Safety Benefit Evaluation Of Intelligent Driving Systems Based On Multisource Data Mining

The rapid development of intelligent driving technologies has attracted much attention.In order to evaluate their effectiveness in reducing the injury caused by traffic accidents,it requires the integration of test scenarios extraction,identification and modeling of the objects to be evaluated,and estimation of occupant injury.Aiming at the problems of limited test scenarios,unknown systems to be evaluated,serious differentiation in the control logic,and lack of Chinese occupant injury model,this thesis presents a safety benefit evaluation methodology of intelligent driving systems based on multi-source data mining.These data are obtained from vehicle operating,accident database and field operational tests.In order to assess different versions of intelligent driving systems,this methodody rationally adopts system identification,estimation of occupant injury and modeling of test scenarios through commercial softwares.It can obtain the advantages of existing methods,such as nonpublic and fair test scenarios,extendibility,and repeatability of the results.The key technologies,including system identification and evaluation index extraction,are studied in this dissertation to propose this evaluation methodody.In order to realize the key parameters identification of vehicles with common state signals under the steady region,an estimation method based on the frequency response characteristics of and vehicle dynamics is proposed.The relationships between characteristic points in frequency domain of vehich subsystems and key parameters are obtained by analyzing the motor-mechanical-tire coupled dynamic systems.This approach estimates parameters through frequency domain identification techniques based on these relationships.Following advantages of this method can be achieved: the information of the vehicle speed and tire longitudinal force isnot needed;and the tire slip ratio and side slip angle are not estimated,which makes the method easy to use;insensitivity to signal error and noise.It is difficult to establish a credible occupant injury model due to the limited samples of in-depth studied accidentes in China.The feasibility of vehicle deformation depth as an evaluation index of occupant injury risk is discussed,and an estimation method of occupant injury risk based on vehicle deformation depth is proposed.Based on the obtained EES data from the accident simulation and the relationship between the EES data and the vehicle deformation depth in the accident,the vehicle collision deformation depth is calculated and the occupant injury risk is estimated based on the established injury risk function.The application of this method does not need in-depth studied accident data.It is conducive to the automobile safety system evaluation regulations in our national conditions.Simulation and experiment platforms are built to realize the proposed safety benefit evaluation methodology.The validity of the key technologies is verified by these platforms under a variety of conditions.A comparision is made between the benefit results of intelligent driving technologies output from the built evaluation platform and the literature.It is shown that the novel methodology can achieve high assessment accuracy and extensive applicability in the technical divergence trend by intergrating the characteristic of current approaches.