an algorithm for event detection based on a combination of loop and journey time data

28
An Algorithm for Event Detection based on a combination of Loop and Journey Time Data Pengjun Zheng, Mike McDonald and David Jeffery Transportation Research Group University of Southampton

Upload: mendel

Post on 15-Jan-2016

30 views

Category:

Documents


0 download

DESCRIPTION

An Algorithm for Event Detection based on a combination of Loop and Journey Time Data. Pengjun Zheng, Mike McDonald and David Jeffery Transportation Research Group University of Southampton. Contents. Event vs. Incident. Event detection for information purposes. Advantages of Loop Data. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

An Algorithm for Event Detection based on a combination of Loop

and Journey Time Data

Pengjun Zheng, Mike McDonald and David Jeffery

Transportation Research Group

University of Southampton

Page 2: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Contents

Event vs. Incident

Event detection for information purposesAdvantages of Loop Data

The proposed algorithm

Preliminary results and conclusions

Page 3: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

The National Traffic Control Centre (NTCC) (1)

is a large-scale project with a budget of EUR212 million.

uses advanced technology from Serco

operates 24 hours/day, 7 days/week

Page 4: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

The National Traffic Control Centre (NTCC) (2)

Covers the Strategic Road Network (motorways + trunk roads) in England.

Some 1,000 CCTV cameras and 4,000 road sensors stream images and data into the facility

Also receives information from HA traffic officers, police, local authorities and weather centres.

Page 5: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Event vs. Incident (1)

• Unplanned Event detection is one of the key services provided by the NTCC

– Unplanned Events can be defined as all events, except Planned Events (e.g. roadworks), having duration of greater than a fixed threshold that could potentially have a material affect on the operation of the Project Network.

– Events are identified in the NTCC by a significant increase of Travel Time lasting for a certain amount of time.

– Events are therefore always associated with excessive delays to travellers in the NTCC.

Page 6: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

An Event

0

1000

2000

3000

4000

5000T

T (

s)

8:20 12:30 16:40 20:500

20

40

60

80

100

120

140

Time

Spe

ed (

km/h

)

TTTT Profile

Page 7: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Event vs. Incident (2)

• How an incident is defined is inconsistent throughout the literature.

– An incident is an unexpected event that temporarily disrupts the flow of traffic on a segment of roadway (Solomon 1991).

– An incident is an event leading or likely to lead to changes in traffic patterns or behaviour resulting in changes in the driving context over some reference period (system).

– It can be argued that any occurrence is an incident provided he/she wants to be aware of it (ie. If no action is to be taken by the control authority the event is not classed as an incident) (operator).

– Anything that might get in his way and cause him inconvenience or delay (driver)

Page 8: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Event vs. Incident (3)

• Incidents do not necessarily cause congestion.

– e.g. Stationary vehicles in the ‘hard shoulder’ lane.

• Events are incidents that cause delays

Page 9: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Event detection for information purposes (1)

• The main purpose of event detection in the NTCC is to disseminate traffic information to the public about actual or likely delays.

– Many incident detection methods cannot be directly applied in the NTCC, e.g. a stationary vehicle in the ‘hard shoulder’ lane is unlikely to cause an event.

• Whilst for the incident detection, minimising the response time is crucial in several aspects.

– Faster treatment for the injured.– Minimising the traffic flow disruption (and potential for secondary

incidents).

Page 10: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Event detection for information purposes (2)

• Events are recognised based on Travel Time Increases.

– Significant increase in Measured Travel Time: The occasions that a Measured Travel Time (MTT) was greater than the Expected Travel Time by more than 12 minutes for each congestion event with MTT greater than the Standard Travel Time by more than 40%.

– Lasting for a certain period: MTT greater than the Standard Travel Time for more than 15 minutes

• Detection of Events is achieved by setting alerts at lower thresholds.

Page 11: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Event detection for information purposes – Problems

• False Alarms.

– A reduction in the threshold will result in the identification of a larger number of events that exceed the new threshold.

• Timing.

– Some Measured Travel Times may increase very quickly to exceed the threshold, which will reduce the effectiveness of using a lower threshold.

– No Measured Travel Times are obtained if the road is totally blocked

Page 12: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Example (false alert)

.

