Daniel C. Chin

TRAFFIC-RESPONSIVE SIGNAL TIMING FOR SYSTEM-WIDE TRAFFIC CONTROL

Abstract A long-standing problem in traffic engineering is to optimize the flow of vehicles through a given road network. Improving the timing of the traffic signals at intersections in the network is generally the most powerful and cost-effective means of achieving this goal. However, because of the many complex aspects of a traffic system—human behavioral considerations, vehicle flow interactions within the network, weather effects, traffic accidents, long-term (e.g. seasonal) variation, etc.—it has been notoriously difficult to determine the optimal signal timing. This is especially the case on a system-wide (multiple intersection) basis. Much of this difficulty has stemmed from the need to build extremely complex models of the traffic dynamics as a component of the control strategy. This paper presents a fundamentally different approach for optimal signal timing that eliminates the need for such complex models. The approach is based on a neural network (or other function approximator) serving as the basis for the control law, with the weight estimation occurring in closed-loop mode via the simultaneous perturbation stochastic approximation (SPSA) algorithm. The neural network function uses current traffic information to solve the current (instantaneous) traffic problem on a system-wide basis through an optimal signal timing strategy. The approach is illustrated by a realistic simulation of a nine-intersection network within the central business district of Manhattan, New York.

A model-free approach to optimal signal light timing for system-wide traffic control

A long-standing problem in traffic engineering is to optimize the flow of vehicles through a given road network. Improving the timing of the traffic signals at intersections in the network is generally the most powerful and cost-effective means of achieving this goal. However, because of the many complex aspects of a traffic system-human behavioral considerations, vehicle flow interactions within the network, weather effects, traffic accidents, long-term (e.g., seasonal) variation, etc.-it has been notoriously difficult to determine the optimal signal light timing. This is especially the case on a system-wide (multiple intersection) basis. Much of this difficulty has stemmed from the need to build extremely complex open-loop models of the traffic dynamics as a component of the control strategy. This paper presents a fundamentally different approach for optimal light timing that eliminates the need for such an open-loop model. The approach is based on a neural network (or other function approximator) serving as the basis for the control law, with the weight estimation occurring in closed-loop mode via the simultaneous perturbation stochastic approximation (SPSA) algorithm. Since the SPSA algorithm requires only loss function measurements (no gradients of the loss function), there is no open-loop model required for the weight estimation. The approach is illustrated by simulation on a six-intersection network with moderate congestion and stochastic, nonlinear effects.<<ETX>>

Evaluation of system-wide traffic signal control using stochastic optimization and neural networks

The problem of system-wide traffic control is one of the most challenging in advanced traffic management. The S-TRAC (system-wide traffic-adaptive control) method was introduced as a means for producing optimal real-time signal timings on a system (network)-wide basis. S-TRAC has several desirable features that make it both practically feasible and theoretically sound in addressing the system-wide control problem. Among these features are: (1) no system-wide traffic flow model is required; (2) S-TRAC automatically adapts to long-term changes in the system (e.g., seasonal variations) while providing real-time responsive signal commands; and (3) S-TRAC is able to work with existing hardware and sensor configurations within the network of interest (although additional sensors may help the overall control capability). The Montgomery County (Maryland) Department of Public Work and Transportation and JHU/APL have collaborated in moving towards a possible field demonstration of S-TRAC in a moderately congested network. The paper presents an innovative measure-of-effectiveness that evaluates the interruptions of the traffic flow caused by the traffic signal and also reflects the needs of traffic engineers in Montgomery County, Maryland. Also, the paper describes some of the practical implementation issues that have been addressed and presents the results of some realistic simulations built from Montgomery County traffic data.

Traffic-responsive signal timing for system-wide traffic control

Due to the many complex aspects of a traffic system, it has been difficult to determine the optimal signal timing. Much of this difficulty has stemmed from the need to build extremely complex models of the traffic dynamics as a component of the control strategy. This paper presents a fundamentally different approach for optimal signal timing that eliminates the need for such complex models. The approach is based on a neural network serving as the basis for the control law, with the weight estimation occurring in closed-loop mode via the simultaneous perturbation stochastic approximation (SPSA) algorithm. Since the SPSA algorithm requires only loss function measurements, there is no system-wide model required for the weight estimation.

First-order data sensitivity measures with applications to a multivariate signal-plus-noise problem

Abstract This paper considers the use of first-order (implicit-function-based) measures of the sensitivity of statistical parameter estimates to certain elements within the data. Although the methods considered are general, the focus is on a maximum likelihood problem in a signal-plus-noise context. We evaluate the accuracy of the measures and give an example of how they will be used in data analysis for a physical system.

