With the rapid development of modern society, the number of vehicles and the need for mobility has increased beyond the current road capacity. This has resulted in congestions, consequent excess delays, reduced pedestrian and vehicle safety, and increased air pollution. A promising solution to these problems exploits the existing infrastructure through efficient dynamic traffic management and control. Specifically, in freeway traffic systems, many measures could be adopted to improve the service quality of freeways, such as ramp metering, route guidance, reversible lanes, speed limits and so on. Among these measures, ramp metering is a well-known method extensively used in present freeway traffic systems. This method regulates the volume of traffic entering a given freeway at its entry ramps, so that the freeway can operate at some desired level of service. When properly designed, ramp metering can efficiently alleviate recurrent and nonrecurrent congestions, which has been proven both by mathematically sound arguments and in practice. It is well known that traditional dynamic programming is limited in applications due to its high computation and storage complexity for high order nonlinear systems, causing the problem of the curse of dimensionality. However, Approximate Dynamic Programming (ADP) is able to artfully circumvent such difficulties by using a critic network for estimating the performance and an action network for generating optimal actions. In order to improve the learning efficiency of ADP and to solve the problem of random selecting initial parameters with ADP, we combine eligibility traces of reinforcement learning and neuro-fuzzy network with ADP, then present the novel ADHDP(λ) and NFADHDP(λ) methods which are used to solve the non-recurrent and recurrent congestions in freeway. Firstly, we pay attention to freeway control problems. We also present the research background and significance of this paper. The research aims, as well as this paper’s main work and missions are introduced. Secondly, we introduce the freeway traffic models; present a general overview of the concept of the parameters of traffic theory and relationships between them. Then we introduce the LWR and MACK traffic flow models which are used in simulation tests to present the freeway optimal control problems. Thirdly, this paper introduces the traditional Dynamic Programming (DP), then presents an overview of the reinforcement learning and ADP theory. For dealing...
修改评论