Spatiotemporal prediction 2-gcn_ LSTM

Reference resources ： https://keras.io/examples/timeseries/timeseries_traffic_forecasting/

Mission ：

Traffic multi-step prediction

Prepare the data

node ：228
Calculate the distance between each node ： 288 * 288
Speed ： 12672 * 288, Time * node

To simplify the description , Only selected 26 Nodes

Training ：6336 * 26
verification ：2534 * 26
test ：3802 * 26

Figure data

Composition ： Adjacency matrix

Model

Input ： batch * time_steps * node * feature
transpose： node * batch * time_steps * feature
GCN： Surrounding message propagation
transpose: batch * node, time_steps, feature
LSTM:
dense: batch * node, pred_time_steps
output reshape: batch, pred_time_steps, node