The city map (whole road network) is diveded into N x N (in this example N = 256) disjoint equal-sized grids.
Script prepares training data for DeepTravel model:
-
maps historical trajectories to grid, generates historical T path and full G path (grid cell sequences that correspond to GPS points)
Travel path G is represented by a sequence of grids it passed by G = {g1, g2, ..., gn}. Recorded GPS points of the path capture the real trajectory T of G in the form T = {t1, t2, ..., tn}, each GPS point pi = {xi, yi, ti} (latitude xi, longitude yi and time stamp ti.
-
extract and aggregate historical speed and time data as short-term and long-term traffict features
For each grid cell gi speed and time spent within the gi data saved to the 5 minutes time bin in short-term features and in week day bin in long-term features.
-
saves generated G and T path for each trajectory and extracted short-term and ling-term traffic features to the separate files.