AVERAGE FUEL CONSUMPTION OF HEAVY VEHICLES USING MACHINE LEARNING

ABSTRACT: This paper introduces a novel approach to developing individualized machine learning models for predicting fuel consumption in heavy vehicles. Unlike traditional methods that rely on time-based intervals, our model leverages vehicle travel distance as the primary basis for predictions. We integrate seven key predictors derived from vehicle speed and road grade to construct a highly accurate neural network model. This proposed model offers the flexibility for straightforward development and deployment across individual vehicles within an entire fleet, thereby enabling optimized fuel consumption management. To achieve this, the model's predictors are aggregated over fixed window sizes of distance travelled. Our evaluation of different window sizes demonstrates that a 1 km window effectively predicts fuel consumption, achieving a coefficient of determination of 0.91. Furthermore, it yields a mean absolute peak-to-peak percentage error of less than 4% for diverse routes, including both city and highway driving cycles. This study highlights the potential of distance-based modeling for more precise and practical fuel efficiency insights in heavy vehicle operations. Keywords — Fuel Consumption, Machine Learning, Neural Network, Heavy Vehicles, Vehicle Travel Distance, Fleet Optimization, Predictors