Smart Charging Strategy Balances Grid Needs and Driver Preferences
As electric vehicle (EV) adoption accelerates globally, the integration of millions of new loads into the power grid presents both a challenge and an opportunity. Uncoordinated charging could strain infrastructure, particularly during peak evening hours when drivers return home and plug in. However, if managed intelligently, the collective battery capacity of EVs can become a powerful asset for grid stability, helping to absorb surplus renewable energy and reduce reliance on fossil-fuel peaking plants. The critical question is no longer whether EVs can support the grid, but how to design systems that make this integration seamless, economically beneficial for all parties, and acceptable to the end user—the driver.
Traditional approaches to managing large fleets of EVs often rely on centralized optimization models. These models treat the entire EV fleet as a single, homogeneous entity, calculating an ideal charging schedule based on aggregate grid conditions and energy prices. While mathematically elegant, these methods face significant practical hurdles. They require immense computational power, often lack granular, real-world data on individual driving patterns, and crucially, they fail to capture the complex, strategic interactions between the various stakeholders involved. The load aggregator (LA), acting as an intermediary between the grid and consumers, has its own profit motives. Individual EV owners, meanwhile, are driven by a mix of economic concerns and personal convenience. A top-down directive to charge at a specific time may be economically optimal for the grid, but it risks alienating users if it conflicts with their daily routines or if the financial incentives are not compelling enough. This disconnect between theoretical models and real-world human behavior has been a persistent roadblock to the widespread deployment of effective vehicle-to-grid (V2G) programs.
A groundbreaking study published in Power System Protection and Control offers a sophisticated solution to this dilemma. Researchers Meixia Zhang, Xiaoqing Wang, Xiu Yang, An Zhang, and Yulin Fu from the College of Electrical Engineering at Shanghai University of Electric Power have developed a novel two-level “Stackelberg game” optimization strategy that explicitly models the strategic interplay between a load aggregator and clusters of EV users. Their model, titled “Stackelberg game optimization scheduling strategy for aggregated electric vehicles considering customer satisfaction and the road network,” moves beyond simplistic assumptions by incorporating real-world travel data, user psychology, and the physical constraints of the transportation network. The result is a framework that not only enhances grid stability and increases renewable energy consumption but also delivers tangible benefits to both the aggregator and the individual EV owner, creating a true win-win scenario.
The foundation of this innovative approach is a profound understanding of human mobility. Instead of relying on synthetic or averaged travel patterns, the research team utilized a rich dataset of real-world driving behavior. Their analysis was built upon a month of trip records from Didi Chuxing, China’s largest ride-hailing service, focusing on the city of Chengdu. This provided a high-resolution picture of actual journeys, including pickup and drop-off locations, travel times, and vehicle trajectories. To transform this raw GPS data into a meaningful model of urban travel, the researchers constructed a detailed digital representation of the city’s road network. Using geographic information system (GIS) software, they mapped the primary roads and intersections, creating a topological graph that the simulated vehicles could navigate. Furthermore, they enriched this spatial model by integrating Point of Interest (POI) data from Gaode Map, a leading Chinese digital map service. This allowed them to classify different areas of the city into functional zones—residential, commercial, work, and public service—providing context for why people travel from one point to another. This fusion of real travel data with a geographically accurate and functionally annotated road network is a significant advancement, ensuring that the simulated EV charging demands are grounded in reality, reflecting actual commuting patterns, shopping trips, and leisure activities.
With a realistic model of travel behavior established, the next step was to understand the charging needs that arise from this mobility. The researchers simulated the journeys of 2,000 private EVs over a representative day. For each vehicle, they tracked its state of charge (SOC), calculating energy consumption based on the distance traveled and the type of road (e.g., highway vs. city street), which affects driving speed and efficiency. A charging event was triggered when the battery level dropped below a safe threshold, typically 20%. This process generated a highly detailed, spatio-temporal map of charging demand across the city, revealing predictable patterns. As expected, a significant charging peak emerged in the late afternoon and evening, between 5 PM and 10 PM, as drivers returned from work and other activities. Conversely, charging demand was much lower in the early morning hours. This granular forecast of when and where charging would occur provided the essential input for the subsequent optimization phase.
The sheer number and diversity of EVs make it impractical for a load aggregator to manage each vehicle individually. To address this, the researchers employed a K-means++ clustering algorithm, an advanced form of unsupervised machine learning. This algorithm grouped the 2,000 EVs into five distinct clusters based on the similarity of their “in-network” and “out-network” times—the periods when they were available to be charged or to discharge energy back to the grid. The K-means++ variant was chosen over the standard K-means because it intelligently selects the initial cluster centers to be as far apart as possible, which dramatically improves the speed and reliability of the clustering process, avoiding suboptimal groupings. The resulting clusters represented different user archetypes: one group that plugged in early in the morning, another that arrived home late in the evening, a third that had a midday charging window, and so on. This clustering is a pivotal step, as it allows the LA to treat each group as a single, more predictable entity with a defined availability window, vastly simplifying the scheduling problem.
