Accessing and constructing driving data to develop fuel consumption forecast model

3rd International Conference on Advances in Energy Resources and Environment Engineering

能源学;生态环境科学

Yamashita, Rei-Jo^1 ; Yao, Hsiu-Hsen^2 ; Hung, Shih-Wei^2 ; Hackman, Acquah^2

Shigun Research Institute Corporation, Ibaraki, Japan^1

Dept. of Computer Science and Engineering, Yuan-Ze University, Taoyuan, Taiwan^2

关键词: Forecasting modeling;    Forecasting models;    Highly-correlated;    K-means cluster analysis;    Mean absolute percentage error;    Neural network model;    Pearson coefficient;    Statistical analysis methods;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/113/1/012217/pdf
DOI  :  10.1088/1755-1315/113/1/012217

学科分类：环境科学（综合）

来源: IOP

【 摘 要 】

In this study, we develop a forecasting models, to estimate fuel consumption based on the driving behavior, in which vehicles and routes are known. First, the driving data are collected via telematics and OBDII. Then, the driving fuel consumption formula is used to calculate the estimate fuel consumption, and driving behavior indicators are generated for analysis. Based on statistical analysis method, the driving fuel consumption forecasting model is constructed. Some field experiment results were done in this study to generate hundreds of driving behavior indicators. Based on data mining approach, the Pearson coefficient correlation analysis is used to filter highly fuel consumption related DBIs. Only highly correlated DBI will be used in the model. These DBIs are divided into four classes: speed class, acceleration class, Left/Right/U-turn class and the other category. We then use K-means cluster analysis to group to the driver class and the route class. Finally, more than 12 aggregate models are generated by those highly correlated DBIs, using the neural network model and regression analysis. Based on Mean Absolute Percentage Error (MAPE) to evaluate from the developed AMs. The best MAPE values among these AM is below 5%.