8

3.2.1 Learn For Successful Trading

• After failed trades, prosumers can adjust cost parameters to obtain successful transactions: buyers decrease bi and sellers increase bi while keeping ai unchanged.These adjustments change marginal costs and gradually match expected profits across prosumers.Source: §3.2.1 · pp. 8–9

3.2.2 Learn For Increasing Traded Power Amount

• To increase traded power, the learning method decreases ai rather than increasing bi, because increasing bi raises the trading price while decreasing ai need not.Source: §3.2.2 · p. 9

4.1 Synthetic System

• The synthetic study uses six prosumers trading hourly and examines six scenarios, including communication changes, role switching, trade weights, and decentralized learning.All variables are initialized at zero; dashed lines represent transaction prices and solid lines represent traded powers.Source: §4.1 · pp. 9–10

4.1.1 Scenario 1 (Pool-based market)

• Scenario 1 Figure 3 results use the algorithm in [20] with ρ = 0.02, φ = 0.021, ψ = 0.021, and κ = 0.99 for the problem without P2P energy trading.In Scenario 2, the total traded power of each peer and energy price are the same as in Scenario 1.Source: §4.1.1 · p. 10, §4.1.2 · p. 10

4.1.2 Scenario 2 (P2P market with buyer-seller complete communication)

• With complete buyer-seller communication, successful trades have identical prices and total peer powers to the pool-based market, despite peers 2 and 5 not trading.The communication graph among successfully traded peers is connected.Source: §4.1.2 · p. 10, Figure 4 · p. 11

4.1.3 Scenario 3 (P2P market with buyer-seller incomplete communication)

• Removing the link between peers 1 and 6 disconnects successful trades, forming two smaller P2P markets with different prices.Peer 1 sells only to peer 4, while peer 3 sells only to peer 6; peers 2 and 5 do not trade.Source: §4.1.3 · p. 10, Figure 5 · p. 11

4.1.4 Scenario 4 (P2P market with role changing of peers)

• When peer 2 switches from seller to buyer, it buys from peers 1 and 3, as do peers 4 and 6; prices and powers differ from scenarios 2 and 3.Peer 5 again does not trade.Source: §4.1.4 · p. 11, Figure 6 · p. 12

4.1.5 Scenario 5 (P2P market with effects of bilateral trade weights)

• Bilateral trade weights substantially change outcomes: peer 6 buys 5.1 kW from fossil-based peer 1 versus 51.2 kW in Scenario 2, and 90.1 kW from renewable peer 3.Prices also differ from Scenario 2 despite complete buyer-seller communication; the weights encode buyer preferences for generation sources.Source: §4.1.5 · p. 12

4.1.6 Scenario 6 (Decentralized learning for successful trading)

• After increasing b2 and decreasing b5 while keeping a2 and a5 unchanged, all peers trade successfully at b2 = 7.53 and b5 = 4.53.The learning strategy is applied to peers 2 and 5, which had failed to trade in earlier scenarios.Source: §4.1.6 · p. 13
• In Scenario 6, compared with scenarios 1 and 2, prosumers 4 and 6 pay less, prosumer 1 suffers a slight loss, and prosumer 3 gains considerably; decentralized learning for prosumers 2 and 5 benefits many, but not all, prosumers.In Scenario 5, trade weights change traded power amounts and are associated with higher payments for prosumers 4 and 6 and more profit for prosumers 1 and 3; choosing the best bilateral weights needs further investigation.Source: §4.1.6 · p. 13

4.2 Modified IEEE European Low Voltage Test Feeder

• The feeder case has 55 nodes, with 25 nodes assigned 5.5 kW rooftop solar and the remaining nodes assigned 3 kWh batteries; trading is hourly.The one-hour load profiles come from, and July solar irradiance data for Spain come from.Source: §4.2 · p. 14, Figure 10 · p. 14
• At noon, the 25 solar-equipped nodes have maximum demands of 3.5 kW and at least 2 kW of redundant solar power to sell, yielding 25 sellers and 30 buyers.The inter-prosumer communication graph has 750 edges.Source: §4.2 · p. 14
• When all prosumers’ parameters ai are randomly regenerated in [0.002, 0.006], absolute total traded power significantly increases, while optimal prices are not much different between the two cases.Similar simulation results are obtained at other time steps.Source: §4.2 · p. 16

4.3 Scalability of The Proposed Approach

• For the two IEEE European Low Voltage Test Feeder cases, proposed decentralized ADMM runtimes are 14.2 s and 21.2 s, just double the synthetic-system runtime; maximum communicated peers (trades) per prosumer is 30.Source: §4.3 · p. 16
• For systems with 75 sellers and 90 buyers, runtime is 117.6 s; with 150 sellers and 180 buyers, it is 1022.9 s, and computation time is reported not to increase exponentially with system size.The authors characterize the approach as scaling well.Source: §4.3 · p. 16, Figure 13 · p. 16
• The reported computational times omit inter-peer communication time and possible delays; communication delay is outside this study’s scope.The authors identify it as future work.Source: §4.3 · p. 16

5 Conclusion

• The proposed decentralized ADMM approach and prosumer learning strategies are based on analytical trading-price and power formulas, with cost-function tuning aimed at successful transactions and desired total traded powers.The formulas relate pairwise trades to each prosumer’s total traded amount, trade weights, and cost-function parameters.Source: §5 · p. 16
• Case studies illustrate the effectiveness and scalability of the proposed approaches.Source: §5 · p. 17
• Future work should investigate more complex participant models, power-flow constraints, market interactions, communication delays, and systematic trade-weight selection.Source: §5 · p. 17

5.1 Proof of Theorem 1

• Without trade weights, a connected graph of successfully traded peers yields a common clearing price and the same total traded energy per peer as the pool-based market; disconnected graphs split into clusters inheriting these properties.Source: §5.1 · p. 17

5.2 Proof of Theorem 2

• The ADMM convergence proof invokes the result in and substitutes the converged solutions into the update relation to obtain the corresponding prices.Source: §5.2 · p. 17