UC Berkeley

This dissertation presents a series of traffic management studies including freeway and intersection traffic modeling, estimation, control methodologies and field implementation tests. First, a traffic flow prediction method that combines the most recent traffic data with historical traffic data is studied. An autoregressive moving average with exogenous input (ARMAX) model is estimated on-line based on the most recent vehicle detector station (VDS) data. Results obtained using empirical freeway mainline and on-ramp data show that this method outperforms methods that rely only on the historical average of the data to perform a prediction, especially during days with unusual traffic flow demands.

Second, two freeway control strategies: coordinated ramp metering (CRM) and variable speed advisory (VSA) are investigated and implemented in field tests. In the control of CRM study, the freeway is modeled by the cell transmission model (CTM) and the control problem is solved by the model predictive control (MPC) scheme. The proposed CRM is deployed in a segment of California State Route 99 Northbound (SR-99N) for a five-week field test. The test results shows that the freeway efficiency can be improved by 7.25% for morning peak hours. In the VSA control study, an advisory speed limit control is designed by using traffic flow stabilization of the Lighthill-Whitham-Richards (LWR) model. The proposed VSA is deployed in a segment of California State Route 78 Eastbound (SR-78E) for a five-week field test. The test results shows that the freeway efficiency can be improved by 8.71% for morning peak hours. Both control strategies indicate freeway efficiency improvement in congested traffic.

Third, the large-scale signalized intersection traffic network control by offset optimization is also studied. The traffic network is described by a directed graph and the traffic dynamic is represented by continuous-time fluid queue model with sinusoidal arrival and departure rate assumption. The original non-convex offset optimization problem can be relaxed into a semidefinite program (SDP). The Burer-Monteiro (BM) method is used for solving the large SDP to avoid conic constraints. Two real-world traffic simulation networks, respectively in Manhattan, NY and in Pasadena, CA are constructed for demonstrating the BM method. Numerical simulation results indicate that BM method has good scalability and it can efficiently recover optimal solutions of the SDP.