Harmonic Estimation and Forecasting in Sparsely Monitored Uncertain Power Systems using Probabilistic and Machine Learning Approaches

Abstract

In the face of transition towards Net-Zero emissions, increasing penetration level of power electronic (PE) interfaced renewable generation connections such as wind farms (WFs) and PV plants are influencing the uncertainties in power system operation and leading to additional often uncontrolled harmonic power flows. These potential harmonic issues could compromise power quality (PQ) and lead to significant financial losses, such as increased operational costs, equipment damage, and regulatory penalties, as well as potential interruptions in electricity supply. Meanwhile, given the financial and technical constraints, comprehensive monitoring of large power systems remains impractical, challenging system operators to estimate harmonics using limited available measurements. To address this problem, the probabilistic harmonic estimation and forecasting of harmonic propagation through sparsely monitored, PE-rich systems is becoming increasingly important. The first part of this thesis illustrates how system components should be modified to be suitable for harmonic simulations. A Monte Carlo-based probabilistic simulation methodology is proposed to incorporate the randomness (e.g., weather variations) and uncertainties (e.g., operational variabilities) of system operating conditions. Building upon these techniques, a Norton equivalent model is developed to represent large-scale WFs as a single harmonic source for probabilistic harmonic studies. The impacts of Norton equivalent impedance, harmonic injection magnitude and phase angle are explored. Appropriate modelling ranges and parameter distributions are recommended, based on simulation results and field measurements from single wind turbine generators. The second part of the thesis extends the general applicability of a recent methodology for estimating harmonic distortions from typical radial residential distribution networks to 33 kV - 132 kV networks, which is typically non-radial (meshed networks). This cost-effective methodology minimizes the need for extensive PQ monitor installations. It combines the Morris screening method for harmonic variation sensitivity analysis and an electrical distance-based sensitivity method to determine the optimal/minimum number and location of PQ monitors. This solves the problem of sub-optimal selection of monitoring locations when harmonic distortions and voltage drops are not highly correlated. The final part of the thesis proposes a novel approach for probabilistic harmonic forecasting in transmission networks with limited monitoring capabilities, utilizing sequential artificial neural networks (ANNs). This forecasting process incorporates both individual-order and total harmonic distortions (THD). A sensitivity analysis is performed to enhance the model's performance, focusing on identifying key parameters that critically influence its accuracy, robustness, and reliability. The main outcomes of this research facilitate the assessment of standard compliance and benchmarking, reduce the extensive monitoring installations, accelerate the evaluation of harmonic propagation and mitigation studies in uncertain PE-rich networks, as well as contribute to forecasting potential harmonic issues in future power networks.

Date of Award	6 Jan 2025
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Jovica Milanovic (Supervisor) & Kostas Kopsidas (Supervisor)