Data-driven modeling, analysis and control of non-linear transient dynamics of microgrids
Restricted (Penn State Only)
- Author:
- Nandakumar, Apoorva
- Graduate Program:
- Electrical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- February 06, 2024
- Committee Members:
- Madhavan Swaminathan, Program Head/Chair
Minghui Zhu, Major Field Member
Bin Li, Major Field Member
Yan Li, Chair & Dissertation Advisor
Daning Huang, Outside Unit & Field Member
Nilanjan Ray Chaudhuri, Major Field Member - Keywords:
- Data-driven modeling
Sparse Identification
Microgrids
Model Predictive Control
Transient Dynamics
Unbalanced Operations
Deep Neural Networks - Abstract:
- The increase in the global energy demand is one of the key factors that necessitates the developments in the field of renewable energy resources. Research in the field of microgrids has seen significant progress in recent years. It is driven by increased deployment, integration of renewable energy and energy storage, smart grid integration, innovative ownership models, grid-interactive buildings, standardization efforts, and a focus on resilience and emergency preparedness. These developments contribute to a more sustainable, reliable, and decentralized energy landscape. The high penetration of power-electronic interfaces in Distributed Energy Resources (DERs) integration makes microgrids highly susceptible to disturbances, causing severe transients, especially in the islanded mode. While the details of the system topology are easily obtainable, it is rather difficult to develop a high-fidelity model that represents the transient dynamics of the different DERs. A novel modularized Sparse Identification of Non-linear Dynamics (M-SINDy) algorithm is developed for effective data-driven modeling of the nonlinear transient dynamics of microgrid systems. The M-SINDy method realizes distributed discovery of nonlinear dynamics by partitioning a higher-order microgrid system into multiple subsystems and introducing pseudo-states to represent the impact of neighboring subsystems. This specific property of the proposed algorithm is found to be very useful while working with re-configurable and scalable microgrids. Dynamic discovery of system transients from measurements can be beneficial for designing control strategies that improves the overall microgrid stability and reliability. Prediction of future states is a difficult, but an essential tool in power systems for determining different control strategies that can aid in maintaining the transient stability of the overall system following a contingency. To understand the system and predict these transient dynamics of a microgrid in different operation modes, an extension of the M-SINDy method - Physics-informed hierarchical sparse identification has been proposed. The developed algorithm has a multi-layered structure to reduce the overall computational cost required to obtain accurate model dynamics. The different functions that affect the system dynamics are developed in the primary layer using the measured data. The terms developed in the primary layer are fit in the secondary layer to determine the exact dynamics of the system subject to different disturbances which can be leveraged to predict the system’s future dynamics. The primary motivation to develop the data-driven prediction model is to incorporate the prediction data into a Model Predictive Control (MPC) framework that can generate an optimal control input to enhance the transient stability of microgrids. This MPC controller is augmented with the conventional droop control for frequency stabilization. Given the inherent fluctuations in typical microgrid operations, stemming from factors such as varying load demands, weather conditions, and other variables, reachability analysis is performed in this work. We aim to facilitate the design of a data-driven prediction model that can be leveraged to implement an effective control strategy to ensure the efficient working of microgrids for a wide range of operating conditions. Another potential challenge in the study of microgrids is caused by system imbalances. Variable loads, single phase DERs, network variations, etc. are some of the major contributing factors which are responsible for making the system unbalanced. Unbalanced transients in a microgrid can result in conditions that can impact the connected loads and damage the system equipments. Minimizing the overall imbalance in the system is important for maintaining the system’s stability, reliability, and optimal performance. We developed a data-driven model using a domain-enriched Deep Neural Network (DNN) architecture that can accurately predict the voltage dynamics in an unbalanced microgrid system, based on dynamic power flow computation. A supervisory control strategy is developed to reduce the imbalance by modulating the power generation of dispatchable units within the microgrid. The overarching purpose of this thesis is to explore the advancements in data science and provide an insight on the role of machine learning in transforming power systems for operation optimization and system enhancements. The integration of data science in microgrids allows for a more informed decision-making on resource allocation and builds a more resilient and sustainable energy infrastructure. It accelerates the transition to a more flexible, decentralized, and intelligent grid.