Abstract

In this paper, a reinforcement learning-based optimal control is developed for the drug administration of biological phenomena in chemotherapy cancer treatment. The treatment is considered as a class of unknown discrete-time systems when the input: drug administration and the output: tumor cells population are only utilized to design the proposed controller. Resulting, a full-state observer is completely neglected. The controller is established by the actor–critic architecture containing two fuzzy-rules emulated networks when IF-THEN rules are imposed by human knowledge according to pharmacokinetic and pharmacodynamic behavior. Furthermore, the discontinuous reward function is proposed to derive the online learning laws that guarantee the robustness and the convergence of adjustable parameters. The validation results are conducted by numerical systems according to the robustness of the group of patients and the closed-loop performance altogether with comparative results.