Source-linked AI summary
Placement of EV Charging Stations --- Balancing Benefits among Multiple Entities
Chao Luo, Yih-Fang Huang, Vijay Gupta
TL;DR
The paper addresses how to place EV charging stations over multiple stages while balancing provider profits, consumer satisfaction, and power-grid effects in a competitive market. It combines nested-logit demand estimation, network modeling, and Bayesian-game placement policies, then evaluates them with a simulation platform. In the San Pedro case study, placements align with EV traffic flow, while providers prefer clustering when offering differentiated charging services.
Problem
EV charging-station placement requires balancing consumer satisfaction, provider interests, and power-grid impacts while accounting for mobility and strategic interactions among competing providers.
Method
The paper combines a nested logit model for charging preferences and demand, transportation and power-network modeling, and Bayesian-game analysis of multi-stage provider placement decisions.
Results
In the San Pedro case study, charging-station placement is highly consistent with the EV traffic-flow heatmap, and providers prefer clustering rather than spatial separation.
Takeaways & Limitations
The case study indicates that differentiated charging services can support clustered provider locations while placement follows observed EV traffic patterns.
Abstract
from arXiv · showhide
This paper studies the problem of multi-stage placement of electric vehicle (EV) charging stations with incremental EV penetration rates. A nested logit model is employed to analyze the charging preference of the individual consumer (EV owner), and predict the aggregated charging demand at the charging stations. The EV charging industry is modeled as an oligopoly where the entire market is dominated by a few charging service providers (oligopolists). At the beginning of each planning stage, an optimal placement policy for each service provider is obtained through analyzing strategic interactions in a Bayesian game. To derive the optimal placement policy, we consider both the transportation network graph and the electric power network graph. A simulation software --- The EV Virtual City 1.0 --- is developed using Java to investigate the interactions among the consumers (EV owner), the transportation network graph, the electric power network graph, and the charging stations. Through a series of experiments using the geographic and demographic data from the city of San Pedro District of Los Angeles, we show that the charging station placement is highly consistent with the heatmap of the traffic flow. In addition, we observe a spatial economic phenomenon that service providers prefer clustering instead of separation in the EV charging market.
I. INTRODUCTION
The paper formulates multi-stage EV charging-station placement as a competition among service providers that balances provider profits, consumer satisfaction, and power-grid effects. It combines consumer-demand modeling, network interactions, and Bayesian-game analysis to derive placement policies and studies the resulting system with simulation.
- Strategic placement: A Bayesian game analyzes strategic interactions among providers, whose objective combines expected profit with penalties for power-grid disturbance while satisfying QoS constraints.The framework addresses providers’ incomplete knowledge of competitors’ placement costs and utility functions.
- Motivation and gap: Existing placement studies often omit overall consumer satisfaction, electric-power-network impacts, EV mobility, or competitive service-provider decision-making.The paper also identifies complete-information assumptions in prior noncooperative models as potentially restrictive.
- Contributions: The proposed strategy places stations across multiple stages as EV penetration increases, accounting for interactions among EVs, roads, and the electric power grid.New stages begin when existing stations no longer satisfy quality-of-service constraints.
- Contributions: A nested logit model characterizes EV charging preferences and predicts aggregated charging demand for station-placement decisions.The model is intended to provide insight into EV owners’ preferences and decision-making.
- Simulation platform: The EV Virtual City 1.0 simulator examines interactions among EV owners, transportation and power networks, urban infrastructure, and charging stations.The study uses geographic and demographic data from San Pedro District of Los Angeles.
- Market model: The charging market is modeled as an oligopoly in which providers choose locations and prices for imperfectly substitutable Level 1, Level 2, and Level 3 services.Providers are assumed to operate affiliated stations with a common retail charging price.
B. The Disturbance on Power Grid Due to EV Charging
The paper models large-scale EV charging as a potential source of power-grid disturbance and incorporates that disturbance into providers’ utility functions. A weighting coefficient controls the trade-off between charging profit and grid stress.
- Grid impacts: Simultaneous large-scale EV charging can disrupt grid operation through frequency variation, voltage imbalance, voltage variation, and power loss.The paper notes that generators conventionally coordinate real and reactive power to maintain stability and regulate frequency and voltage.
