Highway traffic big data contains rich traffic flow characteristic information. Due to the obvious differences in the operating characteristics of various vehicles on the highway, namely, heavy vehicles are larger in size, lower in speed, inflexible in the driving, and excessive consumption of road space, which will affect road operation. This paper studies the method of traffic flow state identification considering the impact of different vehicle types based on big data-driven theory. The BPSO-RF-SVM selective ensemble model is constructed for traffic flow state identification. The proposed method first uses fuzzy c-means algorithm to classify the traffic status. Then, random subspace ensemble algorithms are used to divide feature subsets and train base learner SVM, and the BPSO optimization algorithm is established to select classifier. This method is implemented the JingHa highway, in Jilin Province, China to test the efficiency and applicability of this model. Keywords: Heavy vehicle, Traffic state, BPSO-RS-SVM, Selective ensemble learning

CICTP