Research And Application Of Spatio-Temporal Prediction Model Based On Deep Learning In The Field Of Transportation

In recent years,with the continuous advancement of urbanization,the construction of intelligent city has also become a hot spot,in which the spatio-temporal prediction has great practical significance.Accurately predicting the crowd flow in each area of the city plays an important role in preventing the occurrence of public safety events.Accurately predicting the travel demand in each area of the city is of great value for taxi companies to dispatch taxis in advance.According to the characteristics of spatio-temporal data,using deep learning,this paper constructs two spatio-temporal prediction models,and designs and implements the urban regional taxi demand prediction system.The work of this thesis includes the following three aspects:(1)Based on the analysis of spatiotemporal data and aiming at the shortcomings of CNN model in spatio-temporal data mining,the new network structures involution and Se-Net are introduced into the prediction model,a spatio-temporal prediction model ST-Res Invo Net mixed with CNN n model and involution is proposed,and ST-Res Invo Net is applied to urban regional crowd flow prediction and urban regional travel prediction.The comparative experiments of the models are carried out on two real data sets,According to the experimental results,the prediction effect of ST-Res Invo Net proposed in this paper is the best.(2)In view of the deficiency of using fixed size convolution kernel to model spatial dependence in Spatio-Temporal prediction,the dynamic convolution kernel mechanism is introduced into the prediction model,a Spatio-Temporal prediction model ST-DKNet which can dynamically select the size of convolution kernel is proposed,and ST-DKNet is applied to urban regional passenger flow prediction and urban regional travel demand prediction.The comparative experiments of the models are carried out on two real data sets.According to the experimental results,The st-dknet proposed in this paper has the best prediction effect.(3)Taking the Spatial-Temporal prediction model proposed in this thesis as the core,a taxi demand prediction system in urban areas is designed and implemented based on the micro service architecture.: The system realizes the following functions: taxi demand prediction in urban area,collection,processing and storage of trajectory data and weather data,visual display and message delivery of taxi demand prediction results,taxi management and map visual display,model training and deployment,storage and query of training log,platform data collection and visual display.