MODELING AND STABILITY ANALYSIS OF A DC MICROGRID WITH MACHINE LEARNING TECHNIQUE

Abstract

Stability is a critical aspect of any power system, including DC microgrids, as it prevents power quality issues, equipment damage, and system failures. The incorporation of multiple sources, such as Distributed Generation Systems (DGS) and Energy Storage Systems (ESS), into a DC microgrid can also contribute to inefficiencies or instability, especially if synchronization between various sources and loads is not properly maintained. As DC microgrids grow larger and more complex, with multiple sources and loads, maintaining stability becomes increasingly difficult, particularly when evaluating stability margins using traditional methods such as small signal stability analysis. This conventional approach relies on determining eigenvalues from the system model and assessing stability based on the location of these values. However, as the number of eigenvalues increases in large-scale systems, it becomes challenging to effectively evaluate the stability margin. To overcome these challenges, this thesis proposes the use of artificial intelligence (AI), particularly machine learning (ML), to enhance the stability analysis of DC microgrids. Eigenvalues, along with voltage and current data, is generated using the DC microgrid Simulink model and MATLAB code. Finally, the machine learning model is used to train the generated data to improve system stability. This approach is particularly beneficial for handling large, complex systems where traditional methods fall short. In the proposed model, a Simulink-based DC microgrid consisting of three sources and three loads interconnected in a ring topology is developed. The system's voltage, source current, and branch current is recorded under both stable and unstable conditions. Generated data's are used to train ML using the ensemble models, and federated learning approach, allowing the model to learn from decentralized data sources without compromising privacy. The effectiveness of this machine learning based stability analysis is validated through detailed simulation studies, and the results demonstrate high accuracy in predicting system stability. The federated learning approach is preferred in this research since it combines the benefits of ensemble approaches to improve performance overall. The method ensures enhanced prediction accuracy, scalability, and privacy protection by utilizing federated learning.