A model-based deep reinforcement learning approach to the nonblocking coordination of modular supervisors of discrete event systems

Yang, Junjun; Tan, Kaige; Feng, Lei; Li, Zhiwu

doi:10.1016/j.ins.2023.02.033

Cited by 6 publications

(2 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Q table for the transfer line example is approximated by a fully connected three-layer neural network, where the numbers of neurons of the input layer, the middle layer, and the output layer are 6, 8, and 8, respectively. The details of the DQN algorithm are elaborated in [46].…”

Section: ) Deep Reinforcement Learning Frameworkmentioning

confidence: 99%

See 1 more Smart Citation

Reducing the Learning Time of Reinforcement Learning for the Supervisory Control of Discrete Event Systems

et al. 2023

Self Cite

View full text Add to dashboard Cite

Reinforcement learning (RL) can obtain the supervisory controller for discrete-event systems modeled by finite automata and temporal logic. The published methods often have two limitations. First, a large number of training data are required to learn the RL controller. Second, the RL algorithms do not consider uncontrollable events, which are essential for supervisory control theory (SCT). To address the limitations, we first apply SCT to find the supervisors for the specifications modeled by automata. These supervisors remove illegal training data violating these specifications and hence reduce the exploration space of the RL algorithm. For the remaining specifications modeled by temporal logic, the RL algorithm is applied to search for the optimal control decision within the confined exploration space. Uncontrollable events are considered by the RL algorithm as uncertainties in the plant model. The proposed method can obtain a nonblocking supervisor for all specifications with less learning time than the published methods.

show abstract

Section: ) Deep Reinforcement Learning Frameworkmentioning

confidence: 99%

“…and the values of rewards are shown in Table 3. Other parameters for the DQN algorithm are the same as those in Table 3 in [46]. Fig.…”

Section: ) Deep Reinforcement Learning Frameworkmentioning

confidence: 99%