Discrimination of buried plastic and metal objects in subsurface soil

The electrical conductivity object locator (ECOL) uses electric conductivity maps to distinguish buried foreign objects from the regular soil in the subsurface. Assuming that foreign objects and the regular soil have different electrical conductivities, when an electrical current is induced into the subsurface, the difference in conductance causes an electrical field distortion. Theoretically, one can measure the outside field distortion to solve the conductivity profile. Because the problem is highly nonlinear and has noisy field conditions, mapping the conductivity profile is an interesting and challenging task. In addition, the high contrasts in conductivity values among metallic and nonmetallic objects and soil and the high correlation within the model parameters add to the level of difficulty. The high contrast causes the computational instability in the inversion; the high correlation is due to locating the small objects. The ECOL technology utilizes several techniques to overcome the difficulties and locate the mine-like small objects. ECOL applies a low-amplitude (100-/spl mu/A to 500-/spl mu/A) electric alternating current, single or multiple frequency. The impressed AC current generates AC potentials and magnetic fields throughout the site; these are measured at the surface and the boundary of the site. Also, ECOL establishes a finite element model to compute the surface and boundary values from the amount of current, physical structure, and assumed or previous estimated conductivity profile of the subsurface. ECOL estimates the conductivity profile of the subsurface and the characters of the buried object by minimizing the sum of the square of the differences between the measured and the computed values. The minimization is based on a gradient approximation technique, namely, simultaneous perturbation stochastic approximation.

A system-wide approach to adaptive traffic control

A key problem in traffic engineering is the optimization of the flow of vehicles through a given road network. Improving the timing of the traffic signals at intersections in the network is generally the most powerful and cost-effective means of achieving this goal. Recent efforts have resulted in the development of an approach for optimal centralized signal timing that eliminates the need for an open-loop model. The approach is based on a neural network (NN) serving as the basis for the control law, with the internal NN weight estimation occurring real-time in closed-loop mode via the simultaneous perturbation stochastic approximation algorithm. This paper investigates the application of such a non-network-model-based approach and illustrates the approach through a simulation on a nine-intersection, mid-Manhattan, New York network. The simulated traffic network contains varying short and long-term congestion behavior and short-term stochastic, nonlinear effects. The approach results in a net 10% reduction in vehicle wait time relative to the performance of the existing, in-place strategy.

3-d discrimination of buried object in subsurface soil via magnetic sensors

The Electrical Conductivity Object Locator (ECOL) has been developed with the goal of detecting buried objects. Its specific capability to detect and characterize small-size plastic and metal objects buried at shallow depths is demonstrated. The technique can also detect larger objects at greater depths. The ECOL technique maps the soil subsurface conductivity and identifies variations in the conductivity between buried objects and their surroundings. The subsurface conductivity is mapped in two major steps: 1) Low-frequency (1 to 100 Hz) and low-amplitude (<200 /spl mu/A) currents injected into the soil induce potential and magnetic fields in and around the subsurface soil. The potential and magnetic fields are measured using appropriate sensors placed on or above the soil surface. 2) Using the measured values as boundary conditions, a fast optimization algorithm, and an accurate matrix inversion routine, the subsurface conductivity is estimated. Two field tests are conducted using magnetic sensor in either contact or non-contact technique. Both tests successfully located the buried plastic and metal objects within a radius of 1.2 ft.

Multi-model interpolation of range-varying acoustic propagation

This paper develops a model-fitting technique to perform the interpolation and extrapolation of a nonlinear time-varying system. The development is demonstrated on the problem of transmission loss of underwater sound. The technique involves simplified time-varying multiple models, neural networks (NNs), and multiobjective simultaneous perturbation stochastic approximation (MSPSA). The simplified models represent the local phenomena that change in time, the NNs capture the model variations, and MSPSA trains the NN-weights. The localized multi-model technique has shown accuracy and efficiency in the transmission loss interpolation.

Efficient identification procedure for inversion processing [magnetospheric imaging]

Model extrapolation and data inversion are two important steps in control. Model extrapolation is a step to find control variables that are viable for the control process. Data inversion is a step to estimate the control variables from the sensor data. The assumptions and goals are different in these two steps. The model extrapolation step searches for the control variable that gives minimum errors between modelled and true measurements; the data inversion step searches for the values of the control variables that give minimum errors between the modelled and observed measurements. This paper discusses the control processing used in the magnetospheric image setting. The magnetosphere exists in the region of space that surrounds the Earth several hundred kilometres above the Earths surface, extends and is filled with magnetic field lines passing through the Earths surface. The magnetospheric image process is to estimate the ion-population within the field. The ion-population variables and the magnetospheric images have no easily modelled physical relationship. The gradient formula is hard to define, therefore gradient dependent optimization techniques are not applicable. Chase and Roelof (1995) use the Powell algorithm for magnetospheric model extrapolation. Unfortunately, the Powell method is a trial-and-error type of search algorithm and may break down in data inversion, due to high-noise sensors and multiple root problems. This paper shows that simultaneous perturbation stochastic approximation is more efficient than the Powell algorithm and performs well in the data inversion problems.