NeuralMOVES: A lightweight and microscopic vehicle emission estimation model based on reverse engineering and surrogate learning

TL;DR

MOVES is a trusted but highly complex emission model ill-suited for microscopic real-time applications. The authors introduce a two-step framework: first reverse-engineer MOVES to obtain MOVES_RE, a second-by-second, environment- and vehicle-aware dataset (9.89 GB, 109,367,240 points); then train NeuralMOVES, a lightweight surrogate neural network (~2.4 MB) that reproduces MOVES outputs with high fidelity. NeuralMOVES achieves a global MAPE of 6.013% across 2,296,900 evaluations and is differentiable, enabling gradient-based optimization in control settings. A Model Predictive Control eco-driving use case demonstrates millisecond-scale emissions estimates and environment-dependent trajectory optimization for a car and a bus, highlighting NeuralMOVES’ practical impact for real-time transportation analysis and optimization; the work is open-source, and the methodology provides a framework for reverse-engineering industrial software for transportation scenarios.

Abstract

The transportation sector significantly contributes to greenhouse gas emissions, necessitating accurate emission models to guide mitigation strategies. Despite its field validation and certification, the industry-standard Motor Vehicle Emission Simulator (MOVES) faces challenges related to complexity in usage, high computational demands, and its unsuitability for microscopic real-time applications. To address these limitations, we present NeuralMOVES, a comprehensive suite of high-performance, lightweight surrogate models for vehicle CO2 emissions. Developed based on reverse engineering and Neural Networks, NeuralMOVES achieves a remarkable 6.013% Mean Average Percentage Error relative to MOVES across extensive tests spanning over two million scenarios with diverse trajectories and the factors regarding environments and vehicles. NeuralMOVES is only 2.4 MB, largely condensing the original MOVES and the reverse engineered MOVES into a compact representation, while maintaining high accuracy. Therefore, NeuralMOVES significantly enhances accessibility while maintaining the accuracy of MOVES, simplifying CO2 evaluation for transportation analyses and enabling real-time, microscopic applications across diverse scenarios without reliance on complex software or extensive computational resources. Moreover, this paper provides, for the first time, a framework for reverse engineering industrial-grade software tailored specifically to transportation scenarios, going beyond MOVES. The surrogate models are available at https://github.com/edgar-rs/neuralMOVES.

Figures (6)

• Figure 1: Contribution of the paper.
• Figure 2: Impact of environment and vehicle factors on the instantaneous CO$_2$ emissions extracted by reverse-engineering MOVES. (a) Vehicle age and road grade; (b) Vehicle type and fuel type (Note: no diesel for motorcycle); (c) Temperature and humidity (Note: only 21 combinations are selected as we stated in Section \ref{['sec:DatasetGeneration']}).
• Figure 3: Impact of vehicle dynamics on the instantaneous CO$_2$ emissions extracted by reverse-engineering MOVES, characterized by vehicle types. (a) Motorcycles; (b) Passenger cars; (c) Passenger trucks; (d) Light commercial truck; (e) Transit bus; (f) VT-CPFM PARK2013317.
• Figure 4: Synthetic driving cycles for validating NeuralMOVES. (a) Random speed; (b) Sinusoidal speed; (c) Piecewise speed; (d) Approaching intersection; (e) Eco-driving.
• Figure 5: Emission estimation error of NeuralMOVES against MOVES under various scenarios. (a) Percentage error distribution for all trajectories; (b) Percentage error distribution for the trajectories generated by various strategies; (c) Percentage error distribution for the trajectories generated under various environments and vehicle factors.