Abstract

The aim of this study is to develop a model to predict temporal daily collision by integrating of Discrete Wavelet Transform (DWT) and Artificial Neural Network (ANN) algorithms. As a case study, the integrated model was tested on 1097 daily traffic collisions data of Karaj-Qazvin freeway from 2009 to 2013 and the results were compared with the conventional ANN prediction model. In this method, initially, the raw collision data were analyzed, normalized, and classified via Geographical Information System (GIS). Partial Autocorrelation Function (PACF) was also utilized to evaluate the temporal autocorrelation for consecutive existing daily data. The results of this study showed that the proposed integrated DWT-ANN method provided higher predictive accuracy in daily traffic collision than ANN model by increasing coefficient of determination (R2) from 0.66 to 0.82.

摘要

这项研究的目的是通过集成离散小波变换（DWT）和人工神经网络（ANN）算法来开发一种预测每日日常碰撞的模型。作为案例研究，在2009年至2013年间对Karaj-Qazvin高速公路的1097次每日交通碰撞数据进行了测试，并将该结果与常规ANN预测模型进行了比较。在这种方法中，最初，原始碰撞数据通过地理信息系统（GIS）进行了分析，归一化和分类。部分自相关函数（PACF）还用于评估连续的现有每日数据的时间自相关。