Electric Vehicles Reshape Urban Grids and Roads Through Elastic Demand Modeling
In a groundbreaking study that bridges the disciplines of power systems engineering and urban transportation planning, researchers have unveiled a new mathematical framework capable of capturing the intricate interplay between electric vehicle (EV) drivers’ behavior and the operational dynamics of coupled power and transportation networks. As cities worldwide accelerate their transition toward electrified mobility, understanding how EV users respond to real-time changes in traffic congestion and electricity pricing has become critical—not only for grid reliability but also for equitable and efficient urban infrastructure planning.
The research, led by Shiwei Xie, Kaiyue Chen, Yachao Zhang, Longtao Xie from Fuzhou University, and Qiuwei Wu from Tsinghua University’s Shenzhen campus, introduces a novel two-layer game model grounded in quasi-variational inequality (QVI) theory. Published in the Proceedings of the CSEE, the paper tackles a long-standing challenge in integrated infrastructure modeling: how to account for elastic demand—the phenomenon where the number of trips made by EV users is not fixed but instead dynamically responds to travel costs, including both time spent in traffic and the price of charging.
Traditional models have often assumed that travel demand is static, treating the number of vehicles on the road as predetermined. While this simplification eases computation, it fails to reflect real-world behavior, where potential travelers may choose to delay a trip, switch to public transport, or even cancel a journey altogether if congestion worsens or charging becomes too expensive. By incorporating elasticity, the new model captures a feedback loop: higher travel costs suppress demand, which in turn alleviates congestion and reduces strain on the power grid—a nuance entirely missed by rigid-demand approaches.
At the heart of the methodology is a dual-layer equilibrium concept. The inner layer represents the strategic decisions of individual travelers. Faced with information about road conditions and charging station prices, EV and conventional vehicle users independently select routes that minimize their personal travel costs. This layer results in a user equilibrium where no driver can reduce their cost by unilaterally changing routes—a classic principle in transportation economics, now extended to include charging behavior and elastic trip generation.
The outer layer models the interaction between the transportation network and the distribution grid. Here, the power system operator solves an optimal power flow problem that includes EV charging loads as flexible but price-sensitive demands. The marginal electricity prices derived from this optimization—specifically, locational marginal prices (LMPs)—are then fed back to the transportation side as dynamic charging tariffs. This creates a closed-loop system: traffic patterns determine where and when EVs charge, which shapes the electrical load profile; the grid’s response to that load, in turn, sets prices that influence future driver decisions.
To solve this complex, interdependent system, the team developed a specialized algorithmic architecture. The outer loop employs a fixed-point iteration scheme that alternates between solving the traffic equilibrium and the power flow problem. Within each outer iteration, the inner traffic equilibrium—complicated by the fact that the feasible set of travel demands depends on the unknown equilibrium cost itself—is resolved using a viscous projection approximation algorithm. This technique cleverly handles the set-valued nature of elastic demand by iteratively refining an estimate of the feasible region until convergence.
The model was rigorously tested on a real-world integrated system based on Fuzhou, a major city in southeastern China. The testbed combined a 45-node transportation network with a 56-node radial distribution grid, complete with four strategically located EV charging stations. Simulations revealed several key insights with profound practical implications.
First, ignoring demand elasticity leads to significant overestimation of grid stress. When the model was run with fixed (inelastic) demand, the resulting charging load was consistently higher, driving up operational costs for the utility. In contrast, the elastic model showed that as charging prices or travel times increased, some users opted out of trips, naturally curbing peak demand. This self-regulating behavior not only lowered the total cost of grid operation but also led to a more balanced utilization of charging infrastructure.
Second, the study demonstrated that elasticity promotes price fairness and grid stability. Under inelastic conditions, the marginal prices at different charging stations varied widely, reflecting severe congestion or voltage issues at specific grid nodes. However, with elastic demand, these price disparities narrowed considerably. The reason is intuitive: when a station becomes expensive due to local grid congestion, price-sensitive users divert to alternatives or reduce their travel, thereby alleviating the bottleneck. This dynamic balancing act prevents any single node from becoming a critical failure point.
Third, the research quantified the role of user sensitivity parameters. Using a Logit-based demand function, the team varied two key factors: the “attractiveness” of trip destinations (a proxy for the intrinsic need or desire to travel) and the “negative elasticity parameter” (which measures how strongly users react to cost increases). They found that grid operational costs are far more sensitive to changes in destination attractiveness than to changes in user price sensitivity. In other words, the sheer volume of desired trips is a more dominant driver of grid stress than how fickle users are about costs. However, in specific scenarios—such as when a distant charging station becomes a relief valve during city-center congestion—high elasticity can lead to non-monotonic effects, where increasing demand first raises and then lowers the load at a particular station.
From a policy perspective, these findings are invaluable. Urban planners and grid operators can no longer treat EV adoption as a simple load-addition problem. The behavioral response of users is a powerful, built-in demand-response mechanism. Pricing strategies for public charging infrastructure, therefore, should not only aim to recover costs or manage grid peaks but also be designed to harness this elasticity for system-wide efficiency. For instance, dynamic pricing that is transparent and predictable can guide users to make choices that collectively stabilize both traffic and power networks.
Moreover, the model provides a robust tool for infrastructure investment decisions. By simulating the equilibrium state under various scenarios—such as the addition of a new charging station or the upgrade of a distribution feeder—authorities can forecast not just the direct impact but also the induced changes in travel and charging behavior. This holistic view prevents costly misallocations, such as building a high-capacity charger in a location that users will avoid due to poor road access or high concurrent electricity prices.
The technical contribution of the paper is equally significant. By framing the elastic traffic equilibrium as a QVI, the authors circumvent the limitations of traditional variational inequality (VI) approaches, which require a fixed, convex feasible set. The QVI framework elegantly accommodates the reality that the set of possible travel demands is itself a function of the equilibrium outcome. This theoretical advance opens the door for modeling other complex socio-technical systems where human decisions and physical infrastructure are co-dependent.
The proposed algorithms also mark a step forward in computational tractability. Despite the high dimensionality of the Fuzhou test case, the dual-loop method converged in just six outer iterations, with the inner viscous projection algorithm demonstrating superior speed compared to its non-viscous counterpart. This efficiency is crucial for real-world applications, where models may need to be run repeatedly for planning or even in near-real-time for operational decision support.
Looking ahead, the authors suggest extending their framework to incorporate time-varying elasticity—a natural next step given that user sensitivity to price and congestion likely differs between morning commutes and weekend leisure trips. They also propose exploring differential QVIs in Hilbert spaces, which would allow for a fully dynamic, rather than static, representation of traffic and power flows.
In an era where the boundaries between energy and mobility are rapidly dissolving, this research provides a much-needed analytical lens. It moves beyond siloed thinking to offer a unified, behaviorally grounded model of the urban ecosystem. As cities strive to meet climate targets while maintaining livability, such integrated tools will be indispensable for crafting policies that are not only technically sound but also socially attuned.
For grid operators, the message is clear: the next megawatt of EV load is not just a number on a spreadsheet—it is the aggregate outcome of thousands of individual decisions, each shaped by a complex calculus of time, money, and convenience. By respecting and modeling this complexity, we can build a future where electric mobility enhances, rather than overwhelms, our urban infrastructure.
By Shiwei Xie, Kaiyue Chen, Yachao Zhang, Longtao Xie (Fuzhou University) and Qiuwei Wu (Tsinghua University). Published in Proceedings of the CSEE. DOI: 10.13334/j.0258-8013.pcsee.230715.