In the Twin Cities metropolitan area, freeway ramp metering goes back as early as 1969, when the Minnesota Department of Transportation (MnDOT) first tested ramp metering in an I-35E pilot project. To date, the Twin Cities ramp metering system has grown to include more than 433 ramp meters. Research on better, improved ramp control strategies has continued over the years and MnDOT has implemented minor and major changes in the control logic. Two independent studies both aimed at developing the next generation in ramp metering by focusing on density. Based on these efforts, two new algorithms were developed: the UMN Density and the UMD KAdaptive, named based on the campus at which they were developed. The goal of this project was to implement both algorithms and test them under real conditions. Priorities and technical problems prevented the evaluation of the UMN algorithms, so this report focuses on the evaluation of the UMD KAdaptive algorithm on two freeway corridors in the Twin Cities, MN. The first site, a section of TH-100 northbound between 50th Street and I-394, was selected to compare the then current logic, the Stratified Zone algorithm, with the new one. During the course of this project, the UMD algorithm eventually replaced the Stratified Zone algorithm and was implemented in the entire system. This full deployment also included corridors that were not controlled before. The second evaluation site on eastbound TH-212 was a site that allowed for a with/without control evaluation of the UMD algorithm. This report describes the experiments conducted at both sites and includes a comprehensive review of the state of ramp metering strategies around the world to date.

The main objectives of Task Order 4136 are (1) the design of improved freeway on-ramp metering strategies that make use of recent developments in traffic data collection, traffic simulation, and control theory, and (2) the testing of these methods on a 14-mile segment of Interstate 210 Westbound in southern California. To date, the major accomplishments of this project include (i) the development of a complete procedure for constructing and calibrating a microscopic freeway traffic model using the Vissim microsimulator, which was applied successfully to the full I-210 test site, (ii) a simulation study, using the calibrated Vissim I-210 model, comparing the fixed-rate, Percent Occupancy, and Alinea local ramp metering schemes, which showed that Alinea can improve freeway conditions when mainline occupancies are measured upstream of the on-ramp (as on I-210 and most California freeways), as well as when occupancy sensors are downstream of the on-ramp, (iii) development of computationally efficient macroscopic freeway traffic models, the Modified Cell Transmission Model (MCTM) and Switching-Mode Model (SMM), validation of these models on a 2-mile segment of I-210, and determination of observability and controllability properties of the SMM modes, (iv) design of a semi-automated method for calibrating the parameters of the MCTM and SMM, which, when applied to an MCTM representation of the full I-210 segment, was able to reproduce the approximate behavior of traffic congestion, yielding about 2% average error in the predicted Total Travel Time (TTT), and (v) development of a new technique for generating optimal coordinated ramp metering plans, which minimizes a TTT-like objective function. Simulation results for a macroscopic model of the 14-mile I-210 segment have shown that the optimal plan predicts an 8.4% savings in TTT, with queue constraints, over the 5-hour peak period.

Freeway ramp metering systems have been used to improve urban freeway flow. However, control strategies must be properly adjusted to account for ramp queues overflowing onto surface streets and provide equitable on-ramp control during various operating periods. An improved solution can be obtained by optimizing this problem simultaneously for a group of time slices. This study identifies and examines a microcomputer-based optimization scheme that can assist in developing efficient freeway control strategies for on-line freeway surveillance and control. A multi-level freeway control structure is employed for which ramp metering control algorithms are developed for each level of control. Flow-based and lane occupancy-based system algorithms are presented. Detailed data file requirements are provided for each control level. A microcomputer prototype, or laboratory test version of the system level, will be described in a companion project report.

This dissertation investigates various aspect of the design and testing of on-ramp metering control systems, including optimization-based control and microscopic freeway modeling. A new technique for generating optimal metering plans is developed. As with most predictive designs, the ramp metering rates are found as the solution to a nonlinear optimization problem. In contrast to previous designs, the new approach 1) produces a globally optimal solution to the nonlinear problem, 2) requires only to solve a single linear program, and 3) allows the enforcement of hard constraints on the on-ramp queue lengths. The price that is paid for these features is that the objective function being minimized is not Total Travel Time, but rather a member of a class of "TTT-like" objective functions. A TTT-like objective function is defined as a linear combination of mainline flows with weights that, similarly to the Total Travel Time cost weights, decrease linearly in time from some initial value to zero at the final time. An example application of the technique shows that the globally optimal metering plan with respect to a TTT-like objective function also performs well in terms of Total Travel Time. A macroscopic analysis of local traffic-responsive ramp metering on a short stretch of freeway, with a single on-ramp and no offramps, is also presented. The study compares the performance of two popular local traffic-responsive ramp metering algorithms: Alinea and Percent-Occupancy, and addresses issues pertaining to parameter tuning and loop-detector placement. The second half of the dissertation describes the construction of a detailed microsimulation model of a stretch of Interstate 210 in Pasadena, CA. The VISSIM microsimulation package was used to create this model. Descriptions of the data gathering and processing procedures, bottleneck identification, network coding, and model calibration are provided. The model is used to test the performance of candidate local traffic-responsive controllers. Questions concerning the relative merits of these controllers, parameter tuning, and loop-detector placement are addressed in the context of the large-scale microscopic model.