At the heart of the proposed system is the Stackelberg game, a mathematical framework for modeling hierarchical decision-making. In this scenario, the load aggregator (LA) is the “leader,” and the five clusters of EVs are the “followers.” The game proceeds in a specific sequence. First, the LA, which owns or manages a portfolio of distributed energy resources including wind turbines, solar panels, and a battery energy storage system (BESS), announces its pricing strategy for the day. This strategy consists of two dynamic prices for each hour: a higher “buy” price at which the LA will purchase electricity from EVs that are discharging (V2G), and a lower “sell” price at which it will sell electricity to EVs that are charging. The LA’s primary goal is to maximize its own profit, which is calculated as its revenue from selling power to the grid and to its customers, minus its costs for purchasing power from the grid and maintaining its energy assets.
The EV clusters, as followers, observe the LA’s announced prices and then make their own optimal decisions. Each cluster’s goal is to maximize its “consumer surplus,” which is the difference between the total benefit its members derive from charging and the total cost they incur. This is where the model introduces a revolutionary concept: user satisfaction. Instead of treating EV owners as purely economic agents, the researchers developed a comprehensive utility function that captures two key aspects of the user experience. The first is charging cost satisfaction, which is a straightforward economic metric—users prefer lower prices. The second is usage pattern satisfaction, which measures how much the optimized charging schedule deviates from the user’s natural, unoptimized behavior. A driver who would normally charge immediately upon arriving home at 6 PM might be dissatisfied if the optimal schedule requires them to wait until 9 PM, even if the price is lower. This function quantifies that inconvenience. The overall user satisfaction is a weighted sum of these two components, allowing the model to represent different user types. Some users are highly price-sensitive and are willing to shift their charging time significantly to save money. Others prioritize convenience and are less willing to alter their habits, even for a financial gain.
The brilliance of the Stackelberg equilibrium is that it finds a stable solution where no player can improve their outcome by unilaterally changing their strategy. The LA sets prices that are designed to elicit a specific, profitable response from the EV clusters. The EV clusters, in turn, respond by charging and discharging in a way that maximizes their own satisfaction given those prices. The researchers used an improved genetic algorithm to solve this complex, nested optimization problem, iteratively adjusting the LA’s prices and the clusters’ responses until an equilibrium was reached.
The simulation results demonstrate the profound impact of this approach. When compared to a scenario without demand response, the optimized strategy led to a significant “peak shaving and valley filling” effect on the grid’s load curve. The peak-to-valley difference, a key metric of grid stress, was reduced by over 300 kW, a substantial improvement that enhances grid stability and reduces the need for expensive peaking power plants. The LA’s revenue increased by 12.9%, a direct result of its ability to buy low, sell high, and efficiently manage its own renewable and storage assets. Crucially, the consumer surplus for EV users also increased by an average of 9.7%. This means that despite the strategic shifting of their charging times, users ended up better off, primarily because the financial savings from charging at low prices and selling at high prices outweighed the minor inconvenience of a changed schedule. This dual increase in profit and user satisfaction is the hallmark of a truly successful and sustainable system.
The model also proved highly effective at integrating renewable energy. By aligning EV charging with periods of high solar and wind generation—typically the middle of the day when the sun is shining and the wind is blowing—the system significantly increased the utilization of these clean sources. Instead of this renewable energy being curtailed (wasted) because there is no immediate demand, it is stored in EV batteries. This stored energy is then available to be discharged back to the grid during the evening peak, when demand is high and renewable generation is low. This not only reduces carbon emissions but also improves the overall economics of renewable energy projects.
One of the most compelling findings of the study is its ability to provide personalized service based on user preference. The researchers ran three different scenarios by adjusting the weights in the user satisfaction function. In one scenario, users were modeled as highly price-sensitive. The LA responded by creating a pricing signal with a large spread between peak and off-peak prices. This created a strong financial incentive for users to shift their charging, leading to a very flat load curve and high consumer surplus, though the LA’s profit was slightly lower. In another scenario, users were modeled as highly convenience-sensitive. The LA offered a flatter pricing structure with less variation, resulting in a smaller shift in charging behavior. While these users had a lower consumer surplus, their charging patterns were closer to their natural habits, and the LA was able to achieve its highest profit. This flexibility shows that the system is not a one-size-fits-all solution but a dynamic platform that can be tuned to serve different market segments and user personas.
This research represents a significant leap forward in the field of smart grid management. It successfully bridges the gap between abstract economic theory and the messy reality of human behavior and urban infrastructure. By grounding their model in real travel data, respecting user preferences through a sophisticated utility function, and using a game-theoretic framework to model strategic interaction, Zhang Meixia, Wang Xiaoqing, Yang Xiu, Zhang An, and Fu Yulin have created a blueprint for a future where EVs are not just a mode of transportation but an integral, intelligent, and beneficial component of a resilient, clean, and efficient energy system. Their work demonstrates that with the right design, the transition to electric mobility can be a smooth and mutually beneficial journey for everyone involved.
Meixia Zhang, Xiaoqing Wang, Xiu Yang, An Zhang, Yulin Fu, College of Electrical Engineering, Shanghai University of Electric Power, Power System Protection and Control, DOI: 10.19783/j.cnki.pspc.230925