Performing-time-series based predictions with projection thresholds using secondary time-series-based information stream

• US 9,390,622 B2
• Filed: 04/16/2013
• Issued: 07/12/2016
• Est. Priority Date: 04/16/2013
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Kamarianakis, Ioannis, Wynter, Laura
• Primary Examiner(s): Chad, Aniss
• Application Number:
  US13/863,855
• Publication Number:
  US 20140309976A1
• Time in Patent Office: 1,183 Days
• US Class Current: 1/1
• CPC Class Codes: G08G 1/0116 from roadside infrastructure, e.g. beacons
• CPC Class Codes: G08G 1/0129 for creating historical data or processing based on historical data
• CPC Class Codes: G08G 1/0133 for classifying traffic situation
• CPC Class Codes: G08G 1/0141 for traffic information dissemination
• CPC Class Codes: G08G 1/042 using inductive or magnetic detectors

Abstract

A prediction modeling system, method and computer program product for implementing forecasting models that involve numerous measurement locations, e.g., urban occupancy traffic data. The method invokes a data volatility reduction technique based on computing a congestion threshold for each prediction location, and using that threshold in a filtering scheme. Through the use of calibration, and by obtaining an extremal or other specified solution (e.g., maximization) of empirical volume-occupancy curves as a function of the occupancy level, significant accuracy gains are achieved and at virtually no loss of important information to the end user. The calibration use quantile regression to deal with the asymmetry and scatter of the empirical data. The argmax of each empirical function is used in a unidimensional projection to essentially filter all fully congested occupancy level and treat them as a single state.

Claims

1. A method implemented in a computer system for managing traffic flow on a road network, the method comprising:

receiving, at the computer system, a first time-series data set having one or more values for each time point to be predicted, the first time-series data set comprising traffic occupancy levels obtained from a sensor device associated with a road of said road network;

receiving, at the computer system, a second time-series data set of one or more values per time point with correlation to the first time-series data, the second time-series data set comprising traffic volume levels at the road;

estimating, by the computer system, a functional relationship between the first time-series data and the second time-series data, for each value, over a multiplicity of time points;

determining, at the computer system, an extremal value of the functional relationship of the second time-series data as a function of the first time-series data, said extremal value representing an occupancy level at which a full congested traffic state is reached at the associated sensor device;

modifying, at the computer system, said first time-series data by projecting the occupancy level of the first time series data obtained from the associated sensor device on the extremal value so that first time-series data values that are beyond the extremal value are set to the extremal value;

using, by the computer system, said modified first time-series data in any prediction model to increase accuracy of a future predicted traffic occupancy state; and

regulating a traffic flow of said road network based on said future predicted traffic occupancy state.

2. The method of claim 1, wherein first time-series data set includes a vector variable of interest, m(t), where t is a unit of time, and said second time-series data set includes an auxiliary variable, n(t), wherein said functional relationship between the first time-series data and the second time-series data, for each value, over the multiplicity of time points, is a function n(m).

3. The method of claim 2, wherein said step of determining an extremal value of the functional relationship comprises:

calibrating, for each of the time-series vector variable of interest, a curve that fits most closely data from the variable of interest and the auxiliary variable; and

computing a maximum threshold value τ

, of the auxiliary variable beyond which value of the variable of interest is not predicted.

4. The method of claim 3, wherein said modifying said first time-series data based on the extremal value comprises:

obtaining the maximum threshold value τ

of said calibrated curve, wherein τ

represents an occupancy level at which a full congested state occurs; and

unidimensionally projecting the occupancy level onto that threshold.

5. The method of claim 4, wherein said modifying said first time-series data is based on the following:

ŷ

={y(t),τ

}⁻

,

where {·

}⁻

is a minimum operation, ŷ

is said modified first time-series data, y(t) is said first time series data.

6. The method of claim 5, further comprising:

repeating said receiving first and second time-series data, said estimating, said determining, said modifying and said predicting for all elements of a variable of interest.

7. The method of claim 4, further comprising:

computing a minimal threshold value μ

of the auxiliary variable;

applying a first projection according to;

m′

(t)=min{m(t), τ

},

determining if a minimal threshold value μ

exists, and

if said minimal threshold value μ

exists, applying a second projection time series variable m″

(t)=max{m′

(t), μ

}, wherein said predicting is performed on said time series variable, m″

(t).

8. The method of claim 3, wherein said first time-series data set is road traffic data measuring traffic speeds or traffic occupancies obtained from said associated sensor device, and the second time-series data set is road traffic data measuring traffic volumes, wherein said modifying said first time-series data based on the extremal value comprises:

obtaining the maximum threshold value τ

of said calibrated curve, for every associated sensor device, s, wherein τ

_s represents an occupancy level at which a full congested state occurs at said associated sensor device s; and

unidimensionally projecting the occupancy level onto that threshold according to;

ŷ

_s={y_s,τ

_s}⁻

,

where {·

}⁻

is a minimum operation, ŷ

_s is said modified first time-series data for the associated sensor device s, y(t) is said first time series data for the associated sensor device s.