Intelligent Decision-Making Optimization Model of a Traffic Emergency Based on Learning Bayesian Network

According to the incomplete and uncertain factors contained in the real-time information of traffic incidents, the automatic optimization model of incident response plan was developed based on loop learning Bayesian Network (BN) to support for the intelligent decision-making of network traffic emergency management. The first layer (structure learning) adopted the combination of expert knowledge and the K2 algorithm data training method, the second layer (parameter learning) adopted maximum likelihood estimate of the network parameter optimization methods. Firstly, item data were extracted to transform 19 discrete variables with different factors levels. Secondly, consulting some experts in the field of freeway incident management was applied to representing all the key factors that affect the successful operation of incident. Thirdly, this initial model was built with the method of GTT-K2 (Greedy Thick Thinning). Finally, BNs was optimized and modified according to expert knowledge. The core of the proposed model is a loop learning Bayesian network that quantifies the causal dependencies between incident information and response decisions. The model was validated using incidents data to indicate that the proposed algorithm is reliable and accurate.