Data Driven Modeling And Analyses For New Energy Vehicles Based Profile

Profile modeling is widely used in various scenarios,such as user accounts analysis on social networks,customer behaviors analysis on shopping platforms,user browsing histories or hobbies on other platforms and so on.While they are mainly focused on human subject,the aim is for personalized customization or indivisualized advertisement.When it comes to profile modeling itself,the superiority is to analyze multi-dimensional features comprehensively for a full-scale perception.The increasing popularity of new energy vehicles（EVs）appeals to researcheres to carry on more and more studies.Exsisting works mainly concentrate on a certain aspect of EVs,such as architecture or dynamical system,battery usage and driving behaviors,which is short of systematic and comprehensive analysis.What’s more,many researches on car data processing and analyzing perform not so well due to lack of effective data managements as well as efficient algorithms or models.In order to solve the problems mentioned above,based on the concept of profile modeling,and in real data-driven scenarios,this thesis presents a profile modeling method whose subject is new energy vehicles.It proposes effective data managements methods from systematic and comprehensive analyses in three aspects of EVs.In addition,by designing and constructing efficient machine learning or deep learning models,it can achive favorable results.Specifically,the main contributions of this thesis are as follows:· Present a profile modeling method to analyze new energy vehicles systematically and comprehensively.· First from charging stations which EVs live by,construct XGBOOST based stations use rate prediction model by effective data managements methods and obtain relationships among use rate,stations internal features and surroundings,which can offer advice on charging stations site layout.· Then analyze energy consumption of EVs,propose an analyzing method based on runtime,called trip,and obtain relevant conclusions about relationships between energy consumption and traveling miles as well as traveling time of pure electirc cars and hybrid cars.· Emphasise on driving behaviors and safety of EVs,move forward to propose an analyzing method based on more precise runtime,called interval,for specific drving behavior analysis.Construct LSTM + CNN network for multidimensional car features embedding and prototype learning framework for driving safety prediction,which can recognize danger behaviors and influential factors in runtime.