Improving Viability of Electric Taxis by Taxi Service Strategy Optimization: A Big Data Study of New York City

Chien-Ming Tseng, Sid Chi-Kin Chau, Xue LiuIEEE Trans. ITS 2019 · DOI

TL;DR

This paper tackles the adoption barrier of electric taxis by proposing an optimized taxi service strategy using a Markov Decision Process (MDP) model trained on New York City taxi trip data. The research demonstrates that optimized strategies can significantly improve electric taxi profitability, making them competitive with traditional gasoline taxis, especially with sufficient battery capacity.

What Problem Does This Paper Address?

The widespread adoption of electric vehicles (EVs) is crucial for a low-carbon society, and electrifying public transport like taxis offers a significant opportunity. However, electric taxis face several hurdles: limited driving range, long recharging times, scarce charging infrastructure, and the relatively low price of gasoline. These factors deter taxi drivers from switching from conventional internal combustion engine (ICE) taxis. The paper argues that while these EV limitations are real, optimizing taxi service strategies—how drivers navigate, find passengers, and manage charging—can mitigate these issues and improve the economic viability of electric taxis.

Methods

The core of this research is the application of a Markov Decision Process (MDP) to model and optimize taxi service strategies. An MDP is a mathematical framework for decision-making in situations where outcomes are partly random and partly under the control of a decision-maker.

Here's how they applied it:

  • State Definition: The state of an electric taxi is defined by its current time, location (represented by road network junctions), and battery state (charge level).
  • Action Definition: Actions include driving to a neighboring junction and, crucially for EVs, deciding to recharge at a charging station for a specific duration.
  • Transition and Reward: The model incorporates probabilistic transitions between states (e.g., picking up a passenger, driving to a destination) and defines rewards based on net revenue (fare minus energy cost).
  • Optimization: Dynamic programming is used to find the optimal policy that maximizes the expected net revenue for the taxi driver.
  • Parameter Estimation: Key parameters for the MDP, such as passenger pick-up/destination probabilities, driving times, and energy consumption, are estimated from a large dataset of real-world NYC taxi trips.
  • Energy Consumption Model: A black-box model, adapted from previous research on the Nissan Leaf, is used to estimate EV energy consumption, considering driving speed, distance, and auxiliary loads (like HVAC).
  • Driving Behavior: The model also accounts for different driving behaviors (mild, normal, aggressive) and their impact on energy consumption and revenue.

Datasets

The primary dataset used is the New York City (NYC) Taxi Trip Dataset of 2013. This dataset contains detailed information on taxi trips, including:

  • Taxi ID
  • Trip distance and duration
  • Pick-up and drop-off times and GPS locations
  • Information on charging stations in NYC was also utilized.

The paper uses this data to estimate parameters for the MDP model, such as passenger demand, travel times, and energy consumption patterns.

Key Insights and Ideas

  • Strategy Optimization is Key: The most significant insight is that optimizing taxi service strategies can dramatically improve the economic viability of electric taxis, often compensating for their inherent limitations like range and charging time.
  • Battery Capacity Matters: While optimizing strategies helps, sufficient battery capacity (around 50 kWh or more) is crucial for electric taxis to achieve comparable or better net revenues than ICE taxis, especially by minimizing the need for frequent recharging.
  • Charging Strategy Impact: The choice of charging mode (fast vs. mode 3) significantly impacts revenue, with fast charging generally being more beneficial, particularly for smaller battery capacities.
  • Driving Behavior Influence: Aggressive driving significantly increases energy consumption, but its impact on revenue is more pronounced when battery capacity is insufficient to avoid frequent recharging.
  • Gas Price Sensitivity: As gasoline prices rise, electric taxis become increasingly competitive, with higher battery capacities offering a more stable advantage.
  • Data-Driven Decisions: The research highlights the power of using large-scale, real-world taxi trip data to build sophisticated models that can guide drivers towards more profitable operations.

Key Results

  • ICE Taxi Improvement: Optimizing a conventional ICE taxi's strategy can lead to significantly higher net revenues, placing drivers in the top 0.01% compared to those with unoptimized strategies.
  • Electric Taxi Competitiveness: With a 50 kWh battery and fast charging, electric taxis can achieve comparable net revenues to optimized ICE taxis. With 70 kWh batteries, they can even slightly outperform ICE taxis due to lower energy costs.
  • Revenue Potential: Optimized electric taxi drivers with 70 kWh batteries can earn net revenues ranked in the top 0.07% of traditional taxi drivers.
  • Energy Savings: Electric taxis consume significantly less energy than ICE taxis, with 1000 electric taxis potentially saving up to 15 thousand metric tons of CO2 annually compared to their ICE counterparts.
  • Driving Distance: Optimized electric taxis can cover substantial distances (around 242 km without recharging for morning shifts), with energy consumption manageable by 50 kWh batteries.

Limitations and Future Work

  • Data Limitations: The study acknowledges that the taxi trip dataset primarily captures trip information and lacks full mobility data for roaming periods when taxis are empty. This led to an optimistic estimation of net revenue by using the shortest path for unassigned travel.
  • Simplified Model: The MDP model simplifies certain aspects, such as assuming a fixed number of charging stations and not explicitly modeling the queuing delays at these stations.
  • Single Taxi vs. Fleet: While the paper explores multiple taxis, the complexity of coordinating a large fleet and the potential for market saturation (too many taxis in one area) are areas for further investigation.
  • Future Work: The authors suggest extending the model to consider more realistic charging station availability, queuing dynamics, and exploring the integration of autonomous vehicle technology with optimized service strategies. Further research into the impact of real-time traffic prediction and dynamic pricing could also enhance the model.

Summary updated

BibTeX
@article{tseng2019improving,
  author = {Chien-Ming Tseng and Sid Chi-Kin Chau and Xue Liu},
  title = {{Improving Viability of Electric Taxis by Taxi Service Strategy Optimization: A Big Data Study of New York City}},
  journal = {IEEE Trans. Intell. Transp. Syst.},
  year = {2019},
  doi = {10.1109/tits.2018.2839265},
}
Figure