A Fault Alarm Discrimination Model For Electric Vehicles Based On Deep Learning

Electric vehicles as an important part of the automotive industry have developed rapidly in the past few years.In order to improve the safety of electric vehicles,national ministries and commissions require enterprises to develop specific safety pre-warning mechanisms,build remote vehicle monitoring platforms,and conduct real-time remote alarm monitoring on whether vehicles currently have safety hazards.The alarm accuracy of the remote vehicle monitoring platform is an important indicator to ensure vehicle safety.Currently,the alarm function of the vehicle remote vehicle monitoring platform uses a threshold detection method,using multi-parameter joint overrun alarm to improve the alarm accuracy.However,the threshold exceeding alarm depends on the quality of the detection signal.Due to the unpredictable driving environment,there are a large number of false alarms in the safety alarm signals generated during the actual vehicle using-situation.Under the current alarm rules,enterprises mainly rely on the manual troubleshooting method to distinguish the true and false alarm signals.According to incomplete statistics,from the manual recognition of alarm signals to vehicle inbound detection,it takes about 24 hours for each alarm signal to be disposed of,and at the same time,the owner of the car has to be paid a certain misuse fee.A large number of false alarm signals have caused a serious workload to the staff,increased the operating costs of enterprises,and also affected the user’s car experience.At the same time,the recognition of false alarms also affects the disposal time of real alarms,causing potential vehicle safety hazards.In order to solve the above problems,this paper presents an alarm discriminant model for electric vehicle based on in-depth learning.The model starts with the vehicle alarm data,analyses the historical driving data and alarm disposal records stored in the remote vehicle monitoring platform,uses the convolution neural network to build the basic models,and through large-scale data training and parameter optimization adjustment,finally obtains the alarm discrimination model VA-VGG.Integrating this model into the enterprise’remote vehicle monitoring platform,the false alarm signal is effectively identified and the human and material resources are saved.This paper mainly completes the following work:Firstly,we build a dataset for model training and evaluation.Five alarm contents and61 national standard data collection items required by current GB/T 32960 national standard regulations are studied,and alarm business is analyzed.Then the data flow structure of alarm data is analyzed and designed,the attribute analysis of related data is carried out,the data cleaning rules are established,and the corresponding data cleaning tools are developed by Python 3.7,Mongo DB and Standard Scaler for data preprocessing,and a feature project is built to generate 2030 available alarm datasets whiche provide a data base for subsequent model building.Secondly,using VGG migration learning,an electric vehicle fault alarm discrimination model VA-VGG is constructed.The basic principles of six models are studied which are Le Net-5,RNN,LSTM,Alex Net,Res Net and VGG.Based on the training model of sample set and test set and the accuracy performance is obtained,three basic models,RNN,Res Net and VGG,are selected to further optimize.At last VGG is selected for the migration learning of the model,and the VA-VGG is generated.The experimental results show that the accuracy of VA-VGG for true and false alarm recognition can reach 81% in the sample set.Finally,the VA-VGG deployment is integrated into the actual engineering application.When the production environment is integrated,the alarm model mainly developed by Python and the Java development business system are integrated through Vertex framework,including call integration between interface logic.In the actual industrial production process,this alarm discriminant model provides support for workers to identify true and false alarms,and semi-automatic alarm recognition is achieved from manual alarm recognition.To a certain extent,it really helps the staff to reduce the workload,and also saves the human and material resources for the enterprise.