100 150 200 250 3000

1000

2000

3000

100 150 200 250 300-500

0

500

1000

1500

100 150 200 250 300-50

0

50

100

TTprofile

Loop Speeds (km/h)

100 150 200 250 3000

1000

2000

3000

100 150 200 250 300-500

0

500

1000

1500

100 150 200 250 300-50

0

50

100

TT (s)

JTs (s)

EventNot Events

Time (in 5 minute interval, 24:00 =288)

Page 13: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Example (timing)

60 80 100 120 140 160 180 200

0

1000

2000

60 80 100 120 140 160 180 200-1000

0

1000

2000

60 80 100 120 140 160 180 2000

50

100

Time (in 5 minute interval, 24:00 =288)

JT(s)

V(km/h)

JT(s)

Page 14: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Advantages of Loop Data

• Independent source of information not used in the event definition

• Potential very early alert.

• Can pinpoint event location better

• Provide additional information on the nature of the event e.g. whether it is a capacity reducing or demand increasing type.

Page 15: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

An Event

100 110 120 130 140 150 160 170 180 190 200

0

1000

2000

3000

110 120 130 140 150 160 170 180 190 200-500

0

500

1000

100 110 120 130 140 150 160 170 180 190 2000

50

100

150

JT(s)

V(km/h)

JT(s)

Time (in 5 minute interval, 24:00 =288)

Page 16: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Limitations of Loop Data

• Not all events (ie causing a significant increase in TT) can be detected with loop data (e.g. loop not available, events without significant reductions in speeds)

• Not all significant reductions in loop speeds are events (e.g. false alerts not achieving material affects level)

Page 17: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Loop speed data not available

60 80 100 120 140 160 180 200-1000

0

1000

2000

60 80 100 120 140 160 180 200

0

1000

2000

60 80 100 120 140 160 180 2000

50

100

Time (in 5 minute interval, 24:00 =288)

JT(s)

V(km/h)

JT(s)

Page 18: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

An event without loop speed reductions

120 140 160 180 200 220 240

0

1000

2000

3000

120 140 160 180 200 220 240

0

1000

2000

120 140 160 180 200 220 2400

50

100

150

Page 19: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

A ‘false’ alert

120 140 160 180 200 220 240

0

1000

2000

120 140 160 180 200 220 240

-400

-200

0

200

400

600

800

100 150 200 2500

50

100

Page 20: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

The algorithm (1)

160 170 180 190 200 210 220 230 2400

2000400060008000

160 170 180 190 200 210 220 230 240 250

0

1000

2000

3000

160 170 180 190 200 210 220 230 240 2500

50

100

Any Travel Time increases should be a result of some speed reductions, such speed reductions can usually be detected by one or several loops within the Travel Time Section earlier than the increase of Travel Time.

Page 21: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

The Algorithm (2)

Vfree/V>=1.3 for 10 Min

TT/TTstandard>=1.4To Trig AlertTT-TTprofile>=420

(360) s

Vfree/V<1.3 for 15 Min

TT-TTprofile<420 for 15 Min

To End Alert

TT/TTstandard<1.4

Page 22: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

160 170 180 190 200 210 220 230 240

0

2000

4000

6000

8000

160 170 180 190 200 210 220 230 240 250

0

1000

2000

3000

160 170 180 190 200 210 220 230 240 2500

50

100

TT>=1.4TTstandard TT>=1.4TTstandard TT-TTprofile>=600s

Event TT>=1.4TTstandard TT-TTprofile>=360s

Vfree /V>=1.3

Page 23: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Results (1)• Total number of identified events: 68, of which 39

events are accompanied by speed reductions in the loop data.

• Total number of Alerts: 68, of which 39 are Events, i.e. all events with accompanying speed reduction are detected.

• 37 events detected within 10 minutes or before, 2 events detected 5 minutes before.

• The probability that an alert is ‘good’ is 54%.

Page 24: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Results (2)

• 14 events that were not identified by ‘Congestion Alert’ on time can be identified.

• The false alerts are infrequent compared with JT only method.

Page 25: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Suggestions

• Can detect nearly all events with accompanying speed reductions at low false alarm rate and well before other detection methods.

Event detection based on a combination of loop and TT data is a good practice

Page 26: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Suggestions

• The algorithm using loop information will not affect the other algorithm.

• If combined with Journey time only algorithm (with a low threshold), the detection rate could be very high (close to 100%)

Events without accompanying speed reductions can still be detected using Travel Time based methods

Page 27: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Future Work

• The definition of ‘event’ could be optimally determined

based on public opinion of traffic information

requirements.

• It may be beneficial to disseminate some ‘false’ events if

such have been confirmed from other sources .

• The algorithm can be further developed to enable variable

thresholds.

Page 28: An Algorithm for Event Detection based on a combination of Loop and Journey Time Data

Questions …