Domain

Explanation

Seminars

• Arterial travel time estimation using stop-line detectors

A model to estimate average link travel time of signalized arterials using data obtained from detectors at the stop lines have been proposed and tested. The proposed model makes use of detectors at the upstream stop line of a link to capture the arrival of vehicles, and use a platoon dispersion model to project the vehicle arrival times at the downstream stop line. Using the vehicle departure patterns obtained at detectors at the downstream stop line, the average link travel time is then estimated. The accuracy of the proposed model has been tested, by means of microscopic traffic simulations, for different link lengths and urban street level of service.

• Freeway incident detection using kinematic data from probe vehicle

This research presents an incident detection algorithm based on the speed and acceleration profiles of probe vehicles as they travel along an expressway. It is based on the assumption that when a probe vehicle approaches a detectable incident, it will decelerate from its normal speed and then accelerate back to the normal speed after passing the incident. The incident detection performance of the algorithm, at various percentages of probe vehicles in the traffic stream, has been tested on a set of incident data generated by a calibrated microscopic traffic simulation model.

• Lane closure optimization using GA

This presentation introduces GAPSIM, a hybrid Genetic Algorithm-Parallel SIMulation model for scheduling of lane closure requests that aims to minimize total traffic delay in a network over a 24-hour period. Genetic algorithm is used as the search engine while a microscopic traffic simulation tool is used to estimate traffic delay under a set of lane closure schedule. GAPSIM has been implemented in a UNIX multi-processor server. This presentation covers an example involving the scheduling of 20 lane closures in the road network in Clementi area.

By Liu Qun

• Using new model to estimate travel time between stop-lines using detectors & compare-contrast
• Detectors: at stop-lines; detect movements @ turn-left lanes & straight; data collection – sampling
• Model:
• Travel time = cruise time + intersection delay
• Cruise time factors: free flow speed (less used, due to low volume & no traffic control); running speed (more used)
• Arrival profile prediction: @ upstream & @ downstream
• Platoon dispersion model: statistical possibility of travel time
• Intersection delay: graph of Q vs t between actual & delayed profile
• Experimental design:
• Setup: lanes, network, length L, detectors, vehicle composition, saturation
• Design: PARAMICS calibration – modeled data, not actual data
• Model factors: data collection interval
• Parameters:
• Error: average absolute; average relative
• Aspects: R-square; slope
• Model verification: Actual vs estimated
• Model validation: compare with the following existing models
• British model
• Illinois model
• Iowa model
• Sensitivity analysis:
• Platoon dispersion
• Parameter b

By Qi Hongtu

• Using in-vehicle detection algorithm to predict & detect incidents in real-time with only the probe vehicle & velocity & acceleration profile
• Sensors:
• Static: volume, speed, occupancy
• Mobile: black-box; DGPS (differential global positioning system)
• Incidents: unwanted disruptions to traffic flow resulting not from regular traffic controls, but from accidents, breakdowns, works & maintenance
• AID: automatic incident detection
• Techniques:
• Pattern recognition
• Catastrophe theory
• Low pass filter
• Neural network
• Performance measures:
• Detection rate: of real incidents
• Mean time to detect: real incidents
• Misclassified rate
• False alarm rate: w.r.t. derived thresholds
• FA1: false alarm number / total incidents
• FA2: false alarm number / total incident decisions by TMC (traffic management control)
• Algorithm: MOSA (mobile sensor auto-reporting algorithm)
• On-board probe vehicle
• Inputs needed: velocity & acceleration profile
• Real-time incident reporting
• Simulation of incidents:
• Map overview: site
• Vehicle detection sensors
• Data: location, volume, periods – peak & off-peak
• Selection of parameters:
• Graphs comparison w.r.t. to (parameters) vs. (probe vehicle %, FA)
• Comparison with MLF (multi-layer feedforward neural network)
• Results: less effective than MLF especially false alarms, but MOSA simplicity
• Corollary: fault detection, residual generation, thresholds, reduce false alarm rate

By Ma Wenteng

• Find the optimal lane closure period from within an allowable range of time using GA to minimize the traffic delay
• Lane closure: road open, but specific lane(s) closed for works, accidents or parking
• Hybrid GA:
• Uses existing tool from pavement maintenance scheduling
• Minimize traffic delay: fitness = (total delay) / (total runs)
• Increase convergence & reduce CPU time
• Diagram: cyclic diagram of old population à fitness evaluation à selection à mating à recombination à new population à replacement à repeat
• Chromosome:
• Length: number of lanes for closure along the network
• Each bit: starting time to closure
• Criteria: for declaring optimal selection
• Standard error: standard deviation (Ross 1990) s/sqrt®<=epi, r={r_min,r_max}
• Termination: stops @ threshold – with same optimal for threshold generations
• Running with GA tool for selection of optimal lane closure period schedule & PARAMICS tool for API model of scenarios from GA
• Simulations:
• Without precondition
• With precondition & standard error
• With precondition & termination
• With precondition & traffic assignment model
• Precondition technique: selection of good IC
• Idea: lanes with less flow, higher probability of selection
• Selecting N possible starting times
• Volume for selection
• Comparison: without precondition & with precondition (effective); w.r.t. to time delay
• (Parallel) Distributed simulation:
• Automatic workload distribution among x processors using threads allocated by operating systems (Windows – equal priority)
• Reduce CPU time using more processors
• Additional selection using traffic assignment model:
• Capacity retainment assignment model: I/O à trees à links à optimal à (N) re-connect à (Y) optimal
• BPR 1964: analytical closed-form; what’s the linkage between this model & GA; bound travel time for GA à better IC
• GA knowledge:
• If rapidly converges with few population & generation sizes à better to use other closed-form solution to solve (like BPR 1964)
• Convergence not guaranteed, but with better algorithm & techniques can have better performance (lower traffic delay)

Lessons

• Go early for seminar & bring papers to read during waiting
• Smooth-flowing delivery is best, even if the results are undesirable
• Deep-throated, deep voice conducive to thoughtfulness
• Once questioned, give appreciation good
• Questioning testing & breaches fundamentals of the project à never sound or appear rushed à move to new equilibrium & provide thoughtful answers or ideas
• Within the scope of research
• If out of scope à provides references & gives counter-rationale to support project