A Study On Accident Rate Of Freeway Based On ANN Trained By PSO

With the development of economy, freeway has become an important traffic way. However, lots of accidents which happened on freeway are threatening personal safety, lives and property. Therefore, the traffic police department hopes that valuable rules can be extracted from their traffic accident data and valuable advice can be given as their decision support through of the study.As a way of data mining, neural networks has been widely used various fields for data analysis and prediction. PSO is a popular algorithm recently. So it is used as a training algorithm for ANN in this paper, to optimize the performance of network and to overcome the limitations of BP algorithm.The core of this paper is to analyze the features of freeway traffic accidents in terms of affecting factors, based on ANN trained by PSO, with the real freeway traffic accident data, and finally use MATLAB to realize the model. This paper introduces the basic principle of BP network and PSO algorithm. On this basis, it elaborates the way of using PSO into BP network. The MSE of network equals to the fitness in the PSO algorithm, while both the weighs and the bias are mapped to every particle's dimensions. Processing the original data mainly includes selecting sample data and quantification. Design the topology of the neural network according to the research purpose and the characteristics of sample data, and train the network programmed by MATLAB with sample data which have been processed. Then, this model is used to analyze the main affecting factors and finally useful advice is given out.