[Paper Review] Mind the Gap: Modelling Difference Between Censored and Uncensored Electric Vehicle Charging Demand

Frederik Boe Hüttel, Filipe Rodrigues+1 more|arXiv (Cornell University)|Jan 16, 2023

Electric Vehicles and Infrastructure4 citations

TL;DR

This paper proposes censorship-aware machine learning models to estimate latent electric vehicle (EV) charging demand by accounting for unobserved demand due to occupied charging stations and competing services. Using GPS trajectories from ICE vehicles in Copenhagen, the study shows that up to 61% of demand is censored, and censorship-aware models—particularly censored quantile regression—provide significantly better prediction and uncertainty estimation than traditional models.

ABSTRACT

Electric vehicle charging demand models, with charging records as input, will inherently be biased toward the supply of available chargers. These models often fail to account for demand lost from occupied charging stations and competitors. The lost demand suggests that the actual demand is likely higher than the charging records reflect, i.e., the true demand is latent (unobserved), and the observations are censored. As a result, machine learning models that rely on these observed records for forecasting charging demand may be limited in their application in future infrastructure expansion and supply management, as they do not estimate the true demand for charging. We propose using censorship-aware models to model charging demand to address this limitation. These models incorporate censorship in their loss functions and learn the true latent demand distribution from observed charging records. We study how occupied charging stations and competing services censor demand using GPS trajectories from cars in Copenhagen, Denmark. We find that censorship occurs up to $61\%$ of the time in some areas of the city. We use the observed charging demand from our study to estimate the true demand and find that censorship-aware models provide better prediction and uncertainty estimation of actual demand than censorship-unaware models. We suggest that future charging models based on charging records should account for censoring to expand the application areas of machine learning models in supply management and infrastructure expansion.

Motivation & Objective

• To address the bias in EV charging demand models that arise from censored observations due to limited charging station capacity and competition.
• To demonstrate that observed charging records underestimate true latent demand, especially in high-occupancy or competitive environments.
• To develop and validate censorship-aware models that learn the true demand distribution from censored data.
• To evaluate the performance of censorship-aware models in comparison to standard models under realistic counterfactual scenarios.
• To support more accurate infrastructure planning and operational decisions by estimating the full spectrum of unobserved demand.

Proposed method

• Used GPS trajectories from internal combustion engine (ICE) vehicles in Copenhagen as a proxy for EV user behavior, assuming these vehicles were electric.
• Simulated EV charging demand by modeling battery depletion and charging behavior under different assumptions of charging station availability and competition.
• Applied censorship-aware models, including the Tobit model and censored quantile regression, which incorporate censoring into the loss function to estimate latent demand.
• Constructed a counterfactual scenario where the same trips are re-evaluated under varying levels of station occupancy and market share of competing charging providers.
• Calibrated models using observed charging records while explicitly modeling the upper bounds of demand due to station capacity and competitor presence.
• Validated model performance using metrics like fit and uncertainty estimation, comparing censorship-aware and censorship-unaware models.

Figure 1: The figure shows the two instances of censoring of electric vehicle (EV) charging demand Left: The demand is censored due to lost opportunities, where EV drivers can not charge due to occupied charging stations. Right: The demand is censored due to competing services, where EV drivers can

Experimental results

Research questions

• RQ1 To what extent is EV charging demand censored due to occupied charging stations and competing services in urban environments like Copenhagen?
• RQ2 How do censorship-aware models compare to traditional machine learning models in predicting latent charging demand and estimating uncertainty?
• RQ3 What is the impact of censoring on the accuracy of demand forecasts for infrastructure planning and operational decision-making?
• RQ4 How does the performance of censorship-aware models vary under different market share scenarios for competing charging providers?
• RQ5 Can GPS trajectories from non-EVs serve as a reliable proxy for estimating latent EV charging demand in the absence of EV-specific data?

Key findings

• Up to 61% of potential EV charging demand is censored in certain areas of Copenhagen due to occupied charging stations and competition.
• Censorship-aware models, particularly censored quantile regression, provide a better fit and more reliable uncertainty estimation of latent demand than the Tobit model.
• Censorship-aware models significantly outperform censorship-unaware models in predicting true demand, especially in competitive or high-occupancy scenarios.
• The study demonstrates that ignoring censoring leads to systematic underestimation of actual demand, limiting the utility of standard models for infrastructure planning.
• Using ICE vehicle GPS trajectories as a proxy for EV behavior enables realistic counterfactual simulations of demand censorship without requiring actual EV data.
• The results suggest that future EV charging demand models should incorporate censoring mechanisms to improve accuracy and support strategic infrastructure expansion.

Figure 2: Illustration of the experimental procedure. Firstly, the GPS trajectories of Internal Combustion Engine (ICE) vehicles are modelled as EVs using Algorithm 1 (for detailed information, refer to Appendix A ). Subsequently, the observed charging demand and the true demand are derived from the