A promising measure for solving traffic problems seems to exploit the existing infrastructure through efficient dynamic traffic management and control. For this purpose, two advanced neural techniques, i.e. neural dynamic optimization (NDO) and adaptive dynamic programming (ADP), are adopted for some fundamental traffic control problems in this dissertation. Specifically, the optimal signal timing problem is considered for an urban intersection under oversaturated traffic condition. Based on the technique of NDO, two optimal signal timing controllers are proposed for the two-phase case and the four-phase case, respectively. Based on the technique of NDO, the coordinated control of variable speed limits in highway systems is also considered. By viewing the obedience of drivers inverse proportional to the strength of limiting speeds, a new traffic model is proposed in order to characterize the effect of speed limits. With this new traffic model, the NDO controller is developed and tested. Based on the mechanism of eligibility traces in reinforcement learning, the algorithm of ADHDP(λ) is proposed. For the benchmark problem of Narendra system, two new DHP designs are proposed, where primary utility functions are defined exclusively by information at the current time. Based on the technique of ADP, the problems of local ramp metering and coordinated ramp metering are considered. Simulation studies demonstrate ADP controllers with good control performance and strong robustness. Finally, implementation issues are discussed for those traffic control algorithms proposed in this dissertation. Specifically, a framework is built by combining many advanced concepts or techniques, such as networked control, local simple remote complex (LSRC), mobile agent, and so on.
修改评论