Date of Graduation
Statler College of Engineering and Mineral Resources
Lane Department of Computer Science and Electrical Engineering
Energy balance in electric power systems is continuously disrupted by constant demand changes due to customers' switching in and out, or loss of generating units. Load frequency control (LFC) is very essential for interconnected power systems in order to maintain the energy balance which is assessed through the Area Control Error, a signal that is made up of deviations from their nominal values of the system frequency and power area interchanges. Each balancing authority is responsible for its own energy balance in accordance with North American Electric Reliability Corporation (NERC) standards.;This thesis presents a novel approach to the LFC problem. An adaptive intelligent controller, or agent, changes the gains of a proportional-integral (PI) controller based on the operating conditions. The intelligence and decision making is provided by means of a reinforcement learning (RL) algorithms. This approach keeps the simple design of the PI controllers and in the mean time makes them more adaptive and applicable to different disturbances. Moreover, the developed controller can be applied to different systems with various parameters with almost no change in the controller design due to their ability to learn proper settings through interaction with the environment.;Each control authority should comply with NERC control performance standards CPS1 and CPS2. In order to comply with these standards and decrease the control cost, tight control should be prevented. The second approach in this thesis is to design a reinforcement learning based controller that tunes the gains of the PI controller in a way to achieve this goal. Simulations are performed in MATLAB/Simulink to demonstrate performance of all the proposed controllers.
Eftekharnejad, Sara, "Reinforcement learning-based control design for load frequency control" (2008). Graduate Theses, Dissertations, and Problem Reports. 4368.