New Model Optimizes EV Fast-Charging Networks for County-Level Deployment
As electric vehicle (EV) adoption accelerates across China and the world, one of the most pressing challenges for sustainable transportation lies not in major metropolitan centers, but in the vast network of counties, towns, and rural corridors that connect urban and rural life. While cities have seen a surge in charging infrastructure, many suburban and rural regions still suffer from sparse networks, unreliable access, and inefficient planning—barriers that hinder the broader rollout of EVs beyond city limits. A recent study by researchers from Tsinghua University and Guangdong Power Grid offers a groundbreaking solution: a data-driven, behavior-aware optimization model for fast-charging station deployment tailored specifically to the unique dynamics of county-level transportation networks.
The research, led by Zhuoxu Chen and Zechun Hu from Tsinghua University’s Department of Electrical Engineering, in collaboration with Yujian Wan and Junsong Li from Shantou Power Supply Bureau of Guangdong Power Grid, introduces a comprehensive planning framework that goes beyond traditional cost-minimization models. Their approach, published in Automated Electric Power Systems, integrates real-world user behavior, traffic patterns, and economic constraints into a unified optimization model—ushering in a new era of intelligent, user-centric EV infrastructure planning.
At the heart of the challenge is a paradox: while EV ownership is rising, the charging infrastructure in less densely populated areas remains underdeveloped. According to 2022 data, China added 259,300 new charging points and 37,000 public charging stations, achieving a vehicle-to-charger ratio of 2.7:1. Yet, this average masks significant regional disparities. In rural and county-level regions, coverage remains thin, and during peak travel periods—especially holidays and long weekends—drivers face long queues, limited options, and the ever-present anxiety of running out of power mid-journey.
Existing planning models often fall short in these contexts. Many rely on deterministic assumptions, treating drivers as perfectly rational agents who always choose the nearest or cheapest station. In reality, human behavior is far more complex. Drivers make decisions based on a mix of factors: how far they’re willing to detour, how much they’re willing to pay, how crowded a station appears, and even what amenities are nearby—like cafes, restrooms, or shopping areas. Ignoring these behavioral nuances can lead to suboptimal station placement, underutilized investments, and persistent congestion at popular hubs.
Chen and his team recognized this gap. “Traditional models assume full information and perfect rationality,” explains Hu, the paper’s corresponding author. “But in county-level areas, drivers often lack real-time data on charger availability, pricing, or wait times. They make decisions under uncertainty, influenced by habits, preferences, and incomplete information. Our model accounts for this bounded rationality.”
To capture this complexity, the team developed a stochastic user equilibrium (SUE) framework rooted in discrete choice theory. Unlike deterministic models, SUE acknowledges that users do not always make the “optimal” choice. Instead, they select stations based on perceived utility, which incorporates multiple cost components: detour distance, charging price, station congestion, and surrounding amenities. Each of these factors is weighted according to user sensitivity, allowing the model to simulate realistic decision-making under uncertainty.
The innovation lies in how these behavioral dynamics are embedded into the planning process. Rather than treating user behavior as a static input, the model treats it as an equilibrium condition—meaning that the final station layout must be consistent with how users are expected to behave. This creates a feedback loop: the placement of stations influences user choices, and those choices, in turn, affect the performance and profitability of the stations.
To generate realistic demand patterns, the researchers employed a trip chain simulation method. This approach reconstructs individual travel itineraries by simulating origin-destination pairs, departure times, driving speeds, and battery consumption. Crucially, the model includes external connection nodes—points where intercity highways intersect with local roads—allowing it to capture not only local commuter demand but also transient demand from tourists, delivery vehicles, and long-distance travelers.
The simulation was calibrated using real-world data, including vehicle performance parameters (battery capacity, energy consumption rates), driver behavior (initial state of charge, detour tolerance), and regional travel patterns. Four distinct scenarios were modeled, reflecting variations in total vehicle population (from 120,000 to 200,000), EV penetration (5%), and the proportion of incoming external traffic (ranging from 1% to 40%). This multi-scenario approach ensures that the planning model is robust to seasonal fluctuations and unpredictable demand spikes.
Once demand is simulated, the optimization model kicks in. The objective is twofold: minimize the total cost of investment and operation for charging station developers, and minimize the aggregate detour distance for users—a proxy for user convenience and satisfaction. The model must also respect physical and economic constraints, such as power grid capacity limits, land availability, and construction budgets.
What sets this model apart is its computational tractability. The original formulation involves complex, non-convex equilibrium constraints derived from the Logit choice model—a common tool in transportation economics. These constraints are notoriously difficult to solve, especially when combined with integer variables (e.g., whether to build a station at a given location). To overcome this, the team applied a series of linearization techniques, transforming the problem into a mixed-integer linear program (MILP). This allows the use of powerful commercial solvers like Gurobi to find globally optimal solutions efficiently.
