PaperlayerView exact revision on arXiv

Source-linked AI summary

EVTradeMatch: A Mobility-Aware Multi-Objective Matching Framework for EV--EV Energy Trading

Md. Mahfujur Rahman, Alistair Barros, Raja Jurdak, Darshika Koggalahewa

arXiv:2609.10551v1math.OCcs.LGeess.SY

TL;DR

Limited charging infrastructure makes journey-aware coordination between EV energy providers and consumers difficult. EVTradeMatch combines prediction-derived charging-node suitability with multi-objective MILP matching and NSGA-II search. It improves matching outcomes across coverage, transferred energy, and suitability while exposing mobility cost as an explicit trade-off and offering alternative Pareto solutions.

  • Problem

    EV–EV trading requires coordinated provider–consumer matching under mobility, temporal, energy, and charging-location constraints that existing prediction and matching approaches do not jointly integrate.

  • Method

    EVTradeMatch uses EVNextTrade suitability predictions in a multi-objective MILP and tailored NSGA-II to select feasible matches and charging nodes under operational constraints.

  • Results

    NSGA-II simultaneously improves matching coverage, transferred energy, and charging-node suitability relative to comparison methods, while accepting higher mobility cost as an explicit trade-off.

  • Takeaways & Limitations

    Pareto-front and hypervolume results support flexible selection among competing operational priorities rather than reliance on a single matching objective.

  • Takeaways & Limitations

    The evaluation uses a single-day urban dataset, and repeated decision events are treated independently rather than as longitudinal battery-state updates.

Abstract

from arXiv · show

Peer-to-peer energy trading among electric vehicles (EVs) can improve charging flexibility under limited charging infrastructure, but effective EV--EV trading requires coordinated provider--consumer matching under journey-specific conditions. This paper proposes EVTradeMatch, a prediction-guided multi-objective optimization framework for mobility-aware EV--EV energy trading. Building on the EVNextTrade study, a prior learning-to-rank model for charging-node recommendation, we define charging-node suitability as a prediction-derived score reflecting the appropriateness of assigning a provider--consumer pair to a candidate charging node based on mobility, energy, and contextual trading features. This score is used as a guidance signal and as an explicit optimization objective rather than as a hard selection rule. The EV--EV matching problem is formulated as a multi-objective mixed-integer linear program that maximizes matching coverage, transferred energy, and charging-node suitability while minimizing mobility cost under spatial, temporal, one-to-one matching, and charging-node capacity constraints. To approximate Pareto-efficient solutions in wide-area dynamic settings, we develop a tailored non-dominated sorting genetic algorithm II (NSGA-II). Experimental results show that EVTradeMatch improves transferred energy by 74.2--82.8% and charging-node suitability by 8.3--84.4% compared with proximity- and auction-based state-of-the-art methods, while improving matching coverage by 3.74--25.07 percentage points. Balanced NSGA-II solutions achieve 53.07$\pm$0.76% matching coverage and transfer 1701.94$\pm$17.03 kWh, with higher mobility cost as an explicit trade-off against travel-minimizing methods. Pareto-front analysis shows that the framework supports flexible selection among high-coverage, high-energy, low-mobility-cost, and high-suitability solutions according to operational priorities.

I. INTRODUCTION

EV–EV trading must coordinate mobile providers and consumers under spatial, temporal, energy, and charging-location constraints. EVTradeMatch addresses this gap with prediction-guided multi-objective matching that combines MILP formulation and tailored NSGA-II search.

