Subway Passenger Flow Prediction Based On TCSET-seq2seq Model

Subway passenger flow forecast refers to the forecast of future passenger transport demand,which provides a scientific basis for passenger transport planning.If we rely on inaccurate and unscientific subjective judgment,we will often have problems such as wrong judgment and resource scheduling lag.However,the traditional urban traffic flow prediction model does not consider the similarity between the input data and the output data,so the model is often disturbed;It is often difficult to integrate the subway characteristics into the prediction data,and the importance of the data is not considered in the data processing method.Therefore,the model is weak,and it takes a lot of time to identify the "pros and cons" of the data.In order to solve the problem of similar correlation between input data and output data,this thesis calculates the similarity from the characteristics,analyzes different characteristics such as weather,rest day attributes one by one,proposes a model for selecting similar days of subway passenger flow based on neural network,and selects several most similar passenger flow data when inputting the model to avoid excessive interference to the model.In order to solve the problem that it is difficult for the model to incorporate subway features,this thesis proposes a new convolution method,that is,the convolution kernel is dynamically adjusted by features to enable the model to combine subway features and enhance the flexibility of the model in passenger flow prediction.At the same time,in order to avoid model swelling,the convolution method adds expansion operation,so that the model can obtain a large receptive field under a lower hidden layer.In order to solve the problem of unstable model prediction,this thesis improves the extrusion incentive module,combines the data average characteristics with the similarity characteristics of subway passenger flow similar days to select the model calculation,fuses the data of multiple time periods,and enhances the robustness of the model.In order to extract the timespace characteristics of subway passenger flow,this thesis adds graph convolution network to the model to ensure that the model has higher accuracy.the experiment shows that the accuracy of the subway passenger flow similar day selection model proposed in this thesis is about 5% higher than that of the traditional similar day selection method in similar day selection problems;the subway passenger flow prediction model based on time series convolution and squeeze excitation is about 7% lower in average absolute error,10% lower in root mean square error,and1.5% lower in average absolute percentage error than the traditional sequence pair sequence model when facing the problem of predicting subway passenger flow.