Electric Vehicle Charging Patterns Predicted Using Travel Simulation
A groundbreaking study by researchers from Southeast University has introduced a sophisticated new model for predicting the spatial and temporal distribution of electric vehicle (EV) charging loads. This innovative approach, detailed in the journal Electric Power Engineering Technology, integrates travel chain theory with real-world geographic data to simulate user behavior and forecast charging demand with unprecedented accuracy. The research, led by Shen Xiaoqi, Fang Xin, Tan Linlin, Li Xinguo, and Sun Jiaqi, addresses the critical challenge of randomness and uncertainty in EV charging patterns, which has long hindered effective grid management and infrastructure planning.
The transition to electric mobility is a cornerstone of global efforts to achieve carbon neutrality. As governments worldwide push for the widespread adoption of EVs, the strain on power grids is becoming increasingly apparent. Accurate forecasting of when and where EVs will charge is essential for ensuring grid stability, optimizing the placement of charging stations, and preventing overloads. Traditional models have often fallen short, failing to account for the complex interplay of human behavior, urban infrastructure, and traffic networks. The new model developed by the Southeast University team aims to bridge this gap by creating a more holistic and realistic simulation of EV usage.
At the heart of this research is the concept of the “travel chain,” a sequence of trips made by an individual throughout the day, starting and ending at a home base, typically a residential area. By modeling these travel chains, the researchers can simulate the movement of EVs across a city, capturing the dynamic nature of their journeys. This is a significant departure from simpler models that might only consider a single trip from home to work and back. The inclusion of multiple trips, with varying destinations such as workplaces, commercial centers, and other locations, allows for a much more nuanced understanding of charging behavior. For instance, an EV owner might charge at a fast-charging station during a lunch break at a commercial district, or upon returning home in the evening, depending on their remaining battery level and the distance of their next planned trip.
To bring this travel chain model to life, the researchers incorporated detailed geographic information. They divided a target city into 191 functional zones, including residential, work, commercial, and other areas. This zoning is crucial, as it reflects the real-world layout of urban environments and the different charging opportunities available in each type of area. Residential zones are typically equipped with slower, overnight charging options, while commercial and work zones often feature faster chargers to accommodate shorter parking durations. By assigning different charging characteristics to each zone, the model can simulate the decision-making process of an EV driver, who must weigh the urgency of their need to charge against the availability and speed of charging infrastructure at their current location.
A key technical innovation in this study is the use of the Floyd algorithm for path planning. This algorithm efficiently calculates the shortest path between any two points in a complex road network. In the context of this research, it allows the simulation to determine the most likely route an EV will take between destinations, factoring in the actual distances and travel times. This integration of real road network data ensures that the simulated travel times and energy consumption are highly accurate. The model calculates the energy expended on each leg of a journey based on the distance traveled and the vehicle’s average speed, which is influenced by the type of road and traffic conditions. This precise calculation of energy use is fundamental to predicting when an EV’s state of charge (SOC) will drop to a level that necessitates charging.
The prediction process is a multi-step simulation. For each EV in the model, the system first generates a daily travel pattern, including the number of trips and the number of destinations visited per trip. This is based on statistical distributions derived from real-world travel surveys. The initial departure time for the first trip of the day is also randomly generated, following a normal distribution that reflects typical morning routines. The sequence of destinations is determined using a Markov state transition matrix, a mathematical tool that captures the probability of moving from one type of zone to another. For example, the model knows that after leaving a residential zone, a user is more likely to go to a work zone during a weekday, while on a weekend, a trip to a commercial zone is more probable. This probabilistic approach injects a high degree of realism into the simulation, reflecting the habitual yet somewhat random nature of human travel.
Once the travel chain is established, the model uses the Floyd algorithm to map out the precise route between each consecutive destination. It then calculates the travel time and the corresponding energy consumption for each segment of the journey. Upon arrival at a destination, the model draws a random parking duration from a probability distribution that is specific to the type of zone. For instance, parking times in residential areas follow a Weibull distribution, reflecting longer, overnight stays, while parking in commercial areas follows a generalized extreme value distribution, accounting for shorter, more variable durations. This step is vital, as the length of a parking session directly influences whether there is sufficient time to initiate a charging session.
