Traffic Flow Prediction via Spatial Temporal Neural Network "ResLS-C"

The ResLS-C model is proposed in this paper combining with ResNet, LSTM and Deconvolution. The model not only fully captures the spatial and temporal characteristics of traffic flow, but also improves the interpretability and visibility of spatial traffic flow prediction, and carries out spatial analysis by Pearson index. From the simulation results, the accuracy of the model we proposed in this paper is higher than many traditional deep learning methods such as LSTM+CNN, it brings practical and effective results for intelligent transportation.The specific design idea of this paper is as follows: the vehicle trajectory data is transformed into a two-dimensional feature matrix. The overall spatial features of the research area can be extracted from the residual network, which can extract the high-level spatial features and solve the shortcoming of CNN gradient dissipation. LSTM is used to mine the time series characteristics of traffic flow. By deconvolution operation, the feature properties in the hidden layer of convolution are restored to the original space, which makes it more accurate to calculate the loss of the true value and the predicted value. Moreover, it can effectively improve the intuitiveness, interpretability, operability and accuracy of the spatiotemporal prediction model.