Abstarct: Abstract The SEPIC converter is one of the DC-DC converters that can keep the output voltage constant under different input conditions. The controller design for the SEPIC converter is of interest due to its nonlinear nature, its fourth order and the dependence of the converter performance on load changes. In this thesis, the performance of the SEPIC converter is investigated, and in order to achieve its better performance, the cascade controller is designed with parameter adjustment using the deep reinforcement learning method. In the cascade controller, a combination of sliding model controller and PI controller is used. In the continuation of the research, it is shown that using the deep reinforcement learning approach to adjust the coefficients of the PI controller improves the performance of the SEPIC converter compared to the conventional method of adjusting the coefficients.