事故多发点鉴别本质及基于BP神经网络的鉴别方法研究

Essence and an Approach to Identify the Hazardous Locations of Urban Arterial Link Based on BP Artificial Neural Network

首先对常规事故多发点鉴别方法进行了分析和评述,并通过实例揭示了多发点的鉴别本质。在此基础上,建立了基于三层BP神经网络的城市干道路段事故多发点鉴别模型。该模型考虑了交通事故7个方面的影响因素,并能将常规鉴别方法不易识别出的多发点鉴别出来。其次,应用哈尔滨市市区430个干道路段上1999年至2004年发生的13764起交通事故数据及关联因素数据,对神经网络的权值和偏置值进行了标定,并应用该模型进行了事故多发路段鉴别。最后,分别应用了事故次数概率分布法、矩阵法和质量控制法对430个路段进行了多发点鉴别,并对鉴别结果进行了对比分析。

The advantages and limitations of each kind of generalize methodologies to identify the hazardous locations were analyzed and evaluated, and the essence to identify the hazardous locations was discussed in detail through the analysis of a practical instance. Furthermore, a model to identify the hazardous locations of urban arterial link based on BP artificial neural network was established. The model involves in seven factors affecting the accidents greatly, and it can identify the prone sections that cannot be identified by other generalize models. Lastly, the weights and biases of the BP network were demarcated according to the data of 13764 accidents occurred on 430 arterial links from 1999 to 2004 and the data of exposures. Hazardous locations among the 430 links were identified by probability method of accident frequency, matrix method, quality control method as well as the BP network method just established, and the comparison between the results of different method was performed.