  • Motivation and research gap: Existing EV–EV matching methods emphasize proximity, market clearing, or heuristics but do not jointly optimize mobility feasibility, temporal coordination, transferred energy, and charging-location suitability.The literature includes Gale–Shapley, BMNN, DAMA, and other optimization approaches, but the stated gap concerns their limited integration of journey-specific constraints and competing operational objectives.
  • Framework overview: EVTradeMatch converts EVNextTrade’s prediction-derived charging-node scores into network-level provider–consumer trading decisions.The framework jointly selects consumer and provider decision events, charging nodes, and time slots rather than only recommending charging nodes.
  • Optimization method: The matching problem is formulated as a MILP, while tailored NSGA-II approximates Pareto-efficient trade-offs when conflicting objectives and combinatorial growth limit exact optimization.The optimization represents discrete matching and trading-location assignments under linear feasibility constraints.
  • Framework overview: The framework enforces spatial reachability, temporal compatibility, one-to-one matching, and charging-node capacity while prioritizing promising opportunities through predicted suitability.Suitability scores are used as guidance and as an explicit optimization objective, not as a hard selection rule.
  • Contributions and evaluation: NSGA-II improves transferred energy and charging-node suitability while maintaining comparable matching coverage against heuristic, learning-based, and state-of-the-art methods.The comparison includes random matching, greedy optimization, PPO, BMNN, and DAMA, with mobility cost treated as an explicit trade-off rather than the sole objective.
  • Scope: The framework focuses on operational provider–consumer coordination rather than monetary cost, pricing, social welfare, or dynamic state-of-charge evolution.These factors are identified as outside the present scope and potential directions for future extensions.

III. STUDY DESIGN

EVTradeMatch matches consumer and provider EV decision events at wireless charging pads under mobility, energy, spatial, temporal, and infrastructure constraints. The study uses prediction-derived charging-node suitability within a multi-objective mobility-aware matching framework.

  • Study Design: EVTradeMatch balances matched pairs, transferred energy, mobility cost, and charging-node suitability when matching EV decision events at wireless charging pads.The research question concerns provider–consumer matching under multiple conflicting objectives and operational constraints.
  • Study Design: Feasible matches require both EVs to reach the selected charging pad while satisfying journey energy requirements and spatial-temporal feasibility conditions.Travel distances, arrival times, waiting tolerances, traffic-dependent travel times, and candidate WCP sets define feasibility.
  • Study Design: The framework considers consumer and provider decision events, wireless charging pads, and discrete time slots across feasible candidate matches.Decision events are role-specific requests associated with particular journeys, locations, energy states, and candidate charging-node contexts.
  • Study Design: Prediction-derived scores from EVNextTrade provide fixed charging-node suitability inputs alongside spatial-reachability and temporal-alignment checks.The scores represent candidate-node relevance for consumer and provider decision events and are incorporated into matching optimization.
  • Study Design: The formulation uses discrete slots, infrastructure capacity, and binary feasibility indicators to represent candidate charging sessions.The study defines slot membership, WCP capacity, reachability, temporal matching, and suitability-related notation for the optimization model.

C. Decision Variable

The binary decision variable records whether a consumer decision event and provider decision event are matched at a specific wireless charging pad and time slot.

  • C. Decision Variable: X_ijkt equals 1 when consumer event i matches provider event j at WCP k during time slot t, and 0 otherwise.This variable encodes the selected provider–consumer–location–time assignment.

D. Constraints

The constraints enforce one-to-one participation, WCP capacity, spatial reachability, and temporal compatibility for selected EV–EV charging sessions.

  • D. Constraints: Each consumer and provider decision event can participate in at most one selected charging session within the planning horizon.This prevents repeated assignment across providers, locations, or time slots.
  • D. Constraints: Each WCP’s charging-session count cannot exceed its capacity in any time slot.Capacity is enforced per wireless charging pad and slot.
  • D. Constraints: Selected consumer and provider events must remain within their respective maximum allowable detour distances to the chosen WCP.Spatial feasibility is represented through reachability indicators and distance limits.
  • D. Constraints: Consumer and provider arrivals must be sufficiently close under waiting tolerances and fall within the same discrete charging slot.Arrival times use route travel times under time-dependent traffic conditions, with half-open slots preventing boundary overlap or gaps.
  • D. Constraints: Binary temporal-feasibility and decision-variable constraints restrict assignments to feasible provider–consumer–WCP–slot combinations.The temporal indicator and binary variable conditions are enforced for all indexed events, pads, and slots.

E. Model Assumptions

The model treats prediction-derived suitability as an explicit objective while representing coverage, energy transfer, mobility cost, and feasibility through a constrained multi-objective formulation. NSGA-II then searches the combinatorial solution space using prediction-guided construction, mutation, and repair.

