Incident detection through a neural-net and estimation of information required for traffic control
| Project/Area Number |
05650505
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
交通工学・国土計画
|
|---|
| Research Institution | Shinshu University |
|---|
Principal Investigator |
OKUTANI Iwao Shinshu univ., Eng., Professor, 工学部, 教授 (90026138)
|
|---|
| Project Period (FY) |
1993 – 1994
|
| Project Status |
Completed (Fiscal Year 1994)
|
| Budget Amount *help |
¥2,300,000 (Direct Cost: ¥2,300,000)
Fiscal Year 1994: ¥1,300,000 (Direct Cost: ¥1,300,000)
Fiscal Year 1993: ¥1,000,000 (Direct Cost: ¥1,000,000)
|
|---|
| Keywords | Freeway / Incident detection / Traffic control / Neural network / Fuzzy inference / Shock wave analysis |
|---|
| Research Abstract |
Fundamental explorations are made to develop an automatic incident detection tool as well as a method for estimating travel time between ramps when an incident occurs on a freeway. Four models, Model 1 to Model 4, are constructed as an incident detector. Model 1 applys a three-layr neural network which has the input neurons in accordance with employed traffic variables measured at upstream and downstream end of a study road section of the freeway and has the output neuron to tell the presence of an incident. Model 2 utilizes the fuzzy inference technique assuming the difference in the traffic variables between the section ends. Model 3 (Model 4) is a neuro-fuzzy (fuzzy-neuro) model combining two aforementioned models with the fuzzy inference (neural network) using as an input the output from the neurtal network (fuzzy inference). Widely excuted testings by use of simulated date reveal that (1) Model 1 produces the most stable estimates when speed and occupancy are used as the input variables while the speed only input is sufficient for Model 1 2 ; (2) Model 3 outperforms Model 1 when the traffic demand exceeds a fixed level ; (3) Model 2 is likely to rarely perform better than Model 1 and Model 3 ; (4) Model 4 performs best of all models tested. As regards travel time estimation, developed is a method which makes use of shock wave analysis where the traffic flow is assumed uniform. It is shown that the method can only make coarse estimates with relative errors of 20% to 30% probably because it bases the modelling on too much simplification of the traffic flow.
|
Report
(3 results)
Research Products
(15 results)
