Electric Truck Fleets Get Smarter Charging Forecasts
In a significant leap forward for grid integration of heavy-duty electric vehicles, researchers have developed a novel forecasting model that dramatically improves the accuracy of charging load predictions for electric truck aggregations. This innovation addresses one of the most pressing challenges in the electrification of freight transport: the unpredictable and high-impact power demand these vehicles place on local and regional electricity networks.
Unlike passenger electric vehicles, which typically draw modest power over extended periods, electric heavy-duty trucks—often referred to as electric trucks (ETs)—are characterized by massive battery capacities and extremely high charging rates. A single electric truck can demand as much power as dozens of passenger EVs combined during a fast-charging session. When hundreds of these vehicles operate within a logistics hub or depot, their collective charging behavior can create sudden, steep spikes in electricity demand, potentially destabilizing the grid, increasing peak-load costs, and degrading power quality.
Until now, forecasting models for such fleets have largely borrowed methodologies from passenger EV studies, which fail to account for the unique operational dynamics of freight transport. These include strict delivery schedules, variable cargo weights, complex route planning under time constraints, and the influence of traffic conditions on energy consumption. As a result, existing predictions often fall short in capturing the real-world randomness and volatility of electric truck charging patterns.
A new study published in Electric Power offers a compelling solution. Led by Hang Liu from State Grid Handan Electric Power Supply Company, the research team introduces a load forecasting framework built on a higher-order Markov chain model, explicitly designed to incorporate the logistical realities of electric truck operations. The approach not only refines the temporal prediction of when trucks will plug in but also significantly enhances the precision of estimating the total power draw at any given hour.
At the heart of the model is a sophisticated understanding of the electric truck’s daily workflow. The researchers begin by modeling the delivery network as a graph, with nodes representing distribution centers, customer points, and charging stations. Crucially, they introduce a “soft time window” constraint into the route optimization process. In real-world logistics, customers often specify a preferred delivery window, but early or late arrivals are permissible—albeit at a financial penalty to the carrier. This flexibility is a defining feature of commercial freight that rigid scheduling models ignore. By using a genetic algorithm to minimize a combined cost function of travel expenses and time-window penalties, the team generates realistic, optimized delivery routes for each truck in the fleet.
Once the route is established, the model simulates the vehicle’s journey in detail. It accounts for how traffic congestion—quantified through real-time or historical traffic flow data—affects average speed and, consequently, energy consumption. Different road types (e.g., highways versus urban streets) have distinct energy profiles, and frequent stop-and-go traffic in dense areas can significantly increase power draw per kilometer. By integrating this dynamic energy consumption model, the system can continuously track a truck’s state of charge (SOC) throughout its shift.
The decision to charge is not arbitrary. The model triggers a charging event when a truck’s SOC falls below a critical threshold—set at 20% in the study—or when its remaining energy is insufficient to complete its assigned route. This is a more nuanced trigger than a simple low-battery alert; it’s a prediction based on the vehicle’s immediate operational future. Furthermore, the model respects the logistics schedule. If a truck has a subsequent delivery that must be met, it will only charge enough to reach its next destination on time, rather than charging to full capacity. This prevents unnecessary grid strain and aligns charging behavior with business priorities.
This process yields a precise timeline for each truck’s charging start and end times. Aggregating this data across the entire fleet provides a highly granular forecast of how many trucks will be charging at any given hour of the day. In their validation using real-world data from a bonded logistics park with a fleet of 400 electric trucks, the researchers found their method’s prediction of charging quantity closely matched actual measurements, outperforming a traditional Poisson distribution model by reducing the average error by 8%.
However, knowing the number of charging trucks is only half the battle. The total power load also depends on how much energy each truck needs, which is directly tied to its SOC upon arrival at the charger. This is where the model’s second major innovation comes into play: a “fuzzy two-level discretization” of the SOC.