  • E. Model Assumptions: The formulation assumes decision-event-level matching over a fixed horizon, with physical EV events treated as distinct role-specific requests.The study also uses one-hour optimization segments and assumes wireless charging pads source energy from provider EVs.
  • E. Model Assumptions: The MILP maximizes matching coverage, transferred energy, and charging-node suitability while minimizing mobility cost.Coverage counts selected trading sessions, while mobility cost combines normalized travel distance and waiting-time mismatch.
  • E. Model Assumptions: Suitability scores are transformed into bounded probabilities and averaged across consumer and provider sides for pairwise WCP suitability.The arithmetic mean gives symmetric weight to both participants and reflects the requirement that the selected WCP suit both sides.
  • E. Model Assumptions: The optimization jointly selects feasible pairs, WCPs, and time slots while enforcing uniqueness, capacity, spatial, temporal, and binary constraints.These decisions form a combinatorial search space whose tuples grow with EV events and candidate charging nodes.
  • E. Model Assumptions: NSGA-II uses feasible WCP sets and higher predicted suitability to guide initialization and mutation, then repairs violations of uniqueness and WCP–slot capacity.The heuristic approximates solutions when exact multi-objective MILP solving is impractical for large dynamic scenarios.

1) NSGA-II Approach for Mobility-Aware EV–EV Matching:

The tailored NSGA-II searches a precomputed feasible session universe and returns Pareto-efficient EV–EV matching solutions across four competing objectives.

  • NSGA-II Approach for Mobility-Aware EV–EV Matching:: The feasible session universe restricts evolution to realizable sessions satisfying spatial and temporal feasibility before optimization begins.Each tuple (i, j, k, t) denotes a realizable consumer–provider charging session at a WCP and time slot.
  • NSGA-II Approach for Mobility-Aware EV–EV Matching:: The search begins with feasible provider–WCP–time assignments, repairs initial individuals, and evolves them through crossover, mutation, evaluation, and non-dominated selection.Repair preserves consumer, provider, and WCP-capacity feasibility during population construction and variation.
  • NSGA-II Approach for Mobility-Aware EV–EV Matching:: Fitness evaluation computes four objectives, while non-dominated rank and crowding distance guide selection and preserve solution diversity.Dominance favors higher coverage, energy, and suitability, but lower mobility cost.
  • NSGA-II Approach for Mobility-Aware EV–EV Matching:: The algorithm returns a Pareto front balancing matching coverage, transferred energy, mobility cost, and charging-node suitability.Each returned solution represents a different operational trade-off rather than a single maximum.
  • NSGA-II Approach for Mobility-Aware EV–EV Matching:: A balanced solution is selected from the final Pareto front using normalized objectives, prioritizing coverage and transferred energy before mobility cost and suitability.This post-processing step supports direct comparison with single-solution baselines.

2) Feasible Session Selection:

Feasible Session Selection constructs consumer-centered candidate sessions by enforcing reachability, temporal compatibility, exclusivity, and WCP–slot capacity before scoring alternatives.

  • Feasible Session Selection:: Consumer-centric construction considers each consumer once, while provider reachability, provider exclusivity, temporal feasibility, and capacity checks preserve the original matching structure.Remaining conflicts can be handled later by feasibility repair.
  • Feasible Session Selection:: For each consumer, the procedure enumerates provider–WCP–time combinations satisfying reachability, temporal feasibility, one-to-one matching, and capacity availability.Only sessions passing all session-level checks enter the feasible candidate set.
  • Feasible Session Selection:: Candidate scoring can prioritize prediction-derived suitability, transferable energy, or a hybrid utility combining both signals.The hybrid mode uses 0.4q_ijk + 0.6 Ē_ijk/Ē_max, with tie noise added to utilities.
  • Feasible Session Selection:: The highest-utility feasible session is selected for the consumer, while an empty candidate set leaves that consumer unmatched.The selected session specifies the provider, WCP, and time slot.

3) Feasibility Repair:

Feasibility Repair converts candidate session sets into feasible matchings by removing invalid assignments and resolving consumer, provider, and WCP–slot conflicts using a deterministic score.

  • Feasibility Repair:: The repair score rewards suitability and transferable energy while penalizing travel distance and arrival-time mismatch.Weights λ and µ control the relative importance of the two mobility penalties.
  • Feasibility Repair:: Repair removes duplicate or out-of-universe sessions before enforcing one-to-one and WCP–slot capacity constraints.The feasible session universe contains only spatially and temporally feasible assignments.
  • Feasibility Repair:: When a consumer or provider has multiple assignments, repair retains the highest-scoring session and removes the others.This ensures each consumer and provider decision event appears in at most one selected session.
  • Feasibility Repair:: For overloaded WCP–slot pairs, repair retains the highest-scoring sessions up to the available capacity.The retained set keeps usage within capacity while preserving stronger sessions according to the repair score.