The case study was conducted on a modified 33-node transportation network, designed to reflect the sparse road connectivity typical of county-level regions. Twelve candidate sites were considered for fast-charging stations, each capable of hosting between 8 and 16 chargers rated at 96 kW. After optimization, the model recommended building stations at just five locations: nodes 2, 17, 20, 21, and 26. Notably, node 17 is an external connection point, highlighting the strategic importance of gateway locations for serving through-traffic.
The results reveal several key insights. First, the selected stations are not necessarily the most centrally located or the most accessible from all origins. Instead, they strike a balance between coverage, cost, and user behavior. For example, station n17, though on the periphery, serves a high proportion of external vehicles—60.1% in one scenario—making it a critical node for regional connectivity. Building additional stations at other external nodes was deemed unnecessary, as it would increase capital costs without significantly improving service quality.
Second, the model naturally avoids over-concentration. In scenario-based simulations, user demand was distributed relatively evenly across the five stations, preventing the kind of congestion seen in real-world hotspots. This is a direct result of the SUE framework: when a station becomes too crowded, its perceived cost increases (due to longer wait times), prompting users to consider alternatives—even if they require a slightly longer detour.
Third, the integration of non-travel factors—such as the availability of nearby amenities—proved influential. Stations located near commercial or tourist zones were more attractive, not because they were cheaper or closer, but because they offered a better overall experience. This aligns with growing evidence that EV charging is not just a functional necessity but part of a broader mobility and lifestyle ecosystem.
To validate the model’s accuracy, the researchers conducted extensive sensitivity analyses. They tested how changes in user preferences—such as increased aversion to waiting or higher sensitivity to price—affected the optimal solution. As expected, when users became more sensitive to congestion, the model recommended more distributed networks to avoid bottlenecks. Conversely, when price sensitivity dominated, clustering around lower-cost zones became favorable.
Perhaps most revealing was the analysis of the rationality coefficient—a parameter that controls how closely users follow the “optimal” choice. When set to zero (fully random behavior), the average detour distance was high, and station utilization was low. When set to infinity (perfect rationality), users flocked to the cheapest or closest stations, leading to high congestion. The optimal balance emerged at intermediate values, where users exhibit bounded rationality—a finding that underscores the importance of behavioral realism in infrastructure planning.
From a policy perspective, the implications are significant. As governments push to electrify transportation in rural and suburban areas—part of broader efforts to reduce emissions and promote energy independence—this model offers a scientific basis for investment decisions. It enables planners to move beyond guesswork and anecdotal evidence, replacing them with data-driven, behaviorally informed strategies.
Moreover, the model supports the concept of “smart charging corridors”—strategic networks of fast-charging stations along major travel routes that ensure seamless long-distance EV travel. By optimizing both location and capacity, such corridors can reduce range anxiety, improve grid stability, and enhance the overall user experience.
The research also has practical applications for utility companies and private charging operators. For Guangdong Power Grid, which funded the study, the model provides a tool for assessing the impact of new charging stations on local distribution networks. By integrating power flow constraints, the model ensures that new stations do not overload transformers or cause voltage instability—a common issue in rural grids with limited capacity.
For private operators, the model helps identify high-value investment opportunities. By simulating user behavior and revenue potential under different pricing schemes, it allows for more accurate financial forecasting and risk assessment. This is particularly important in regions where demand is uncertain and competition may emerge.
Looking ahead, the team plans to extend the model to include renewable energy and energy storage. Integrating solar panels or battery systems at charging stations could reduce grid dependence, lower operating costs, and enhance sustainability. The framework is also being adapted for dynamic pricing strategies, where real-time signals adjust based on congestion, electricity prices, and grid conditions.
In an era where artificial intelligence and big data are reshaping every aspect of transportation, this study stands out for its balance of technical rigor and practical relevance. It does not rely on black-box algorithms or opaque machine learning models. Instead, it builds on well-established principles of economics, game theory, and operations research, making its results interpretable, auditable, and actionable.
The success of the EV revolution will not be determined solely by advances in battery technology or manufacturing scale. It will also depend on the intelligence of the infrastructure that supports it. As this research demonstrates, the future of charging is not just about building more stations—it’s about building the right ones, in the right places, for the right people.
By grounding infrastructure planning in human behavior, this model offers a blueprint for equitable, efficient, and resilient EV networks—ones that serve not just urban elites, but the full spectrum of drivers who rely on the open road.
Chen Zhuoxu, Hu Zechun et al., Automated Electric Power Systems. DOI: 10.7500/AEPS20230731003