| The high fuel consumption and emissions from motor vehicles is always the serious problem hindering the healthy and sustainable development of traffic system in urban areas.Among those influence factors for vehicle fuel using and emissions production,driver behavior is one of the most critical.Thus,it is necessary to improve driver behavior towards energy conservation and emission reduction.As with obvious advantages of low-cost,little risk and most cost-effective,eco-driving has become popular measurements to improve vehicle fuel efficiency and decrease emissions in many developed countries.Besides,eco-driving has been reported to have great potential in reducing vehicle fuel consumption and emissions.In fact,the cores of eco-driving behavior application and promotion are: 1)what is eco-driving behavior;2)how to estimate the eco-characteristics of driver behavior and;3)how to correct and optimize driver behavior.Above them,the essence is to understand the characteristics of eco-driving,reveal the hidden relationship between driver behavior and vehicle fuel consumption and emissions,and improve drivers’ acceptance for eco-driving.However,it is challenging to resolve these issues because of many influence factors,big randomness,high uncertainty and personalization as well as the limitation of traditional statistical analysis techniques(e.g.,classification and regression)in expressing complex relationship.The current bottleneck limited the eco-driving behavior characteristic description and accurate estimation,and then influenced the development of personalized eco-driving behavior optimization.Thus,the effectiveness of eco-driving training was negatively impacted.With the fast development of information technology,microscopic driver behavior data could be collected and integrated,based on some key technologies of on-board intelligent equipment,cloud computing platform,vehicle and road collaboration and etc,which provides multiple,fine-grained and high-accuracy data base for thoroughly analyzing and mining driving behavior.Meanwhile,combining with the advantages of graph in complicated information visualization and machine learning in recessive characteristics mining,it is possible to express the highly nonlinear relationship with data driven orientation,and further contribute to describe and identify eco-driving behavior.Based on these,eco-driving feedback optimization methods facing driver personality would be developed,along with drivers’ individual attributes and various training approaches.Therefore,based on microscopic driver behavior big data and oriented by data driven,this paper aims at improving eco-characteristics of driver behaviors by adopting methods of graph theory and machine learning.The focus of this study is eco-driving behavior description,estimation and feedback.The main contents are as follows:Firstly,depending on driving simulator and vehicle monitoring technologies,driver behavior perception platform aiming at driver performance and vehicle operating collection was established respectively.Furthermore,eco-driving behavior training experience platform based on driving simulator as well as eco-driving behavior dynamic feedback platform based on vehicle monitoring and smart-phone was developed.On one hand,they lay a foundation for basic data collection and aggregation;on the other hand,they provide research tools for studying eco-driving behavior perception,estimation and feedback.Meanwhile,through exploring the relationships between driving behavior and vehicle fuel consumption,the data reliability was verified,which further contributing to study the influence,characteristics,identification and optimization of eco-driving behavior.Secondly,based on the experiment platform and combing three experimental tests of driving simulation,vehicle monitoring and follow-up surveys,the study results indicated that the general training of eco-driving rules was effective in improving driver behavior and thus contributing to reduce vehicle fuel consumption and emissions.Meanwhile,the effectiveness of different training methods was significantly various.Besides,the effectiveness of eco-driving training will be weakened as time going by.Additionally,eco-driving behavior brings benefits in decreasing traffic accident rates and has little impacts on vehicle operating efficiency;the vehicle fuel consumption reduction percentages resulting from eco-driving behavior was much higher than that in travel time increase in the same driving conditions.The study results tested and verified the necessity and feasibility of applying and promoting eco-driving behavior.Thirdly,due to multiple influence factors,strong microscopic and high sensibility of driving behavior,it is necessary to grasp its scope and time change process to describe the characteristics of eco-driving behavior accurately.Taking graph theory,driver performance characteristic graph and vehicle operating characteristic graph under the constraints of vehicle fuel consumption rank was respectively constructed.Based on these,the features of driver performance and vehicle operating status with different fuel consumption were intuitively displayed.In addition,depending on similarity identification and statistical distribution,eco-driving behavior features in the aspects of driver performance and vehicle operating was explored.Furthermore,there is a good consistency in describing eco-driving behavior for both driver performance and vehicle operating graphs.Fourthly,as driving behavior is influenced by many factors,the relationship between driving behavior and vehicle fuel consumption is not obvious.To identify eco-characteristics of driving behavior preciously,it is necessary to explore the hidden relationship between driving behavior and vehicle fuel using.According to eco-driving behavior characteristic graph and combing the method of mathematical statistic analysis,eco-driving behavior characteristic index system from two aspects of driver performance and vehicle operating was established firstly.Then,depending on back-propagation(BP)neural network,two BP models with three level structures towards eco-driving estimation from driver performance and vehicle operating were developed,respectively.The optimal model structure and parameters were obtained by experimental simulation test.As a result,the average prediction accuracy for these two models was 92.89% and 96.89%.Finally,because the demographic among various drivers is apparently different,driver personality must be considered when applying eco-driving behavior optimization to improve effectiveness.According to the principle of social psychology,driver classification method facing individual value and goal orientation was proposed based on driver behavior data in the process of driving.Then,the appropriate training method and feedback contents depending on the demands and preferences of different driver types were studied and designed.Therefore,personalized eco-driving optimization method were developed,which was tested and verified to be more effective in saving fuel compared to general eco-driving training course.In a summary,this research firstly developed an experiment platform for driving behavior data collection and eco-driving behavior studying,and then comprehensively evaluated the influence of eco-driving on decreasing vehicle fuel consumption,emissions,operating safety and efficiency.Based on these,an optimization mode including eco-driving behavior description,estimation and feedback was developed,which lays a foundation for eco-driving behavior application and promotion.Furthermore,the study results also contribute to propose a method system of micro-driving behavior delicacy management facing safe,green and smooth transport in big data era. |
PDF Full Text Request