Research On Operational Vehicle Risk Assessment Model Based On Improved Hidden Markov Chain

With the continuous development of China's industry and manufacturing industry,China has become the country with the largest automobile production and sales,but the road traffic safety problems it faces are also becoming increasingly prominent.At present,road traffic accidents have become one of the most serious problems that threaten people's public safety.The safety management of the team has always been the focus of the company and the society.Most of the fleet management still has the problem of extensive management theory and measures,which cannot meet the refined and intelligent requirements in the vehicle safety management process.In the entire road traffic system,there are many factors that affect traffic safety.Traffic accidents caused by the mutual influence of drivers and motor vehicles are particularly serious.In addition,traditional car insurance is mostly priced based on the price of the vehicle and the occurrence of the accident last year.This method of post-analysis cannot make a good estimate of the risk of flexible vehicles.Vehicles cannot be differentiated and the auto insurance industry is losing money and facing huge risks.Therefore,vehicle management departments and major insurance companies began to study vehicle risks in order to solve such problems.Based on modern big data theory and artificial intelligence technology,this paper proposes a "risk evaluation model for operating vehicles based on improved hidden Markov chain".Firstly,data preprocessing is performed on the vehicle network alarm data,and the market data required for modeling is formed after the irrelevant indicators and outliers are eliminated.According to the impact on the risk of vehicles,the indicators are divided into A-type indicators and B-type indicators to build an indicator system.Secondly,the random forest is used to calculate the transfer matrix between the B-type indicators,the hidden Markov model is used to modify the transfer matrix and the B-to-A-type emission matrix is calculated,and the improved hidden Markov model of the forest is constructed.Then,the data obtained from the model is used for cluster analysis to determine the number of hidden dangers of the vehicle every day,and is corrected by the naive Bayes classifier.Finally,combined with the Heinrich's rule of management,the combined model of vehicle risk assessment is constructed and the vehicle risk score is calculated.The results of the combined model are verified by the association rule method and the original data.The verification results show that the model has high correlation with the risk assessment results of the test vehicles,and the model is considered effective.In this paper,the real-time alarm data of vehicles is used,combined with various algorithms to establish a combined operational vehicle risk assessment model.Through the adaptive adjustment of the model,the correlation coefficient between the revised data and the risk is continuously updated,and the traditional static assessment is converted into the dynamics of risk factors.Intelligent assessment.Transform the traditional vehicle risk after-the-fact analysis into an ex ante analysis.The model is beneficial to the vehicle safety management system to timely monitor the vehicle,which is beneficial for the insurance company to differentiate the different vehicles according to the risk assessment results,and also helps the driver to optimize the driving behavior.It has opened up new avenues for the development of public transportation safety and vehicle insurance industry.There are still several shortcomings in this paper,mainly reflected in two aspects:First,from the dataset point of view,the risk indicators are still not comprehensive enough,and the objective factors such as road conditions,weather and drivers are not taken into account.Second,since the model is in the preliminary exploration stage,only two levels of indicator systems have been established,and the indicators are not divided into multiple levels.There are likely to be multiple levels of relationships between indicators,and further exploration is still needed.