Instantaneous Vehicle Fuel Consumption Estimation Using Neural Networks

Fuel consumption prediction is vital in improving vehicle fuel economy. The paper describes a method for reducing air pollution caused by road transport by estimating the vehicle's instantaneous fuel consumption using neural networks that processes driving data to make real-time fuel consumption estimations. The data used in the study was collected from a 2010 Toyota Camry HEV runs. An Artificial Neural Network (ANN) model processes GPS position, speed, acceleration, and road conditions to predict fuel consumption. An analysis found that the best ANN type for this task is Long Short-Term Memory (LSTM). It belongs to the Recurrent Neural Networks (RNN) class specifically developed to address the issue of long-term dependencies in time series datasets, making it the best fit for fuel consumption prediction tasks. The model architecture and hyperparameters were tuned using various training methods to converge to the best fuel prediction model regarding accuracy, complexity, and stability. The proposed model showed good performance compared to other approaches described in the literature and had the potential to monitor the in-use fuel consumption of the vehicle accurately.