Artificial Intelligence Aided Adaptive Control Strategies for Converter Based Power Systems

European Union (EU) has set targets to reduce greenhouse gas emissions by at least 55% (compared to 1990 levels) by 2030. EU aims to become climate-neutral by 2050, with the exception of Denmark, which has set an even more ambitious goal of achieving climate neutrality by 2045. In the coming years, one of the inevitable efforts to achieve these targets is the expansion of Renewable Energy Source (RES) such as wind turbines and PhotoVoltaic (PV) in the electrical power grid. Smart AFE converters play a crucial role as the key interface components connecting RESs to the grid. Increasing the number of RESs results in small-signal stability issues, especially in cases where the RES is connected to a weak grid.

In certain cases, stability challenges arise from variations in grid conditions, such as changes in grid impedance. To enhance system reliability and performance, it is essential to adapt the control parameters to the new grid conditions. This thesis presents a novel method for the active identification and adaptive tuning of multiple control parameters using ANNs. The proposed method utilizes the EKF algorithm to identify the grid parameters. These identified parameters are subsequently utilized in an ANN-based parameter searching algorithm. This algorithm ensures pole-tracking-based stabilization by determining the optimal solution for the PI control parameters. The main innovation of this method lies in its capability to perform online tuning of multiple control parameters in AFEs. To validate the effectiveness and performance of the proposed approach, a comprehensive evaluation is conducted, consisting both simulation results and experimental tests conducted at PowerLabDK. The results demonstrate the effective and adaptive stabilization capability of the proposed method in the presence of varying grid impedance.

The active (intrusive) identification method, due to the injected disturbance, may temporarily degrade the performance of the converter. Moreover, when the amplitude of the injected disturbance is large, it can potentially activate nonlinearity in the system’s behavior. This thesis proposes an online method to address these issues by tuning the parameters of the model predictive controller for AFE using a passive identification algorithm. In the proposed method, the grid is represented by a state space structure. The incorporation of augmented state variables, besides the system parameters, yields a dynamic state-space model characterized by nonlinearity. The EKF is utilized in this research to observe the state variables and identify the system parameters. While the concepts of the EKF and MPC are employed to control the AFE system, the algorithm’s performance relies on the selection of appropriate weighting factors. To determine the optimal weighting factors, an innovative algorithm based on ANN is developed. This ANN-based algorithm is specifically designed to identify the optimal weighting factor from a predefined set of options, taking into account the unique characteristics of the grid at each iteration. An offline Particle Swarm Optimization (PSO) algorithm is executed to generate the required data for the predefined set of weighting factors options. The proposed method offers several advantages, which can be enumerated as follows: identification of grid impedance and inductance, guaranteeing hard constraints on the amplitude of input and output variables, enhancing the performance of CCS MPC through parameter updates in each iteration, and increasing the prediction horizon. The proposed method was implemented experimentally in PowerLabDK, and the results of the experiment demonstrate that the algorithm effectively stabilizes the grid-tied converter in weak grid conditions, even when facing significant changes in grid impedance.

In addition to addressing stabilization concerns, the converters have the capability to improve grid performance. Another challenge in grid-tied converters involves unbalanced conditions, which can result in power losses and power quality issues. This thesis also presents an IoT-based reference generator for voltage unbalanced compensation. The proposed method takes into account the challenges coming from packet losses and time delays. The measured current and voltage data taken at the remote transformer are transmitted to the edge devices through the cloud interface. The edge device controller compares the measured data from the transformer with the reference data to generate an appropriate reference signal for the converter. However, there are two primary challenges associated with cloud-based control: packet losses and delays in response time. This thesis introduces a novel sampled-data secondary controller, designed to be robust against packet losses and time delays, implemented in edge devices. The Intelligent Redesign (IRD) method aims to convert a pre-designed analog controller into its equivalent digital counterpart. The method establishes sufficient conditions, expressed in terms of Linear Matrix Inequality (LMI), to ensure both the stability of the closed-loop digital system and the minimization of errors. The secondary controller is designed by solving the LMI conditions. To assess the applicability and efficiency of the proposed approach, the method is experimentally implemented in PowerLabDK.