The tram signal priority strategy is a practical approach to solving the conflict between trams and other vehicles by giving trams appropriate priority signals. In the existing research, many models-based multi-objective optimization algorithms are used, while a few others use Deep Q-learning Networks (DQN), all based on entirely observable states. Accordingly, this paper establishes a Partially Observable Markov Process (POMDP) model based on an intersection environment. The Double Deep Recurrent Q-learning Network (DDRQN) which combines the Long Short Term Memory Model (LSTM) and the Double Deep Q-learning Network (DDQN) is used to solve tram signal priority strategy under the condition of missing or abnormal traffic data, enhancing the model's robustness. The simulation experiment shows that, compared with other DQNs, DDRQN's average reward is 44% and 23% higher than DQN and DDQN respectively at 90% observable state, proving that this algorithm has better anti-interference ability with incomplete data, reduces the average delay and improves the passing efficiency at the intersection.
Deep reinforcement learning method for POMDP based tram signal priority
2023-09-24
3094241 byte
Conference paper
Electronic Resource
English
Signal Timing Calculation Method for Tram Signal Priority
European Patent Office | 2018
|Tram priority at signal-controlled junctions
Online Contents | 2012
|Tram passive signal priority strategy based on the MAXBAND model
Springer Verlag | 2014
|Tram passive signal priority strategy based on the MAXBAND model
Online Contents | 2014
|