With the rapid growth of electric vehicle (EV) ownership, the uneven spatiotemporal distribution of charging demand becomes a critical bottleneck for grid stability. Although dynamic pricing serves as a core economic lever for load regulation, existing studies often overlook the non-cooperative gaming effects among charging stations induced by price fluctuations, and therefore fail to quantitatively characterize users’ substitution behaviors across alternative charging options. To address this limitation, this paper proposes a price-signal-guided prediction framework, termed the Residual-Coupled Graph Neural Network (RC-GNN), which employs a decoupling mechanism to separately model routine spatiotemporal demand patterns and price-driven competitive interactions. The framework features a novel dual-stream architecture: the Trend Stream captures periodic dependencies based on physical proximity, while the Competition Stream introduces a Peak-Shift Graph to explicitly reconstruct load transfer pathways among potential competitors, thereby relaxing the constraints of geographic adjacency. Furthermore, a network-wide price impulse response experiment is conducted to quantify cross-price elasticity between regions. The results reveal a latent price gaming mechanism embedded beneath the physical network structure: charging demand does not diffuse uniformly, but instead reallocates toward specific competitive stations following spatiotemporal arbitrage patterns. Moreover, this gaming behavior exhibits pronounced asymmetry, whereby the fleeing effect induced by price increases exceeds the attraction effect of price reductions. Specifically, the cumulative cross-elasticity under price increases reaches 0.13, substantially exceeding the 0.09 observed under price reductions. Overall, this work not only enhances forecasting accuracy, but also provides an interpretable analytical foundation for identifying key competitive clusters and designing coordinated pricing strategies in urban charging networks.
Uncertainty in future demand poses a critical challenge for decision-making, particularly in strategic Emergency Medical Services (EMS) logistics planning and public service resource allocation, as it directly affects decision outcomes. For example, in EMS systems, failing to properly account for demand uncertainty during planning may lead to suboptimal capacity decisions, resulting in insufficient or inefficient resource allocation. To address this challenge, we propose a novel hybrid scenario generation method, called Residual Scenario Generation (RSG), which constructs representative uncertainty scenarios from forecasting residuals while maintaining a compact scenario set. The method is evaluated in the context of strategic-tactical EMS capacity planning using a bi-objective two-stage stochastic programming model that considers ambulance base location and fleet sizing decisions under demand uncertainty, with personnel requirements implicitly determined by the crew composition associated with each vehicle type. The model considers a multi-period planning horizon in which long-term configuration decisions are evaluated while ensuring that long-term demand can be adequately satisfied. Using real data provided by the Portuguese EMS provider (INEM), we compare the RSG method with Monte Carlo, Bootstrap Sampling, and Latin Hypercube Sampling. Our results show that the different scenario generation methods produce comparable planning outcomes, while RSG achieves substantially lower computational effort and stable performance using a considerably smaller number of scenarios. These characteristics make RSG particularly suitable for stochastic planning settings requiring repeated optimization across multiple configurations or sensitivity analyses. Managerial insights are also provided, allowing for better informed decision-making.
The rapid expansion of electric bus fleets has made charging scheduling a critical decision-intelligence problem in urban fleet-energy logistics. Reinforcement learning (RL) is promising for this problem, but reward function design still relies on manual trial-and-error and transfers poorly across scenarios. This paper develops an artificial intelligence and operations research (AI-OR) decision intelligence framework that integrates large language model (LLM)-driven reward shaping with RL-based sequential decision-making. The framework represents reward functions as executable code and iteratively optimizes them through four modules: reward initializer, performance analysis, reward exploration, and rollback. An event-triggered two-stage charging reallocation mechanism enforces the action-level operational constraints through action masking, while the battery safety bound is controlled through readiness conditions, reward shaping, and the acceptance criterion. Experiments on a synthetic scenario and two real-world bus stations in Guangzhou compare three mainstream LLMs under different initialization conditions. Results show that LLM-driven reward iteration outperforms traditional methods across all scenarios, reducing costs by up to 32.6 % in nominal within-day charging cost and by up to 22.5 % after an approximate adjustment for the energy left on board. Ablation and sensitivity analyses validate the contribution of each module and the framework’s adaptability to multiple RL algorithms. Reward-evolution analysis further identifies design principles for improving charging efficiency, supporting valley-period energy shifting, and enhancing fleet-energy logistics resilience.
Operational disturbances in scheduled mobility systems manifest primarily as delays, where events occur later than planned and the resulting lateness can cascade through schedules. While rescheduling theory often explores re-sequencing strategies to recover from delays, real-world scheduled mobility systems, from metros to Connected and Automated Vehicles (CAV), often strictly maintain service order to adhere to infrastructure constraints, preserve network connectivity, and ensure real-time computational feasibility. Motivated by these operational realities, this paper develops a unified sequence-preserving event graph framework. On this event graph, the basic mitigation strategy (BMS) computes the componentwise least feasible timing plan, while the hybrid mitigation strategy (HMS) introduces a system-level recovery–postponement trade-off. Building on this representation, we propose a computationally efficient online mitigation strategy that addresses a twofold uncertainty, accounting for stochasticity in both exogenous disturbances and the system’s internal delay absorption capacities. We prove that the incremental method exactly maintains the current-information timing plan under monotone information updates. For a stylized time-invariant setting, we derive an event-class approximation of persistent delay growth and characterize the stabilization boundary within the fixed-graph sequence-preserving policy class. Numerical experiments on a CAV rhythmic-control network and an urban metro network demonstrate the framework’s versatility in enhancing robustness across both emerging and existing transportation paradigms.