- Penalty model: The utility function subtracts a weighted grid-disturbance penalty from charging profit: U_k = Π_k − wB_k.Here, Π_k is charging profit, B_k is the penalty from large-scale charging, and w reflects tolerance to that penalty.
- Penalty model: w = 0 ignores charging’s grid impact, whereas larger nonzero w assigns greater importance to the grid penalty.The paper states that w should be determined using heuristic and empirical data.
C. Quality-of-Service Constraints
The placement problem imposes quality-of-service constraints based on average service delay probability and average service coverage. Because these quantities depend on interacting travel and infrastructure conditions, the paper estimates them with Monte Carlo simulation.
- QoS metrics: The two QoS metrics are average service delay probability Υ_k and average service coverage Ξ_k for each provider.The metrics evaluate delayed charging attempts and accessible provider-specific stations along EV routes.
- Metric definitions: Υ_k averages each EV owner’s delayed-charging probability, while Ξ_k averages accessible Level k stations along routes from origin to destination.Delay probability is the ratio of delayed charging events to total charging attempts.
- Estimation: Both QoS metrics are random variables affected by all EV travel patterns, the urban road network, and charging-station locations.The paper therefore uses a Monte Carlo method rather than a simple closed-form formula to estimate them.
D. Multi-stage Charging Station Planning Scheme
At each planning stage, providers seek an optimal station-placement policy subject to predetermined QoS constraints. The paper identifies demand prediction, grid-impact measurement, and tractable policy derivation as the central solution questions.
- Planning problem: Providers solve for an optimal placement policy at each planning stage while satisfying predetermined QoS constraints.The planning problem is framed around selecting station locations under service-quality requirements.
- Solution questions: The formulation requires predicting aggregated demand at each candidate station, characterizing EV charging impacts on the grid, and deriving a tractable placement policy.These are stated as the three principal questions confronting the solution method.
- Demand prediction: A nested logit model is used to estimate charging demand by analyzing EV-owner behavior and charging choices.The resulting demand estimates support station-level planning.
- Solution approach: The placement optimization is described as intractable in its direct formulation, motivating additional modeling and solution procedures.The supplied passage introduces this tractability issue before continuing the derivation.
III. CHARGING DEMAND OF EV CHARGING STATION
The paper defines aggregated charging demand at a candidate station by combining EV owners’ station-choice probabilities with their electricity requirements. A nested logit model represents station attractiveness using behavioral and station-related factors.
- Aggregated charging demand equals the sum of each EV owner’s station-choice probability multiplied by the electricity required for charging.
A. Nested Logit Model And Probability of Choice
The nested logit model represents EV charging choice as utility maximization among charging alternatives, with provider-level and station-level components. Choice probabilities incorporate observable attributes, destination and detour effects, amenities, and correlated unobservable utility within providers’ nests.
- The nested logit model treats consumers as utility maximizers choosing the alternative that provides the highest utility.
- Unobservable utilities follow a generalized extreme-value structure and are correlated within each provider’s charging nest but uncorrelated across nests.
- EV owner utility is decomposed into the utility of selecting a service provider and the utility of selecting a station within that provider’s nest.
- Observable utility combines weighted attributes of charging stations and EV owners.
- Charging time, retail price, and EV-owner income differentiate charging services through their effects on utility.
- Choice utility also accounts for whether a station is near the destination, detouring distance, and nearby restaurant, shopping-center, and supermarket amenities.
- The model averages over random unobservable utility to obtain each EV owner’s probability of choosing a station.
B. Charging Demand Estimation
The paper converts EV station-choice probabilities into predicted station demand by modeling the electricity purchased by each EV owner. Model coefficients can be calibrated from preference survey data to produce statistically meaningful aggregate predictions.
- Predicted demand at a station is obtained after computing EV owners’ probabilities of choosing that station.
- Each EV owner’s purchased electricity is modeled as a uniformly distributed random variable between lower and upper charging-demand limits.
- Nested-logit coefficients can be estimated from preference surveys, yielding statistically meaningful aggregate charging-demand predictions despite individual deviations from modeled probabilities.
IV. THE IMPACT OF EV INTEGRATION ON POWER GRID
The paper evaluates how large-scale EV charging affects power-grid operation through voltage, frequency, and generator-load changes. It incorporates grid stress into charging-station deployment to reduce the added impact of EV charging.
