基于深度学习的生态驾驶行为判别模型及矫正方法研究

Eco-driving is a primary countermeasure for solving the fuel consumption and gas emission problem. To establish the ‘classification-diagnosis-optimization’ process of driving behavior eco-characteristics, it is critical to uncover the hidden-relationships between exterior factors of traffic system, driving behavior and fuel consumption and gas emission. Therefore, in order to optimize eco-driving behavior, the key of this study is to build the driving behavior eco-characteristics identification and diagnosis model based on Deep Belief Networks (DBNs), through mining multi-dimensions and multi-causal eco-driving behavior recessive characteristics, on the basis of quantitatively analysis of big data of micro driving behavior. This study mainly aims to: 1) plot the specific chromatogram expression of the time-space evolving pattern of eco-driving; 2) establish the DBNs identification model, optimizing the critical parameters of behavior characteristics, and reaching the precise diagnosis of eco-driving behavior, by the means of the causality of input characteristics variables, network structure parameters and results of eco-driving identification; 3) build individualized correction and optimization model, and finally form an optimization theory and methodology, considering the whole process of identification-diagnosis-optimization of eco-driving behavior and the differences of drivers’ individual value and target orientation. This theory and methodology will lay the foundation of the popularization and application of eco-driving, and further deductively explore a method system of micro-driving behavior delicacy management, facing safe transport, green transport and smooth transport in big data era.

生态驾驶行为是解决交通领域能耗排放问题的主要手段，驾驶行为生态性“判别-诊断-优化”是解决问题的关键，挖掘交通系统外部因素、驾驶行为与能耗排放的隐性关联是问题的核心。课题以生态驾驶行为优化为目的，以微观驾驶行为大数据为背景，以数据驱动为导向，以多维度、多致因生态驾驶行为隐性特征挖掘为特点，以基于深度信念网络（DBNs）的驾驶行为生态特性判别诊断为核心，重点研究生态驾驶行为时、空演变规律下的特征图谱表达；构建生态驾驶行为DBNs判别模型，借助输入特征变量、网络结构参数、生态判别结果间的因果关系，优化行为特征关键参数，实现生态驾驶行为精确诊断；考虑驾驶员个体价值和目标取向差异，构建个性化矫正优化模式，最终形成涵盖生态驾驶行为判别、诊断、矫正全过程的面向个体特性的优化理论及方法。为生态驾驶行为的推广应用奠定基础，进而推演形成大数据时代面向交通安全、绿色、顺畅的微观驾驶行为精细化管理的方法体系。

生态驾驶行为是解决交通领域能耗排放问题的主要手段，驾驶行为生态性“判别-诊断-优化”是解决问题的关键，挖掘交通系统外部因素、驾驶行为与能耗排放的隐性关联是问题的核心。课题以生态驾驶行为优化为目的，以微观驾驶行为大数据为背景，以数据驱动为导向，以多维度、多致因生态驾驶行为隐性特征挖掘为特点，以基于深度信念网络（DBNs）的驾驶行为生态特性判别诊断为核心，重点研究生态驾驶行为时、空演变规律下的特征图谱表达；构建生态驾驶行为DBNs判别模型，借助输入特征变量、网络结构参数、生态判别结果间的因果关系，优化行为特征关键参数，实现生态驾驶行为精确诊断；考虑驾驶员个体价值和目标取向差异，构建个性化矫正优化模式，最终形成涵盖生态驾驶行为判别、诊断、矫正全过程的面向个体特性的优化理论及方法。为生态驾驶行为的推广应用奠定基础，进而推演形成大数据时代面向交通安全、绿色、顺畅的微观驾驶行为精细化管理的方法体系。