4) Constraint-Aware Fitness Evaluation:

Constraint-Aware Fitness Evaluation scores each candidate matching on four objectives and aggregates residual constraint violations for NSGA-II ranking and selection.

  • Constraint-Aware Fitness Evaluation:: Each solution is evaluated using matching coverage, transferred energy, mobility cost, prediction-derived suitability, and an aggregate violation score.The resulting fitness tuple is used for non-dominated sorting, diversity preservation, and selection.
  • Constraint-Aware Fitness Evaluation:: Matching coverage counts selected sessions, transferred energy accumulates energy caps, mobility cost combines normalized distance and arrival-time mismatch, and suitability sums prediction scores.These quantities are computed over every selected session in the candidate solution.
  • Constraint-Aware Fitness Evaluation:: Constraint violations quantify duplicate sessions, spatial and temporal infeasibility, consumer and provider over-assignment, and WCP–slot capacity excess.The aggregate is Viol(x) = vC1 + vC2 + vC3 + vC4 + vC5 + vC6.
  • Constraint-Aware Fitness Evaluation:: Solutions are repaired before evaluation, while the violation score remains as a diagnostic safeguard for residual infeasibility.The implementation expects evaluated individuals to be feasible after repair.

H. Baseline Methods

The baselines use the same feasible matching set and differ only in how they select matches.

  • All baselines operate on an identical feasible matching set induced by the study’s constraints.
  • Performance differences therefore reflect match-selection strategies rather than different feasibility spaces.
  • The shared feasible set provides a common basis for comparing baseline algorithms.

1) Random Matching:

Random Matching constructs feasible solutions by sequentially sampling tuples while preserving uniqueness and capacity constraints.

  • Random Matching samples feasible tuples uniformly and adds them sequentially until no further feasible tuple can be added.
  • Each rollout preserves consumer-event uniqueness, provider-event uniqueness, and WCP–slot capacity feasibility.
  • Thirty independent rollouts are run per evaluation seed, retaining the best feasible solution.

2) Greedy Matching:

The comparison evaluates greedy, reinforcement-learning, auction, and proximity-based matching methods under common objectives, constraints, data, and feasible opportunities.

  • 2) Greedy Matching:: Greedy Matching iteratively selects the feasible tuple maximizing a scalar utility derived from the MILP objectives.
  • 2) Greedy Matching:: The utility combines transferable energy, normalized mobility cost, and predicted suitability through nonnegative weights.
  • 2) Greedy Matching:: After each selection, conflicting tuples are removed until no feasible tuples remain.
  • 2) Greedy Matching:: PPO sequentially constructs feasible matchings with an actor–critic policy and a scalarized reward combining four objectives.
  • 2) Greedy Matching:: DAMA clears bids and asks independently within each wireless charging pad and time slot, whereas BMNN uses mutual nearest-neighbor selection.
  • 2) Greedy Matching:: The experiments use a Chicago-area dataset with 431 consumer and 388 provider decision events across 128 charging stations.

B. Structure of the Feasible Opportunity Space

The feasible opportunity space is uneven, motivating NSGA-II to coordinate matching across competing objectives and trade-offs. NSGA-II achieves the strongest overall balance in coverage, transferred energy, and suitability, while mobility cost remains an explicit trade-off.

  • The feasible opportunity set is highly uneven, with fragile events having only one or two feasible sessions and flexible events having many alternatives.
  • NSGA-II achieves 79.60 ± 1.14 matched sessions and 53.07 ± 0.76% matching coverage, the highest average coverage among the compared methods.
  • NSGA-II transfers 1701.94 ± 17.03 kWh and reaches a suitability score of 71.12 ± 0.79, exceeding PPO, Greedy, and Random on both objectives.
  • Greedy has the lowest mobility cost at 18.65 ± 0.62, while NSGA-II records 19.11 ± 0.79 and remains close despite stronger system-level performance.
  • NSGA-II’s hypervolume increases until stabilizing around generation 173, and its final hypervolume exceeds the fixed levels of the baseline methods.
  • The Pareto front exposes a coverage–mobility-cost trade-off and generally links more matched pairs with higher transferred energy.

