High-precision vehicle engine fuel consumption model based on data augmentation

With the increasing prominence of environmental issues in transportation, reducing energy usage and emissions has emerged as a crucial research direction. However, existing data-centric models for estimating vehicle transient fuel usage often depend heavily on extensive training data. This study suggests a computational method for vehicle engine fuel consumption that addresses the model accuracy problem caused by insufficient original data through data augmentation. The method has three parts. First, the experimental data are preprocessed. Second, a generative adversarial network is employed to create augmented data samples (ADSs). Lastly, the effect of different data augmentation scales on model performance is analyzed. This analysis determines the optimal ADS. The proposed method is tested on four machine learning (ML)-based fuel consumption models using comprehensive statistical metrics. It reduces the mean absolute percentage error from 10 % to 4.3 %. This study improves prediction reliability while lowering data dependency. The approach not only helps adopt advanced ML algorithms in industry but also reduces data collection needs.