ABSTRACT
A framework for the integration of dynamic programming, discrete simulation, and reinforcement learning is described to take advantage of the simulation procedures that self-adjust to meet goals for scheduling purposes in a manufacturing environment. The framework contributes by understanding the decision process for scheduling planning and integrating analytical techniques to learn effective scheduling policies. Furthermore, it is designed to achieve learning using this hybrid modeling approach and with the use of signals of the environment to measure the achieved state or goal and calculate the system’s performance criteria. An example of scheduling jobs in machines illustrates how the learning mechanisms allow the system to adjust to new situations. In addition, the technique in reinforcement learning, such as Q-Learning, is applied using a neural network to learn policies.