1) Performance of the Proposed NSGA-II Against State-ofthe-Art Methods:

NSGA-II delivers stronger matching coverage, transferred energy, and charging-node suitability than BMNN and DAMA, while accepting higher mobility cost. Its Pareto front also provides configurable trade-offs across system objectives, and suitability guidance improves energy transfer without materially changing coverage.

  • 1) Performance of the Proposed NSGA-II Against State-ofthe-Art Methods:: 53.07 ± 0.76% matching coverage is achieved by NSGA-II, versus 49.33 ± 0.00% for BMNN and 28.00 ± 0.00% for DAMA.This corresponds to improvements of 3.74 and 25.07 percentage points over BMNN and DAMA, respectively.
  • 1) Performance of the Proposed NSGA-II Against State-ofthe-Art Methods:: 1701.94 ± 17.03 kWh transferred energy and 71.12 ± 0.79 suitability are obtained by NSGA-II, exceeding both BMNN and DAMA.Transferred energy improves by approximately 82.80% over BMNN and 74.21% over DAMA, while suitability exceeds their reported scores.
  • 1) Performance of the Proposed NSGA-II Against State-ofthe-Art Methods:: 19.11 ± 0.79 mobility cost is recorded by NSGA-II, compared with 15.76 ± 0.00 for BMNN and 11.63 ± 0.00 for DAMA.The higher cost is exchanged for stronger matching coverage, transferred energy, and charging-node suitability.
  • 2) Pareto-Front Convergence and Trade-off Analysis Against State-of-the-Art Methods:: NSGA-II converges toward a stable Pareto front and exposes trade-offs in which more matched pairs generally require higher mobility cost and correspond to higher transferred energy.The balanced solution in a representative run achieves 78 matched pairs, 52.00% matching rate, approximately 1713 kWh, and mobility cost approximately 20.13.
  • 2) Pareto-Front Convergence and Trade-off Analysis Against State-of-the-Art Methods:: Table VII spans objective-specific alternatives: best-f1 reaches 54.00% coverage, best-f2 transfers 1716.04 kWh, best-f3 lowers mobility cost to 7.4857, and best-f4 reaches suitability 72.7895.These solutions give decision-makers flexibility to select among conflicting system priorities.
  • E. Ablation Study of Prediction-Derived Suitability Guidance: Removing suitability guidance reduces transferred energy from 1701.94 to 1407.70 kWh while matching coverage changes only from 53.07% to 52.67%.Average transferred energy per match also falls from 21.39 to 17.82 kWh, indicating that guidance selects higher-quality trading sessions within the feasible opportunity set.

H. Sensitivity Analysis

Sensitivity analysis selects NSGA-II settings for balanced multi-objective performance, while broader results show feasible trade-offs among matching, energy, suitability, and mobility cost.

  • Hyperparameter selection: 250 generations were selected because they remained within 0.22 percentage points of the highest matching percentage while achieving the highest transferred energy and suitability.The longer run also produced substantially lower mobility cost than 50 generations.
  • Hyperparameter selection: A population size of 250 was selected because it stayed within 0.45 percentage points of the best matching value while improving mobility cost and transferred energy over size 200.This choice follows the study’s balanced multi-objective selection rule.
  • Hyperparameter selection: The final NSGA-II configuration used population size 250, 250 generations, crossover rate 0.2, and mutation rate 0.6.Crossover rate 0.2 achieved 79.67 matched pairs, a 53.11% matching rate, and suitability of 71.0978 in the sweep.
  • Results: NSGA-II identifies compromise solutions that improve matching performance, transferred energy, and suitability simultaneously, but not mobility cost.The method improves the main system-level objectives rather than maximizing one metric alone.
  • Results: The Pareto analysis shows that reducing mobility cost can lower matched pairs, transferred energy, and suitability, making single-objective evaluation misleading.Balanced solutions are preferred when multiple operational goals must be satisfied simultaneously.
  • Results: Hypervolume increased and stabilized, indicating convergence toward a stable Pareto front with broader, higher-quality trade-off coverage than comparison methods.This broader coverage supports operator selection across different priorities.
Loading 2609.10551v1…