Reinforcement learning in pulse energy converters, process knowledge learned to transfer perspective framework

The article presents the task-based methodological framework of reinforcement learning (RL) for the control of energy transfer using the example of pulse energy converters (PEC). The focus is on the evaluation aspects of RL design as reflected in the formulation of knowledge learned and the ability to transfer for safe system application. The designer’s awareness of the process knowledge that is critical to the design of the control system begins with the definition of the necessary initial knowledge about the process and continues with the transformation of the RL knowledge, including storage and internal or external transferability. The transferable knowledge is inductive knowledge and not knowledge about the hyperparameters of the higher-level RL formulation, but it does contain this information. Not as a method for modelling deterministic certainty, but for modelling deterministic uncertainty. Modelling hybrid systems provides the imaginative deterministic foundations for the implementation of heuristics and RL formulations. The mild mathematical expressions in the form of definitions, assumptions, remarks and theorems serve to support the idea of transferable knowledge formulations that start from already inherited knowledge. The emphasis is on the inductive acquisition of process knowledge and the awareness of the epistemic connotation of the learning algorithm to ontics with the clear transformation. Markov Decision Processes (MDP) is a clear mathematical tool and modelling framework that merges the mathematical spaces of process states with our probability spaces and heuristics-based decision making in real time.