Traditional models often divide the 0-100% SOC range into a handful of broad bins (e.g., 0-20%, 20-40%, etc.). For a vehicle with a 300 kWh battery, a single bin can represent a massive 60 kWh of potential energy, creating a large margin of error in load prediction. The new model introduces a nested approach. It first creates five primary fuzzy sets that capture the qualitative state of the battery: “very low,” “low,” “normal,” “high,” and “very high.” Within each of these primary sets, it then creates a finer subdivision into n smaller intervals. This double-layered discretization allows the model to capture subtle but critical differences in charging demand. A truck arriving at 18% SOC has a very different charging profile than one at 2%, even though both fall into the same “low” category in a coarse model.
With this high-resolution SOC data, the researchers then apply their higher-order Markov chain. A standard (first-order) Markov model assumes that a truck’s future SOC depends only on its current state. In reality, its trajectory is influenced by its recent history—its SOC over the past several hours. A higher-order model, in this case a second-order one, incorporates this memory. It calculates multi-step transition probabilities, asking not just “Where will this truck’s SOC go from here?” but “Given its SOC two hours ago and its SOC now, where is it most likely to be in the next hour?”
This historical context is vital for capturing the inertia and trends in a fleet’s behavior. For instance, if a large number of trucks have been on long-haul routes for the past few hours, the model can anticipate a wave of deep-discharge arrivals at the depot in the evening, leading to a predictable and substantial load surge. By leveraging this richer historical data, the higher-order Markov chain produces a far more accurate and stable forecast of the aggregate power demand.
The results of their simulation are striking. The model successfully predicted the characteristic daily load curve of the electric truck fleet, with a major peak in the late evening (reaching 8.1 MW in their test case) as trucks returned from their day’s deliveries. When compared against a standard first-order Markov chain model, their higher-order approach reduced the average prediction error by an impressive 22.24%. This level of accuracy held even when they scaled the simulation up to larger fleets of 600 and 800 trucks, demonstrating the model’s robustness.
The implications of this work extend far beyond academic interest. For a utility or a grid operator, a 22% improvement in forecast accuracy is a game-changer. It allows for much more effective planning of generation resources, reduces the need for expensive peaking power plants, and minimizes the risk of local grid overloads. For the logistics company itself, a precise load forecast is the foundation for implementing smart charging strategies. They can negotiate better electricity rates by shifting non-urgent charging to off-peak hours or even participate in demand response programs, earning revenue by curtailing their load during system emergencies.
Moreover, the model’s ability to track not just charging but also the potential for vehicle-to-grid (V2G) services is a forward-looking feature. The paper notes that trucks with a “high” or “very high” SOC and no immediate delivery orders can serve as a distributed energy resource, feeding power back to the grid during peak periods. Accurately forecasting this available capacity is a critical first step toward unlocking this valuable grid-balancing service.
While the current model is a major advancement, the authors acknowledge its limitations. It does not yet account for the long-term degradation of battery health, which is a crucial factor in commercial fleet economics. Future work could integrate battery aging models to create a more holistic operational framework that balances grid needs with vehicle longevity.
In an era where the electrification of heavy transport is accelerating rapidly, this research provides a vital tool for ensuring that this transition is not only environmentally beneficial but also technically and economically viable for the power grid. By grounding a sophisticated mathematical model in the gritty realities of logistics operations, the team has built a bridge between the world of freight transport and power system engineering. Their work paves the way for a future where massive electric truck fleets are not a threat to grid stability, but a predictable, manageable, and even supportive asset in the clean energy ecosystem.
Authors: Hang Liu, Hao Shen, Yong Yang (State Grid Handan Electric Power Supply Company); Ling Ji (Guodian Nanjing Automation Co., Ltd.); Yang Yu (State Key Laboratory of Alternate Electrical Power System With Renewable Energy Sources, North China Electric Power University). Published in: Electric Power, Vol. 57, No. 5, May 2024. DOI: 10.11930/j.issn.1004-9649.202306066.