Learning-based voltage regulation in direct current micro-grids with constant power loads

Power electronics-based direct current (DC) distribution systems are becoming widespread due to their advantages in power systems such as automobiles, aircraft, ships and satellites. However, power electronics converter loads can cause instability in the system when tightly controlled. Therefore, it is of great importance to achieve fast and dynamic response in DC microgrids (MG) to ensure stable operation of the system. Within the scope of the thesis study, firstly, the risk of constant power load (CPL) with negative impedance characteristics, which causes instability in DC DC, was examined. It has been determined that a DC/DC converter with an ideal arrangement can lead to unstable behavior when combined with an input filter. Various situations have been examined and stability conditions have been established. In addition, transfer functions of the unstable dynamic behavior of DC/DC converters have been derived by relating them to CPL. These transfer functions have been used to analyze open and closed loop instabilities under control changes and resistive load situations. Various control methods have been proposed in the literature for DC/DC power electronic converters loaded by CPLs. Passive damping strategies are simple and effective, but are costly and limited by physical constraints. Active damping methods have the ability to stabilize the system by mimicking virtual elements. However, these methods run the risk of compromising load performance. Therefore, the use of nonlinear control methods becomes imperative to ensure stability in terms of large signal. In order to demonstrate the effectiveness of the controller designed in the thesis study, simulation results were compared with robust control strategies commonly used in the literature. The first of these strategies, backsteping control (BSC), is an effective nonlinear control method to solve the voltage regulation, stability and reference power tracking problems of a DC/DC amplifier. Working in conjunction with the BSC control, the nonlinear disturbance observer (NDO) is used to predict uncertain load variation, improving system dynamics. The BSC controller we use for comparison purposes is designed using an adaptive backstepping algorithm. Using coordinate transformation and NDO, control parameters are preset to ensure system stability and monitor load variation. The second one, the model predictive control (MPC) method, is known as an effective method to improve performance in the field of power converters control. The basis of the MPC method is a method that solves an optimization problem involving future behavior over a certain period of time and obtains a control signal by considering nonlinear situations. MPC calculates real-time optimal control decisions based on predictions. In the literature, the MPC algorithm has been applied to most of the power converters and it has been stated that it gives better results than traditional methods. In this thesis study, the learning-based nonlinear control model developed using artificial neural network (ANN) will provide a solution to the CPL problem. The designed controller aims to design a robust control strategy against the negative effects of CPL loads of an independent DC MS managed with a decentralized control hierarchy. In the learning-based designed model, the voltage on the DC source side is transmitted to a mixed load consisting of CPL and ohmic resistor via a DC/DC boost converter. Learning-based control design is an effective method to ensure stability, fast response and accurate voltage monitoring during CPL load changes, as shown in simulation studies performed in MATLAB/Simulink. This design offers advantages such as lower voltage oscillations and faster settling time compared to traditional control methods and the BSC and MPC control methods mentioned above. The design of the ANN involves training a neural network using 200000 lines of data obtained from the BSC controller. The trained artificial neural network is then used in model design for voltage regulation and control. As a result, extensive simulation studies are performed to evaluate the performance of the designed control strategy. The results obtained are compared with those obtained using conventional PI, BSC and MPC controllers. Simulation results clearly demonstrate the effectiveness and superiority of the proposed control strategy over PI, BSC and MPC.