Battery Electric Truck Infrastructure Co-design via Joint Optimization and Agent-based Simulation

Abstract

As zero-emission zones emerge in European cities, fleet operators are shifting to electric vehicles. To maintain their current operations, a clear understanding of the charging infrastructure required and its relationship to existing power grid limitations is needed. This study presents an optimization frame-work for jointly designing charging infrastructure and schedules within a logistics distribution network, validated through agent-based simulations. We formulate the problem as a mixed-integer linear program and develop an agent-based model to evaluate various designs and operations under stochastic conditions. Our experiments compare rule-based and optimized strategies in a case study of the Netherlands. Results show that current commercial solutions suffice for middle-mile logistics, with central co-design yielding average cost reductions of 5.2% to 6.4% and an average 20.1% decrease in total installed power. While rule-based control effectively manages charging operations and mitigates delays, optimizing charge scheduling significantly reduces queuing times (99%), charging costs (13.5%), and time spent near capacity (10.9%). Our optimization-simulation framework paves the way for combining optimized infrastructure planning and realistic fleet operations in digital-twin environments.

Figures (8)

• Figure 1: A network sketch where nodes represent retailers connected to a central distribution center (DC). This paper deals with the optimal location and sizing of charging stations (CS) for battery-electric truck (BET) operations.
• Figure 2: Flowchart depicting the overarching methodology proposed. Input data on existing operations is given for each vehicle trip as the set $\mathcal{L}^k$. Then, through the optimization module, the optimal design variables $X_i^r$ (quantity, location and power rating of chargers) and $Y^{r,t}_{k,l}$ (charging schedule of each truck) are obtained. Alternatively, the optimization module may be bypassed and rule-based decision making is applied to obtain the charging infrastructure and charging schedules are defined in-simulation.
• Figure 3: Schematic of the optimization problem solved and its main decision variables: $k$ BETs have pre-specified delivery tours comprised of multiple legs $l_k$ that span between locations (DCs or retailers) denoted by $i$ (origin) and $j$ (destination). The number of chargers of type $r$ available at each location $i$ is given by the integer decision variable $X^{r}_{i}$. BETs have the decision to charge before each trip leg $l_k$, at a given time step $t$, at a charger of type $r$. This is modelled with a binary decision variable defined as $Y^{k,l}_{r,t}$. These decisions add up to a final power requirements curve denoted as $P(t)$. For simplicity's sake, in this diagram chargers are only installed at the DC (i.e: $X^r_{i\neq1}=0$).
• Figure 4: Unified modeling language class diagram for the agent-based simulation model. Four main collections are given to the model: the truck itineraries, the truck charging schedules, the fleet characteristics and the distribution center characteristics. The environment then initializes the DC agents and truck agents in their geographical location and the metrics resulting from the interactions are recorded.
• Figure 5: Activity diagram for a truck $k$. This statechart can be classified in three operational states: Charging, Loading/unloading and Transport. In the case of the optimized schedule testing, the charging time and power is given in a separate charging schedule. In the rule-based case, the charging is performed depending on the instantaneous battery level.