Prediction Of Taxi Passenger Hot Spots Based On Residual Neural Network

With the acceleration of urbanization in China and the advent of the Internet era,various kinds of online ride-hailing services have emerged in an endless stream,bringing serious traffic problems to cities.However,due to the large population and the limited taxis,there will be road congestion in many areas during rush hour.Therefore,it is necessary to explore an appropriate means to obtain the promptly and accurate space-time distribution hot spots of taxis,so as to provide a foundation for taxi planning and scheduling management.Therefore,this paper demonstrates ARes Net based on the residual neural network and accomplish the visualization system.The reminder of the thesis is structured as follows:(1)To solve the problem of difficulty in model training caused by the disappearance of gradient,using the residual error of the neural network shortcut links,also,introducing standardized processing,strengthening the gradient in the process of back propagation sensitivity.Accordingly,increasing the depth of the neural network,which can capture the spatial correlation between the far area,avoiding complex due to the design of multi-layer convolution training,using multiple channels to learn the internal influence between the passenger volume and the number of cars up and down,and strengthening the model generalization.(2)In terms of the problem about weights allocation for the original characteristics of the channel,by means of incorporating the channel domain attention mechanism in the residual neural network,extracting from the original feature weights,recalibrating the original features and changing their distribution.Then,generating the channel domain attention,strengthening the effective features and suppressing invalid ones,speeding up model convergence and the ability of feature extraction.(3)Regarding the impact of external factors on the model prediction,the two-layer fully connected neural network was constructed to input weather and holiday factors into the network,furthermore,the residual neural network output was aggregated after the fusion with the parameter matrix to obtain the passenger hot spot prediction results.By preprocessing the two data sets,the ARes Net model was compared with CNN,ST-Res Net and Quad-Res Net models on the data sets,and the experimental results showed that the ARes Net model has higher prediction accuracy.(4)The hot spot visualization system for taxi passengers was programmed and visualized by Python,Flask,HTML and Echarts.The historical data of Hai Kou were statistically analyzed at different time intervals,and the predicted results were visualized and analyzed according to proximity,similarity and periodic time interval.