Secure decentralized state estimation and control for smart distribution grids

Abstract

Modern power distribution networks face significant challenges due to the increasing integration of renewable energy resources and electric vehicles, leading to higher uncertainty and dynamics in grid states. This necessitates accurate state estimation for effective control actions. This dissertation proposes multiple novel approaches for distribution system state estimation (DSSE) that can be scalable to large distribution networks and control of inverter-based resources, offering a robust solution to these challenges. Current sparsity-based (DSSE) methods face limitations in scalability to large distribution networks or fail to effectively account for the temporal dependency within the measurement data. To address the scalability issue, we propose a hierarchical spectral clustering-based network partitioning algorithm followed by distributed compressive sensing (DCS)-based state estimation. This method utilizes an alternating direction method of multipliers (ADMM) for iterative information exchange among sub-areas for DSSE. We also demonstrate the robustness of ADMM-based distributed CS against various cyber-attacks through simulations on IEEE 37-bus and IEEE 123-bus networks. We address the second challenge by proposing two dynamic sparsity-based state estimation strategies, locally weighted matrix completion (LW-MC) and Bayesian matrix completion with Kalman filter prediction (BMC-KF). Evaluations on IEEE 37 and IEEE 123 bus test systems demonstrate that BMC-KF outperforms LW-MC. The above-proposed methods work efficiently for large distribution networks. However, they suffer from high computational complexity or slow convergence time. Hence, they become unsuitable for real-time state estimation. Therefore, we propose a model-based neural network called deep unfolding, for fast and accurate estimation of system states in low-observable distribution networks. This approach achieves accurate state estimation with reduced simulation time by incorporating linearized power flow constraints, and the efficacy is validated through extensive simulations on IEEE-123 bus and 559 bus networks. This deep learning approach does not provide any guarantee with regard to the estimate fidelity. Therefore, we propose a deep unfolding approach augmented with Bayesian neural networks to estimate the mean and covariance of system states, demonstrating its efficacy in estimating system states and providing associated confidence intervals through simulations on the IEEE 123-bus network. In addition to state estimation, the increasing integration of IBRs necessitates robust and decentralized control strategies to ensure system stability and efficient power sharing. The existing literature uses PID control for voltage and frequency restoration that suffers from asymptotic convergence. Hence, this dissertation introduces a novel approach that leverages game theory principles to design distributed secondary control for inverter-based islanded microgrids. The proposed approach optimizes power sharing among interconnected IBRs while enhancing system frequency stability. Simulation results on a four-bus microgrid demonstrate that the game-theoretic secondary control offers improved power-sharing accuracy, stability, and load balancing compared to the distributed-averaging proportional-integral (DAPI) method. In addition to this, to make a microgrid stable against cyber attacks, we propose an adversarial robust multiagent reinforcement learning secondary control that mitigates adversarial impacts and ensures system resilience. Together, these contributions address the evolving landscape of power distribution systems, by providing effective decentralized state estimation and control strategies to ensure stability and efficiency amidst large-scale integration of renewable energy resources.