State Estimation and System Identification of Distribution Grids

Abstract

As we combat climate change, transitioning to a carbon-neutral economy has become imperative, particularly in the electricity sector. This shift necessitates moving from fossil fuel-based electricity generation to the integration of distributed energy resources (DERs) such as solar power, batteries, and electric vehicles. However, integrating DERs presents significant challenges, such as over-voltage; larger negative demand during sunny days resulting in negative power flows, known as the duck curve. Addressing these challenges requires the development of advanced control strategies for DERs within the distribution grid (DG). Effective distribution network management and operational strategies are crucial for facilitating a smooth transition to a sustainable and reliable energy system.

To improve the control of DERs, it is essential to have accurate knowledge of the current state of the DG and its network model. However, the system states are often unknown, and network line parameters are typically either unknown or inaccurate for a given DG. Therefore, estimating system state variables and line parameters is crucial for the successful management of DERs and maintaining grid stability. This thesis focuses on distribution system state estimation (DSSE) and line parameter estimation, utilizing data from readily available devices, such as smart meters (SMs) in the DG.

An accurate and computationally efficient power flow model is essential for any estimation problem. Chapter 3 proposes a new power flow model called the modified Distflow model, a modified version of the widely known Distflow model. It incorporates line losses and is expressed by explicit equations, eliminating the need for an iterative solver, and thereby maintaining accuracy at a lower computational cost.

Chapter 4 introduces a comprehensive DSSE framework and proposes a new DSSE approach using the modified Distflow model. It focuses on estimating the system state variables with limited availability of measurements. An analysis of the accuracy with measurements available at different locations has been performed. All required gradients for the approach are provided. Here it is assumed that the network parameters are known.

However, line parameter information is typically unavailable in the DG. Hence it is important to estimate the line parameters. Chapter 6 proposes a new method for estimating the line parameters and system state variables with noisy measurements at all nodes. Additionally, the voltage magnitudes are estimated accurately. This is achieved using a novel combination of Linearized Distflow and expectation maximization (EM) with Bayesian regression.

Despite achieving accurate results in the previous study, it utilized a linearized power flow model that neglects line losses. This approach requires more data and higher accuracy can be achieved with a power flow model considering losses. Hence, in Chapter 7, the non-linear modified Distflow model is utilized, including line losses. As a non-linear model, it requires a different solution approach. EM is still employed, but a first-order Taylor expansion to approximate the distribution of state variables is implemented using the efficient square root form. This changes the cost function and improves accuracy compared to the method proposed in Chapter 6 and other similar studies.

In the previous chapters, noisy measurements at all nodes were considered for line parameter estimation. Chapter 8 addresses the challenging case of missing measurements in the DG while estimating line parameters. While EM is used as a solver, handling missing data significantly alters the problem formulation. Despite these challenges, accurate results are achieved with as little as 50% measurement coverage. Accurate results were achieved even with noise levels of 4.5%. Additionally, the impact of missing measurements at different locations in the DG is analyzed to assess their impact on estimation accuracy. Show more