Enhanced fuel consumption prediction using dimensionality reduction and ensemble learning
研究利用自动编码器降维和极端梯度提升模型，基于真实车辆能量数据预测燃油消耗，达到98%准确率，同时降低计算复杂度。
Accurate fuel consumption prediction is crucial for focusing on pressing environmental concerns, optimising fuel efficiency, and reducing operational costs in the transportation sector. Despite its importance, existing prediction models often struggle with high dimensionality, complexity, and low computational speed, simultaneously achieving high prediction accuracy. To address these gaps, this study explores an approach using Vehicle Energy Data to predict the fuel consumption of Internal Combustion Engine vehicles with enhanced performance. Firstly, the Autoencoder was used on a large real-world vehicle energy dataset to convert into a low-dimensional spaces. Secondly, low-dimensional data was fed to the Extreme Gradient Boosting model to predict fuel consumption accurately. To prove the effectiveness of the proposed model, its performance was compared with state-of-the-art approaches. The results revealed that the proposed model performs better than other state-of-the-art models with 98% accuracy. Further sensitivity analysis disclosed that the proposed model simultaneously reduces the computational complexity and increases the predictive performance of the downstream model. The study demonstrates an effective and innovative method for analysing fuel consumption during real vehicle operating conditions in terms of accuracy and real-time capability. This study can be applied to other high-dimensional data to extract various insights regarding fuel consumption.