Systems and methods for traffic flow prediction

A method of generating a traffic flow prediction, comprising:

• receiving traffic data from a plurality of reporting sources, the traffic data being associated with roads in a geographic area and indicating reported information about the roads;
• forming a plurality of initial graphs, each created by fusing the traffic data received over first lengths of time;
• generating a plurality of completed graphs based on the plurality of initial graphs by at least removing noise from the plurality of initial graphs;
• generating a plurality of feature vectors that represent spatial relationships and temporal relationships in the plurality of completed graphs;
• outputting a first feature vector of the plurality of feature vectors corresponding to a first time window as a first output;
• caching a copy of the first feature vector with a set of feature vectors of the plurality of feature vectors corresponding to a second time window;
• connecting the first feature vector with the set of feature vectors;
• outputting a result of the connecting as a second output, where the second time window is longer than the first time window;
• fusing the first output with the second output to produce fused outputs;
• and
• generating a traffic prediction based on the fused outputs.

A non-transitory computer-readable medium for generating a traffic flow prediction and including instructions that when executed by one or more processors cause the one or more processors to:

• receive traffic data from a plurality of reporting sources, the traffic data being associated with roads in a geographic area and indicating reported information about the roads;
• form a plurality of initial graphs, each created by fusing the traffic data received over first lengths of time;
• generate a plurality of completed graphs based on the plurality of initial graphs by at least removing noise from the plurality of initial graphs;
• generate a plurality of feature vectors that represent spatial relationships and temporal relationships in the plurality of completed graphs;
• output a first feature vector of the plurality of feature vectors corresponding to a first time window as a first output;
• cache a copy of the first feature vector with a set of feature vectors of the plurality of feature vectors corresponding to a second time window;
• connect the first feature vector with the set of feature vectors and output a result as a second output, where the second time window is longer than the first time window;
• fuse the first output with the second output to produce fused outputs;
• and
• generate a traffic prediction based on the fused outputs.