Research On Short-term Traffic Flow Prediction Freeway Under Different Weather Conditions

Short-time freeway traffic flow forecasting has become a popular topic in recent years as an important basis for short-term operational adjustments by freeway management.However,most of the existing highway traffic flow forecasts only consider the normal environment,and less consider the influence of weather conditions on the short-time highway traffic flow,but in relatively closed highways,weather changes are more likely to cause traffic interruption or increased delays,so it is especially important to carry out research on short-time highway traffic flow forecasting under different weather conditions.In this paper,we firstly compiled and summarized the traffic flow data of a highway in G province,accurately crawled the historical weather data of each toll station using Python,and completed the pre-processing and spatio-temporal matching study of the data to provide accurate data support for the subsequent work;secondly,we tested the normality of the traffic flow data,selected the Spearman correlation coefficient for the spatio-temporal correlation analysis of the traffic flow data,further studied the correlation between different Then,based on the convolutional neural network(CNN)and long and short-term memory(LSTM),we introduced the attention mechanism to construct a CNN-LSTM-Attention model that can adaptively capture the dynamic relevance of the input,and used the control variable method to adjust the model parameters;finally,by comparing different models,the prediction efficiency of the proposed model is verified,and the model is used to compare the prediction of traffic flow under different weather conditions with or without integration,and the traffic flow prediction under single weather conditions is deeply explored to complete the study of the applicability of the prediction model under different seasons.The proposed prediction model is trained and validated by the toll data of a highway in G province,and the results show that:CNN-LSTM-Attention has a better prediction effect,and the R~2 is improved from 0.8945 to 0.9645 compared with the ordinary CNN-LSTM model;comparing the prediction results of models with and without fused weather factors,it is found that the R~2 improved from 0.9567 to 0.9645 after fusing weather factors.In the prediction of traffic flow by fusing individual weather factors,the model with fused temperature T has the highest prediction accuracy,and the evaluation index R~2 is as high as 0.9856,which strongly verifies the correlation between temperature and traffic flow.In the prediction of traffic flow under different seasons,the model has good prediction accuracy(R~2 is greater than 0.96).The research results verify the effectiveness and universality of the proposed short-time traffic flow prediction model under different weather conditions,and lay a theoretical foundation for realizing fast,accurate and efficient traffic guidance and regulation.