Research On Vehicle Travel Time Prediction Based On Hidden Markov Model

Intelligence Traffic System (ITS) is applied in urban traffic to dredge road network. This system has profound influence in the traffic control, vehicle management and traffic flow analysis. The main target of construct Intelligence Traffic System is how to build road network and monitoring the vehicles on the road. And vehicle travel time prediction is one of the main issues in the traffic problem. However, most travel time prediction research is based from the angle of the whole traffic flow and the short time traffic system. The study of the single vehicle travel time prediction is not well addressed.As the topic of big data is becoming a hot research theme in the world, at the same time, traffic data as a source of big data has also been widespread concern. Traffic data from road vehicles tend to have relevance, heterogeneity, uncertainty and time sequence and other characteristics, but there are few vehicles travel time prediction method is specifically designed for traffic data. In addition, the road characteristic data reflection has not been taken into account. Therefore, this article focuses on the bicycle in a short time traffic travel time prediction, prediction model based on the characteristics and road traffic data, integrated forecasting method based on historical data and real-time data to solve problems in real-time forecasting cycling road travel time.In this thesis, we present a hidden Markov model based on cluster using the temporal and relevance criteria, this prediction model effectively solve the discrete distance measure problem of the traffic time series data. By combining the entropy of information in the information theory, we can improve the initial time series data cluster method and realized the unification of the temporal and relevance character of the traffic data. To each criteria, we can formalize a distance measure method, in the meanwhile, to reinforce the robustness of the algorithm, an adaptive clustering method (TSCTSC) is exploited to compute the distance so that through the clustering stage the number of cluster can be adaptively secured. As to the problem of finding the best state based on the observation data on lines, an improved Viterbi algorithm is designed to iteratively find the best probability of the model. On the basis of it, to ensure the cluster based hidden Markov model can also deal the multi-step prediction, an iteration refinement method is applied in the model, and an mathematical proof has been given. The experimental results show that, the proposed algorithm can effectively predict the data flow, and improve the multi-step prediction, and also can solve the discrete problem of the distance measurement.However, the cluster based Markov model is formalized in historical data, with the improvement of the traffic data collection technology, the accuracy and quickness of the data collection is largely increased. Hence the need for the research of the real-time traffic data algorithm. As the off-line model is time expensive and hidden Markov model has some limitation on the state number, these all decrease the efficiency of the static prediction algorithm. In this paper, we proposed an online Non-parameter Hidden Markov model, utilizing the non-parameter method to extend the hidden Markov model. Therefore, the finite state model transformed to an infinite state model. To accelerate the learning speed of the model parameter and hyper-parameter, an original truncate sample method is proposed. By introducing an auxiliary variable, the infinite sample sequence can be truncated into a finite state, so that the sample can be conducted in a more effective fashion. This algorithm make the training parameter of the model converge faster with less time consumption. The experimental results on authentic dataset, public traffic data and vehicle data show that this algorithm takes the least time consuming and performs well and reliably in the accuracy and the real-time performance aspects on the hidden Markov model.To realize the single vehicle travel time prediction, on the basis of the cluster based hidden Markov model and online Non-parameter Hidden Markov model, in the thesis we construct a model combining the online and offline algorithm. After analyze the situation of the road traffic, we proposed an N order road network model. It is built on the original and destination of the road include managing the correlation between the front and rear sections of the road. It iteratively computed the intersection of the road from a first order neighbor road and finally gained the N order neighbor road set. These operations will solve the relevance problem of the road and increase the controllability of the model. Based on the road network model, the hidden Markov model can be transform to a multi-state reliance Matrix, it is more suitable than the traditional first order hidden Markov model. To update the traffic model in a real-time circumstance, an adaptively update clustering algorithm is deployed. The experiment results demonstrate that, the algorithm is more accurate in the environment of the single vehicle travel time prediction from the original to the destination. And also can be effectively applied to the real-time traffic prediction in complex conditions.