The core of the model is the charging decision logic. When an EV arrives at a destination, the simulation checks its remaining SOC. If the SOC is below a critical threshold, typically set at 25%, a charging event is triggered. Even if the SOC is above this threshold, the model will initiate a charge if the remaining energy is insufficient to reach the next destination in the chain. This forward-looking logic mirrors real-world driver behavior, where individuals plan their charging to ensure they won’t be stranded. The duration of the charging session is then determined by the available parking time and the charging power of the station in that zone. The model differentiates between slow charging in residential areas and fast charging in commercial and work zones, using a detailed charging power profile that reflects the two-stage charging process of a typical lithium iron phosphate battery.
The researchers applied their model to a real city, simulating a fleet of 100,000 EVs distributed across its eight administrative districts. The results revealed a distinct “double peak” pattern in the city’s total charging load. The first peak occurred around 7:00 AM, driven by EVs returning home and beginning to charge after their morning commutes. This was followed by a period of low demand during the midday hours. The second, and larger, peak emerged in the evening, around 10:00 PM, as users returned from work and other activities and plugged in for overnight charging. This evening peak represented a demand that was 1.28 times the daily average, highlighting the significant stress that EV charging can place on the grid during peak hours.
The analysis also uncovered stark differences between functional zones. In residential areas, the load curve showed a pronounced peak during the late evening and night, with a peak-to-trough ratio of 1.65. This aligns perfectly with the common practice of home charging. In contrast, the load in work zones peaked much earlier in the day, around 7:00 AM, as users arrived at their workplaces and took the opportunity to charge. This peak subsided by mid-morning, resulting in a much lower load for the rest of the day. A particularly striking finding was the disparity between residential and work zone loads in the late evening. From 9:00 PM to midnight, the residential charging load was a staggering 134.17% higher than that in work zones, a clear testament to the dominance of home-based charging.
The study further demonstrated that the charging patterns of individual administrative districts are heavily influenced by their internal functional zoning. For example, one district (Area A) was dominated by work zones, and its overall charging load curve closely resembled the pattern of a typical work zone, with an early morning peak. Another district (Area B), which had a majority of residential zones, exhibited a load curve that mirrored the residential pattern, with its highest demand occurring in the evening. This finding is of paramount importance for urban planners and utility companies, as it suggests that the charging infrastructure needs of a district can be predicted by analyzing its land-use composition.
To validate the superiority of their model, the researchers compared its predictions against those of a previous method from the literature. The comparison showed that the new model produced a more accurate representation of the charging load, particularly in the afternoon. The older model underestimated the load because it did not adequately account for the higher proportion of fast charging that occurs in commercial and work zones during the day. Furthermore, the new model predicted a smaller difference between peak and valley loads, a result of its more realistic simulation of multiple daily trips, which spreads out the charging demand more evenly throughout the day compared to models that assume only a single daily commute.
The implications of this research are far-reaching. For power grid operators, this model provides a powerful tool for anticipating and managing peak loads, enabling them to implement demand-response programs or schedule maintenance during low-demand periods. For city planners and charging station operators, the ability to predict the charging demand in specific zones allows for a more strategic and cost-effective deployment of charging infrastructure. The study’s conclusion that charging load is primarily concentrated in residential areas strongly suggests that prioritizing the installation of charging points in homes and residential complexes should be a top priority.
Moreover, the model’s ability to simulate different scenarios makes it invaluable for long-term planning. Policymakers can use it to assess the impact of various incentives, such as time-of-use electricity pricing designed to encourage off-peak charging. By analyzing how different regional types respond to such policies, authorities can tailor their strategies to maximize grid stability and minimize the need for costly grid upgrades. The model also provides a framework for understanding the future impact of autonomous vehicles and shared mobility services, which could drastically alter travel chains and, consequently, charging patterns.
In summary, the research by Shen Xiaoqi and her colleagues at Southeast University represents a significant leap forward in the field of EV load forecasting. By seamlessly integrating travel chain theory, detailed geographic data, and advanced path-planning algorithms, they have created a model that captures the true complexity of human mobility. This work moves beyond simplistic assumptions and provides a robust, data-driven foundation for the sustainable integration of millions of electric vehicles into our power systems. As the world accelerates its shift toward electrified transportation, tools like this will be indispensable for building a resilient, efficient, and clean energy future.
Shen Xiaoqi, Fang Xin, Tan Linlin, Li Xinguo, Sun Jiaqi, Southeast University, Electric Power Engineering Technology, DOI: 10.12158/j.2096-3203.2024.03.014