Distributed and Data-Driven Decision-Making for Sustainable Power Systems

Abstract

With a large-scale integration of renewable generation, especially wind power and solar energy, significant uncertainty and volatility have been introduced to the control and operation of electric power systems. The increase of power electronics converter-based renewable generation dramatically reduces the inertia level of power systems, leading to faster dynamics and greater control difficulty. Meanwhile, the rapid proliferation of distributed energy resources (DERs), such as solar panels, wind turbines, energy storage, electric vehicles, thermostatically controlled loads, etc., has been witnessed in distribution grids and the demand side. The coordinated dispatch of massive DERs is expected to unlock significant power flexibility that can accommodate a high penetration of renewable generation and enhance system-level efficiency, reliability, and resilience.

This dissertation studies the schemes that exploit the power flexibility of massive DERs to improve the control and operation of sustainable power systems and focuses on addressing the two critical challenges: 1) uncertainty and unknown model information, and 2) the scalability issue of coordinating a vast amount of DERs. Accordingly, this dissertation develops data-driven and distributed decision-making algorithms to tackle these two challenges. In particular, this dissertation studies four key problems in sustainable power systems, including frequency regulation, voltage control, power flexibility aggregation, and demand response in Chapter 2 - Chapter 5, respectively.

Chapter 2 develops a fully distributed automatic load control (ALC) algorithm for frequency regulation by reversely engineering power system dynamics as primal-dual solution schemes. The ALC algorithm only needs local measurement and local communication, and globally converges to an optimal operating point that minimizes the total cost, restores the nominal frequency and the scheduled tie-line power flows, and respects the load capacity limits and the line thermal constraints. The global exponential convergence and tracking performance of ALC are analyzed as well.

Chapter 3 proposes a model-free optimal voltage control (MF-OVC) algorithm based on projected primal-dual gradient dynamics and the continuous-time zeroth-order method (extremum seeking control). The MF-OVC algorithm i) operates purely based on voltage measurements and does not require any other model information, ii) can drive the voltage magnitudes back to the acceptable range, iii) satisfies the power capacity constraints all the time, iv) minimizes the total operating cost, and v) is implemented in a decentralized fashion. We prove that the MF-OVC algorithm is semi-globally practically asymptotically stable and is structurally robust to small measurement noises.

Chapter 4 studies distribution-level power flexibility aggregation strategies to effectively harness the collective flexibility from massive DERs for transmission-distribution (T-D) interaction. We propose two schemes to model and quantify the aggregate power flexibility, i.e., the net power injection achievable at the substation, in unbalanced distribution systems over time, considering network constraints and multi-phase unbalanced modeling. A distributed model predictive control (MPC) framework is then developed for the practical implementation of T-D interaction.

Chapter 5 studies the user selection problem and the air conditioner (AC) load control problem in incentive-based residential demand response (DR). The critical challenge is that the user opt-out behaviors are uncertain and unknown in practice. Hence, we adopt online learning techniques to learn the user behaviors through interactions and observations, and integrate them with the decision-making process. Based on the Thompson sampling framework, we propose online learning and decision algorithms to select the right users and optimally control AC loads in DR events.