Traffic Accident Prediction Based On Improved Long Short Term Memory Neural Network And Convolutional Neural Network

According to data from the World Health Organization,road accidents are responsible for up to 1.35 million deaths and around 50 million injuries worldwide.Therefore,with the rapid development of road transport,it is not possible to overlook the improvement in the level of road safety.Predicting traffic accidents has become an vital part of traffic accident prevention,which can help decision-makers understand the development trend and change the characteristics of traffic accidents.It provides a reasonable basis for formulating targeted prevention measures,traffic laws and regulations,and reasonable traffic safety management objectives,which are of great significance for reducing property losses and protecting personal safety.Road traffic accident data has the characteristics of nonlinear and randomness,and the neural network has stronger nonlinear mapping ability,robustness,and strong self-learning ability,which is suitable for processing complex data.As a result,experts and researchers have started to apply the deep learning approach to the prediction of accidents.Based on this,this study is based on a area’s road traffic accident data,,respectively predicts the number of traffic accidents from the macro level and individual perspective,and makes an overall evaluation of the internal causes of traffic accidents.The aim is to predict the possibility of traffic accidents,understand the internal action mechanism,provide theoretical support for road safety agencies and policymakers,and proposed some measures to reduce the number of deaths and economic losses resulting from road accidents.The main contents of this thesis are as follows:(1)Optimization of traffic accident number forecast model based on EEMD and PSO(PSO).On the basis of a single model,the noise reduction algorithm EEMD is introduced to reduce the noise of traffic accident time series.The LSTM network structure is optimized by using PSO,and the time characteristic information of data is extracted under the optimal network structure of LSTM for prediction.Firstly,the model uses EEMD to decompose the accident time series to get several sub-sequences and a residual term,which are respectively input into the LSTM model network optimized by PSO algorithm for prediction.The final prediction result is obtained by summing the prediction results of each sub-sequence and residual.(2)In this paper,a new model is proposed,which combines channel attention and convolutional neural networks,which are based on the previous convolutional neural networks.The channel attention module is used to optimize the weight of the data,distinguish the influential factors of different importance degrees from the important features of different traffic accidents.Moreover,different weights are given to the features with different significance levels,so that the precision of predicting the severity of traffic accidents can be increased.(3)Based on the multivariate logit model,the main factors that influence the severity of road traffic accidents and the intrinsic mechanism of these factors are identified.And the road safety agencies and policy makers are given clearer ideas to actively take targeted measures to reduce the harm caused by traffic accidents.