Grid/load integration of photovoltaic system with shadow effect for energy continuity

As the grid integration of renewable energy sources increases, improving the low voltage ride-through capability of these systems has become critically important for grid stability and energy security. This thesis addresses the enhancement of the low voltage ride-through capability of a grid-connected photovoltaic system. The two-stage three-phase grid-connected photovoltaic system is modeled in Matlab/Simulink. Three different deep reinforcement learning based controllers distinguished from conventional methods by their ability to adapt to dynamic and uncertain grid conditions and to generalize across various fault scenarios are developed. Their performances under balanced and unbalanced faults are compared with one another and with a proportional-integral controller optimized by the symbiotic organism search algorithm. The results show that transfer learning, which improves generalization capacity, provides superior performance. In addition, to increase the robustness of the photovoltaic system, the low voltage ride-through capability is considered together with shading conditions arising from environmental factors. The deep reinforcement learning method enhances system robustness by increasing generalization capacity compared with the perturb and observe algorithm.