Fine-Grained Fuel Consumption Prediction

Abstract

The high costs and pollutant emissions of vehicles have raised the demand for reducing fuel consumption globally. The idea is to improve the operations of vehicles without losing the output power such that the engine speed and torque work with the minimum fuel consumption rate. It relies on the complete map of engine speed and torque to fuel consumption rate, known as the engine universal characteristic map. Unfortunately, such a map is often incomplete (fuel consumption rate not observed under most engine speed and torque combinations) and inconsistent (different fuel consumption rates observed under the same engine speed and torque combination). In this paper, we propose to predict the fine-grained fuel consumption rate of each engine speed and torque combination, by learning a model from the incomplete and inconsistent observation data. A novel FuelNet is designed based on Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs). Deconvolution is employed to predict the incomplete fuel consumption rates, while the discriminator can successfully tolerate the inconsistent fuel consumption rate observations. Experiments show that our FuelNet outperforms the existing approaches in both imputing the incomplete and repairing the inconsistent fuel consumption rates. Remarkably, we deploy the predicted fine-grained fuel consumption rates in a mobile application to assist driving, and show that the fuel consumption can be reduced up to 12.8%.