On Approximate Dynamic Programming And Its Applications In Transportation

This dissertation focused on some issues on approximate dynamic programming (ADP) and its applications in transportation. The main work and key contributions were summarized as the following:1. Based on Steffensen's method and former weights'values, a new optimized accelerated algorithm was presented for neural network. On the basis of this new optimized accelerated algorithm, a new ADHDP(Action-Dependent Heuristic Dynamic Programming) method based on former weights'value data (ADHDP(D) for short) was proposed. A detailed analysis was made between the ADHDP and ADHDP(D). Simulation results showed the good performance of this new algorithm. Furthermore, an initial weights compositional method was proposed. Simulation results demonstrated the convergence property is improved effectively in comparison with initial weights randomly preassigned.2. Considering on that traffic system is a strong nonlinear and uncertain system, an identification method based on ADP was developed to estimate the parameters of the general discrete-time nonlinear traffic flow system. With rigorous analysis, it was shown that the proposed identification scheme, independent of the precise traffic flow model, can guarantee the convergence. A number of simulation results were provided to the efficacy of the proposed approach.3. An adaptive single-exponent smoothing based on ADP was put forward to select the smoothing coefficient dynamically. With rigorous analysis, it was shown that the proposed predictionscheme can guarantee the convergence. The simulation results verifid the effectiveness of the proposed algorithm.4. The optimal signal timing problem was investigated for an urban intersection.Based on the technique ADP, the optimal signal timing controllers were proposed for the two-phase intersection and three-phase intersection, respetively. The simulation examples showed that the control algorithm based on ADP can allocate green time rationally and achieve the equilibrium of queue length. Furthermore, taking the optimal signal timing of three-phase intersection for example, simulations verifid that the initial weights affect the convergence result.5. Based on the technique of ADP, the problems of local ramp and coordinated ramp metering were discussed. The ADP controller was designed to avoid the difficulty of traffic flow modeling. Simulation results demonstrated the new controllers have better transient response, prevent congestion and increase traffic throughput.