Sa-fql, Algorithm Applied Research In The Area Traffic Control

As opposed to single intersection traffic control, area traffic control can obtain global optimization. In traditional methods of area traffic control, it needs to establish mathematical models for traffic systems. But traffic systems are typical nonlinear, time-varying, stochastic, complicated and large scale systems, so it's hard to establish precise mathematical models for them. Q-Learning algorithm is not necessary to establish the mathematical models and can learn the policy in the real environment, so it's fit to be applied in area traffic controlHowever, there are some shortcomings in Q-learning algorithm, generally, its learning speed is quite slow, and it's difficult to balance between exploration and exploitation of action selection and so on. In order to resolve these issues, Q-learning algorithm is modified in this thesis. The main contents of the thesis are the following four aspects:1. Some classical modified Q-learning algorithms are summarized and their shortcomings are discussed, then SA-FQL algorithm, a new modified Q-learning algorithm is presented based them. For accelerating the learning process, the prior knowledge is embedded into the fuzzy rules in the proposed algorithm. Theεvalue of theε-greedy policy is adjusted by changing the temperature in the algorithm, and then the balance between exploration and exploitation is achieved2. A new optimizing method for area traffic control which based on SA-FQL algorithm is proposed. The common cycle of the traffic network is optimized by using SA-FQL algorithm, and based on the common cycle the offset of each arterial in the network is optimized by using the same algorithm. Finally, the split of each intersection is adjusted according to its traffic volume.3. The new method is simulated by TSIS simulation software. Results show that, compared with the method based on Fuzzy Q-Learning algorithm and the method based on Q-Learning algorithm, the proposed method can significantly accelerate learning and improve traffic efficiency.