Machine learning-based wind turbine control systems for demand-oriented scenarios

Li, Tenghui (2024) Machine learning-based wind turbine control systems for demand-oriented scenarios. PhD thesis, University of Glasgow.

Abstract

When wind power has an increasing share towards a 100% renewable society, wind energy conversion systems (WECSs) need to consider a requirement of the grid generation-consumption equilibrium, i.e., wind turbines (WTs) should be able to adjust their output according to power demand. However, current WTs focus on maximum power capture, which has intrinsic disadvantages in power scheduling. Hence, this study aims for machine learning (ML) based control systems that realize flexible wind capture in demand-oriented scenarios.

First, this study reviews various turbine components and establishes corresponding control models. Second, aerodynamic modelling relies on an artificial neural network (ANN) to predict thrust, torque, and power from the turbine state. Subsequently, a novel online power strategy (OPS) based on an aerodynamic model solves the 2-degree-of-freedom (DOF) optimization of the rotor speed control (RSC) and pitch angle control (PAC), which has two implementations: power reference point tracking (PRPT) and reinforcement learning (RL). Besides, the OPS has a local linearization to estimate thrust and torque sensitivities for optimal control configuration. When a wind processing unit updates wind velocity and direction signals, the OPS receives the velocity signal to calculate the 2-DOF solution and unwraps the direction signal as the command of the yaw angle control (YAC), which achieves a complete 3-DOF regulation of rotation, pitch, and yaw. Besides, the OPS framework has four control implementations: one model-free controller and three model-based controllers. In addition, our turbine control system can integrate wind forecasting to enhance the capability of handling wind stochastics.

The case study verifies and proves the accuracy and reliability of the OPS-based control framework in four simulation cases. The proposed turbine control can track different power targets and ensure reliable output in stochastic winds. Therefore, this control framework can contribute to intelligent WTs for large-scale grid integration.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: T Technology > TA Engineering (General). Civil engineering (General)
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Supervisor's Name: Yang, Dr. Jin and Ioannou, Dr. Anastasia
• Date of Award: 2024
• Depositing User: Theses Team
• Unique ID: glathesis:2024-84643
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 31 Oct 2024 14:25
• Last Modified: 31 Oct 2024 14:31
• Thesis DOI: 10.5525/gla.thesis.84643
• URI: https://theses.gla.ac.uk/id/eprint/84643

Related URLs

• Article DOI
• Article DOI
• Article DOI
• Article DOI