Abstract:

An algorithm based on the learning vector quantization(LVQ) neural network with selforganizing feature maps was developed to detect the urban expressway incidents. A concept of detection based on each lane separately was proposed. The input data screening, the input vector mode, the quantity of neural element, and the detection time interval selection were investigated. The highfrequency noise in the input data was filtered using the db3 wavelet denoising technique.The long vehicle flow volume was introduced into the input vector, and the effect of its introduction was studied comparatively. The implement process of the proposed algorithm and the selection the thresholds were studied using the traditional California algorithm as reference of estimation.

Key words: engineering of communications and transportation, urban expressway, incident detection, California algorithm, learning vector quantization(LVQ) neural network