System and method for multi-task learning for prediction of demand on a system

Patent metadata

Field	Value
Publication number	US-9349150-B2
Application number	US-201314140640-A
Country	US
Kind code	B2
Filing date	Dec 26, 2013
Priority date	Dec 26, 2013
Publication date	May 24, 2016
Grant date	May 24, 2016

Abstract

A multi-task learning system and method for predicting travel demand on an associated transportation network are provided. Observations corresponding to the associated transportation network are collected and a set of time series corresponding to travel demand are generated. Clusters of time series are then formed and for each cluster, multi-task learning is applied to generate a prediction model. Travel demand on a selected segment of the associated transportation network corresponding to at least one of the set of time series is then predicted in accordance with the generated prediction model.

First claim

What is claimed is:

1. A multi-task learning method for predicting travel demand on an associated transportation network, comprising:
collecting observation data corresponding to a plurality of observations of the associated transportation network;
generating a set of time series corresponding to transportation network data of segments of the associated transportation network;
forming a plurality of clusters of time series, each time series in the set thereof in one cluster;
for each cluster, applying multi-task learning to generate a prediction model, including:
determining a common prediction model for all series in a cluster thereof, and
determining a specific prediction model for each series in the cluster thereof, wherein the prediction model generated via the multi-task leaminq application is a sum of the common prediction model and the specific prediction model for each cluster; and

predicting travel demand on a selected segment of the associated transportation network corresponding to at least one of the set of time series in accordance with the generated prediction model,
wherein at least one of the collecting, generating, forming, and predicting is performed by a computer processor.

2. The method according to claim 1 , further comprising collecting transportation network data from a plurality of collection components disposed on the associated transportation network.

3. The method according to claim 2 , wherein the collected transportation network data includes data representative of current usage of at least one of a route, a vehicle, or a stop of the associated transportation network.

4. The method according to claim 3 , wherein generating the set of time series includes generating a time series for each route, vehicle or stop of the associated transportation system corresponding to usage thereof.

5. The method according to claim 4 , wherein forming the plurality of clusters further comprises:
for each pair of time series in the set thereof, measuring a similarity therebetween via at least one selected process; and
forming the plurality of clusters of time series in accordance with the similarity measurement associated therewith.

6. The method according to claim 5 , wherein forming the plurality of clusters further comprises applying a k-means clustering methodology in accordance with the similarity measurements for each pair of time series.

7. The method according to claim 6 , wherein the at least one selected process is dynamic time warping.

8. The method according to claim 7 , wherein the observation data comprises at least one of historical data, traffic data, weather data, or calendar data.

9. The method according to claim 8 , wherein the multi-task learning includes support vector regression.

10. The method according to claim 9 , further comprising:
receiving a selection of the segment of the associated transportation network from an associated user;
identifying a time series associated with the selected segment; and
inputting collected observation data into the common prediction model and the specific prediction model associated with the identified time series, wherein the travel demand is predicted for the selected segment in accordance therewith.

11. The method according to claim 10 , wherein the segment of the associated transportation network is selected from the group consisting of a route, a vehicle, a stop, a zone, a series of routes, a set of vehicles and a sequence of stops.

12. A computer program product comprising a non-transitory recording medium storing instructions, which when executed on a computer causes the computer to perform the method of claim 1 .

13. A system comprising memory storing instructions for performing the method of claim 1 , and a processor in communication with the memory which implements the instructions.

14. A multi-task learning system for predicting travel demand on an associated transportation network, comprising:
a time series generator component configured to generate a set of time series corresponding to segments of the associated transportation network;
a multi-task learning module configured to generate a prediction model for each of a plurality of clusters of time series, wherein the multi-task learning module determines a common prediction model for all series in a cluster thereof, and determines a specific prediction model for each series in the cluster thereof, and wherein the generated prediction model is a sum of the common prediction model and the specific prediction model for each cluster;
memory which stores instructions for:
collecting, from a plurality of collection components, transportation network data corresponding to the segments of the associated transportation network,
receiving observation data corresponding to a plurality of observations associated with the transportation network, and
predicting travel demand on a selected segment of the associated transportation network in accordance with the received observation data and the prediction model of the cluster with which the time series corresponding to the selected segment is associated; and

a processor in communication with the memory which executes the instructions.

15. The system of claim 14 , wherein the segments of the associated network correspond to at least one of a route, a stop, or a vehicle.

16. The system of claim 15 , further comprising a similarity module configured to measure a similarity measure between each of a pair of time series via dynamic time warping.

17. The system according to claim 15 , further comprising a clustering module configured to apply a k-means clustering method to the time series in accordance with the similarity measure of each pair of time series.

18. The system according to claim 17 , wherein the observation data comprises at least one of historical data, traffic data, weather data, or calendar data.

19. The system according to claim 18 , wherein the prediction models are generated via multi-task learning support vector regression.

20. A computer-implemented multi-task learning method for predicting travel demand on an associated transportation network, comprising:
collecting data corresponding to the associated transportation network, the data including at least one of observation data and transportation network data;
generating a set of time series corresponding to segments of the associated transportation network in accordance with at least one of the observation data or the transportation network data;
measuring a similarity between each pair of the set of time series via dynamic time warping;
forming a plurality of clusters of time series based upon the measured similarity of each pair of time series;
for each cluster, applying multi-task learning support vector regression to generate a prediction model, the application comprising:
determining a common prediction model for all series in a cluster thereof; and
determining a specific prediction model for each series in the cluster thereof, wherein the prediction model generated via the multi-task learning application is a sum of the common prediction model and the specific prediction model for each cluster; and

predicting travel demand on a selected segment of the associated transportation network corresponding to at least one of the set of time series in accordance with the generated prediction model.

Assignees

• Xerox Corp

Inventors

• Chidlovskii Boris

Classifications

• G06Q10/04 Primary

  Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" (market predictions or forecasting for commercial activities G06Q30/0202) · CPC title

• G08G1/0129

  for creating historical data or processing based on historical data · CPC title

• G06Q50/26 Primary

  Government or public services (business processes related to the transportation industry G06Q50/40) · CPC title

• G06N5/025

  Extracting rules from data · CPC title

• G06Q50/28

  Physics · mapped topic