How to Identify Patterns of Citywide Dynamic Traffic at a Low Cost? An In-Depth Neural Network Approach with Digital Maps

Figure 2

Our proposed method framework. The input is a series of NRTM with the size of . They are fed to a stacked convolutional autoencoder and transformed into a series of three-dimensional feature vectors. The distances between vectors are measured by cosine similarity. Then, the statistical analysis describes their distribution characteristics.