The increasing adoption of electric vehicles underscores the urgent need for efficient and reliable charging infrastructure. Fixed charging stations are crucial for meeting surging demand and ensuring convenient access, yet accurately predicting demand remains highly challenging, because charging patterns change in response to where stations are placed, which creates a cyclical dilemma for planning. However, mobile charging stations (MCSs) offer a novel solution by flexibly relocating across urban areas, they can both deliver energy and act as dynamic probes to collect real-time data on charging demand. Existing studies, however, typically assume prior knowledge of demand distributions, which is rarely available in emerging EV markets or where privacy concerns limit data access. This paper proposes a Deep Reinforcement Learning (DRL) approach, formulated as a Partially Observable Markov Decision Process (POMDP), to optimize the relocation of MCSs in conjunction with fixed charging stations, while simultaneously uncovering latent demand patterns. We employ an Advantage Actor–Critic (A2C) algorithm with Long Short-Term Memory (LSTM) networks to capture temporal dependencies and adapt to stochastic demand. A dynamic Mixed-Integer Programming (MIP) model is developed as a benchmark that represents an idealized case with perfect foresight of demand. We compare the DRL agent against this optimization model in two settings: (i) a synthetic toy environment for controlled testing, and (ii) a realistic simulation of the Frederiksberg municipality in Denmark, calibrated with real charging data. The results show that the DRL framework effectively adapts to stochastic demand, outperforms the optimization baseline under uncertainty, and scales efficiently to larger problem instances. Beyond methodological contributions, the findings highlight how MCSs can serve a dual role as infrastructure supplements and as demand-discovery tools, offering valuable insights for data-driven and adaptive EV charging planning.

The bias-variance trade-off is a well-known problem in machine learning that only gets more pronounced the less available data there is. In active learning, where labeled data is scarce or difficult to obtain, neglecting this trade-off can cause inefficient and non-optimal querying, leading to unnecessary data labeling. In this paper, we focus on active learning with Gaussian Processes (GPs). For the GP, the bias-variance trade-off is made by optimization of the two hyperparameters: the length scale and noise-term. Considering that the optimal mode of the joint posterior of the hyperparameters is equivalent to the optimal bias-variance trade-off, we approximate this joint posterior and utilize it to design two new acquisition functions. The first one is a Bayesian variant of Query-by-Committee (B-QBC), and the second is an extension that explicitly minimizes the predictive variance through a Query by Mixture of Gaussian Processes (QB-MGP) formulation. Across six simulators, we empirically show that B-QBC, on average, achieves the best marginal likelihood, whereas QB-MGP achieves the best predictive performance. We show that incorporating the bias-variance trade-off in the acquisition functions mitigates unnecessary and expensive data labeling.

Shared mobility services require accurate demand models for effective service planning. On the one hand, modeling the full probability distribution of demand is advantageous because the entire uncertainty structure preserves valuable information for decision-making. On the other hand, demand is often observed through the usage of the service itself, so that the observations are censored, as they are inherently limited by available supply. Since the 1980s, various works on Censored Quantile Regression models have performed well under such conditions. Further, in the last two decades, several papers have proposed to implement these models flexibly through Neural Networks. However, the models in current works estimate the quantiles individually, thus incurring a computational overhead and ignoring valuable relationships between the quantiles. We address this gap by extending current Censored Quantile Regression models to learn multiple quantiles at once and apply these to synthetic baseline datasets and datasets from two shared mobility providers in the Copenhagen metropolitan area in Denmark. The results show that our extended models yield fewer quantile crossings and less computational overhead without compromising model performance.