Fuel efficiency of dump trucks is affected by real world variables such as vehicle parameters, road conditions, weather parameters, and driver behavior. Predicting fuel consumption per trip using dynamic road condition data can effectively reduce the cost and time associated with on-road testing. This paper proposes new models for predicting fuel consumption of dump trucks in surface mining operations. The models combine locally collected data from dump truck sensors and analyze it to enhance their capabilities. The architectural design consists of two distinct parts, initially based on dual Long-term Short-Term Memories (LSTMs) and dual dense layers of Deep Neural Networks (DNNs). The new hybrid architecture improves the performance of the proposed model compared to other models, especially in terms of accuracy measurement. The MAE, RMSE, MSE and R2 scores indicate high prediction accuracy.

Keywords: LSTM algorithm, DNN, density, prediction, fuel consumption, quarries

Global Positioning System (GPS) data acquisition devices have proven to be useful tools for collecting real-world motion data. The data collected by these devices provide valuable information when studying the vehicle movement parameters. For vehicle modeling, this data is invaluable for analyzing fuel consumption and vehicle performance. The study presents a methodology for developing the driving cycle of special cars, during which the speed profile of a particular type of vehicle is studied, loaded and processed, and noisy data is filtered for the purity of the experiment. The test data for severe operating conditions are analyzed. A city driving cycle has been developed for a special truck concrete mixer truck in the conditions of the city of Tyumen. Estimated fuel economy of the specified vehicle is estimated.

Keywords: driving cycle, fuel efficiency of concrete mixer truck, noisy data, data filtering, GLONASS/GPS