- Large-scale EV integration can disrupt grid stability through power loss, frequency variation, and voltage imbalance.
- Power-flow analysis solves voltage, real-power, and reactive-power conditions across generators and substations.
- The deployment model considers alleviating EV-charging stress on the power grid when determining optimal station placement.
- Voltage and frequency are key power-quality factors because reactive-power imbalance changes voltage while active-power imbalance shifts frequency.
- Generator-power fluctuations with and without EV charging provide a metric for evaluating charging-station stress on the grid.
A BAYESIAN GAME
The paper models competing charging service providers as a Bayesian game because providers lack complete information about rivals’ placement costs and actions. At each stage, providers choose placement policies and prices to maximize expected utility under stated assumptions.
- Game formulation: The Bayesian game captures strategic interaction when providers know their own placement costs but not those of two competing providers.Placement costs include equipment, installation, construction, and land rental costs.
- Game formulation: Each provider’s strategy space contains all possible placement policies across its L candidate locations, yielding 2^L policies.The type space is an L-dimensional vector of placement costs.
- Competition and pricing: For simplicity, competing placement policies follow a Binomial(2^L, 0.5) distribution, although the analysis can use other distributions.This distribution represents each provider’s conjecture about how rivals will act.
- Competition and pricing: The model assumes Bertrand competition, so providers do not cooperate and choose actions to maximize their own utility.Retail prices are determined for each combination of placement policies using first-order conditions.
- Equilibrium policy: A provider selects placement policy l when its type falls within the hypervolume where that policy yields higher expected revenue than alternatives.The hypervolume is defined by the intersection of inequalities comparing placement strategies.
VI. SIMULATION PLATFORM AND CASE STUDY
The EV Virtual City 1.0 integrates transportation, power-grid, geographic, demographic, and travel data to simulate staged charging deployment in San Pedro District. Results show traffic-aligned placement, differentiated provider strategies, near-linear station growth with EVs, and clustering among providers.
- Simulation platform: The EV Virtual City 1.0 integrates geographic, demographic, road-network, power-network, travel-pattern, and traffic-flow data in a flexible simulation platform.The platform can include or exclude modules for different simulation needs.
- Case study: The case study uses San Pedro District data, including Census road and ZCTA shapefiles, activity-location centroids, California power-grid maps, and travel statistics from the 2009 NHTS.The simulated power system uses the IEEE 118-bus test case and MATPOWER to calculate locational marginal prices and generator outputs.
- Simulation design: The simulation evaluates four stages with 5,000, 10,000, 15,000, and 20,000 EVs, using EV movement, traffic heatmaps, and Bayesian-game deployment decisions.Figures 5–8 show stage-specific placements, while Figure 10 tracks station counts as EV penetration increases.
- Results: Optimal charging-station deployment is consistent with the EV traffic-flow heatmap.The authors interpret this consistency as evidence that the model captures EV mobility and supports convenient charging access.
- Results: Level 1 stations outnumber Level 2 and Level 3 stations, while providers differ between even-area coverage and concentration at hot locations.The paper attributes the larger Level 1 count to its longer charging time and average-delay constraints.
- Results: Station counts grow almost linearly with EV numbers except initially, when Level 1 and Level 2 providers place more stations to satisfy service-coverage constraints.As station numbers increase, service coverage becomes less concerning to providers.
- Results: Service providers prefer clustering because differentiated charging products soften price competition without requiring spatial separation.The products differ in voltage, current, charging speed, and charging price.
VII. CONCLUSION
The paper proposes a multi-stage EV charging-station placement solution that balances EV-owner, station-owner, and power-grid interests. A Bayesian game and the EV Virtual City 1.0 support the analysis, whose case study finds traffic-aligned placement and provider clustering.
- Conclusion: The proposed placement solution balances the benefits of EV owners, charging-station owners, and power-grid operators.The framework formulates provider competition as a Bayesian game to obtain optimal placement policies.
- Conclusion: The San Pedro District case study finds charging-station placement highly consistent with EV traffic flow and providers preferring clustering over separation.The simulation software used for the case study is The EV Virtual City 1.0 on Repast.