Advancing Intelligent DC Microgrids: AI-Enabled Control, Cyber security, and Energy Management

Microgrids are regarded as one of the cornerstones of the future smart grid, contributing to the development of zero-carbon cities and diverse energy systems. They consist of various distributed generation resources (DERs), energy storage systems (ESSs), and loads, serving as energy routers to enable the grid to withstand major disasters and enhance the energy security and resilience of the country. DC microgrids are distinguished among different microgrid configurations due to their higher efficiency, compatibility with renewable energy sources, scalability, resilience, cost savings, and alignment with the growing use of DC systems.
In DC microgrids, various modules operate both independently and coordinately to achieve different operational objectives. The hierarchical control framework is recognized as a standardized solution for DC microgrids. The local control at the primary control layer manages voltage and current control, as well as preliminary power sharing among different units. The secondary control layer regulates the voltage and ensures accurate power sharing. The tertiary control layer is responsible for advanced control goals, including energy management, economic dispatch, and power flow control, among others. However, factors such as the complex model, parameter availability and accuracy, dynamic operating conditions, and related aspects pose challenges to the existing DC microgrid control system.
This research aims to revolutionize microgrid operations by incorporating advanced artificial intelligence (AI) technologies into the existing multilayer control framework, addressing the aforementioned challenges. The research encompasses three primary areas: 1) at the local control level, a model-free learning-based controller operating in real-time using standard commercial microprocessors, ensuring optimal converter control and enabling independent unit operations; 2) at the secondary control layer, enhancing the system’s cybersecurity by identifying and detecting intelligent cyber-attacks using a data-driven approach; and 3) at the system level, the autonomous energy management of the system with an innovative battery degradation modeling approach under dynamically changing operating conditions. Ultimately, this research endeavors to develop an AI-enabled control system that enhances local control, cybersecurity, and energy management of DC microgrids.
At the primary control layer for local control, the interfacing power converter ensures proper operation at the device level. Conventional advanced controllers, such as finite-control-set model predictive control (FCS-MPC), are extensively applied due to their straightforward and flexible formulation of control objectives and constraints. However, this approach suffers from parameter sensitivity, unmodeled dynamics, and time-consuming optimization of the cost function. To address these issues, a guideline based on reinforcement learning (RL), aiming to automate weighting factor design in FCS-MPC, is introduced. In addition, a model-free self-learning controller for the converter is proposed, achieving comparable control performance to conventional FCS-MPC without requiring prior system knowledge. Furthermore, a safety framework is proposed to enforce physical limits and enhance the learning efficiency of RL agents. These findings offer opportunities for implementing online RL-based controllers in practical power electronics systems.
Furthermore, at the secondary control layer, the communication-based distributed control, widely adopted for module coordination control, potentially exposes DC microgrids to cyber-attacks. Various advanced cyber-attack detectors have been proposed, yet they face critical challenges when confronted with astute adversaries introducing innovative attack patterns. To address this, RL is first employed to identify the vulnerabilities of the system to cyber-attacks by emulating intelligent attackers, and autonomously generate stealthy attack patterns that bypass conventional cyber-attack detectors. Conversely, a data-driven cyber-attack detector is proposed to enhance the capability of state-of-the art cyber-attack detectors by identifying these attacks. Consequently, an ML-enabled framework for cyber-attack identification and detection is developed. This development opens up possibilities for iteratively implementing the proposed cybersecurity framework, enabling the identification and detection of a broader spectrum of cyber-attacks in DC microgrids.
Ultimately, at the tertiary control layer, accurate modeling of battery degradation is crucial for the economic operation of DC microgrids to minimize the overall operation cost. Different from the conventional methods that employ a single-stage battery degradation model, this research proposes a multi-stage battery degradation modeling method to accurately capture the varying aging patterns throughout a battery’s lifetime, thereby enabling more precise evaluations of capacity losses under diverse operating conditions. In addition, to adapt to varying operating conditions and mitigate system uncertainty, this project explores an RL-based energy management strategy that takes into account battery degradation for the economic and autonomous operation of DC microgrids.
In conclusion, the contributions of this Ph.D. project include the development of an AI-enabled hierarchical control framework, spanning from model-free optimal power converter control at the primary control layer to a data-driven cyber-attack identification and detection scheme at the secondary control layer, and extending to an RL-based energy management strategy incorporating a multi-stage battery degradation model at the tertiary level. The proposed AI-enabled multilayer control framework is expected to significantly advance the development of more intelligent, secure, and efficient DC microgrids.