Time-space traffic flow prediction method based on hidden time-space feature mining

The invention belongs to the field of intelligent traffic, and particularly relates to a space-time traffic flow prediction method based on hidden space-time feature mining, which comprises the following steps: taking a road intersection as a road network node, acquiring historical traffic data of the road network node, and converting the historical traffic data into a traffic situation embedding matrix; fusing the granularity information with the traffic situation embedding matrix to obtain a traffic situation pixel matrix; capturing features from multiple dimensions of time and space according to the traffic situation pixel matrix and fusing the features; and prediction of future traffic flow is realized through the non-linearly activated feedforward neural network. Aiming at traffic privacy data leakage and dynamic space-time correlation, the method effectively captures a dynamic change relation between long-time dependence and multi-dimensional characteristics by capturing a time-space dependence relation