Traffic Flow Prediction Based On Hidden Markov Model

With the rapid development of urbanization and motorization of road transport carriers.In view of limited land resources,the speed of the road construction is far from meeting the actual traffic demand.The road congestion problems that come with it has become one of the core factors that boycott the development of the city with a higher level.In order to alleviate the congestion of urban traffic conditions,it has an important strategic significance for the seamless connection of urban roads and rail transit networks and the optimization of high-quality traffic monitoring that urban traffic flow prediction in accurate,efficient,stable and real-time can better describe the regional road traffic situation.In view of the fact that people's travel activities drive a large amount of road traffic flow data,which promotes road traffic patterns in different road condition,there is a huge system in existing transportation built by people,cars,roads and environment.Taking into consideration of the volatility and non-linear characteristics of the traffic flow itself,the previous traffic flow prediction research considers only that the characteristic data is relatively simple and the traffic network model is too complicated or simple.So that the ability of generalize the model is weak and the predicted road traffic flow cannot accurately reflect the actual road traffic state and so on defects.With the continuous improvement of big data technologies such as Internet of Things and cloud computing,it provides a guarantee for the construction of urban traffic state pattern recognition and monitoring systems that has access to massive and multi-dimensional sensor traffic data.Based on this,this paper uses regional multi-feature traffic data,adopts semi-supervised learning method in feature engineering stage,and combines features according to information entropy theory decision tree algorithm.An optimized hidden Markov model is proposed for traffic flow prediction and pattern recognition of traffic status.Research on the combination of regional traffic data in Linzi District of Shandong Province,the main work content process is as follows:To begin with,the paper analyzes the current situation of the traffic flow prediction model at home and abroad,and recognizes the essential factors affecting the traffic flow change.According to the objective facts,the actual acquired regional traffic data is marked,and the characteristic engineering operation processing is performed in the experimental process at the same time.Secondly,in order to reduce the potential noise impact of the model algorithm at the input end,a decision tree algorithm is added before the construction of the traffic flow prediction model,and the characteristics of the road traffic state are filtered to perform corresponding fusion.At the last but not least,the optimal state parameters for the original Hidden Markov Model(HMM)are difficult to select and the redundancy of the number of states determined during the parameter training process leads to the problem of over-fitting and weak generalization of the model.Combined with Bayesian information criterion and Akaike information criterion theory,an optimized Hidden Markov Model for traffic flow prediction is proposed.It is verified by actual data that enhances the generalization ability and stability of the model effectively.