A reinforcement‐learning approach for individual pitch control

Abstract

Individual pitch control has shown great capability of alleviating the oscillating loads experienced by wind turbine blades due to wind shear, atmospheric turbulence, yaw misalignment, or wake impingement. This work presents a novel controller structure that relies on the separation of low-level control tasks and high-level ones. It is based on a neural network that modulates basic periodic pitch angle signals. This neural network is trained with reinforcement learning, a trial and error way of acquiring skills, in a low-fidelity environment exempt from turbulence. The trained controller is further deployed in large eddy simulations to assess its performances in turbulent and waked flows. Results show that the method enables the neural network to learn how to reduce fatigue loads and to exploit that knowledge to complex turbulent flows.When compared to a state-of-the-art individual pitch controller, the one introduced here presents similar load alleviation capacities at reasonable turbulence intensity levels, while displaying very smooth pitching commands by nature.

Comment on wes-2024-56

“…The blade load-based estimator is verified numerically. What we consider as the true measurements of the out-of-plane bending moments are provided by LES performed with an in-house wind turbine simulation environment (Chatelain et al, 2017;Balty et al, 2020;Coquelet et al, 2022). The latter relies on a Vortex Particle-Mesh method (Chatelain et al, 2013) in which wind turbine blades are modeled by Immersed Lifting Lines (Caprace et al, 2019).…”

“…Russell et al (2024) additionally demonstrated how LiDAR-assisted feedforward individual pitch control was improving load alleviation results for turbines operating in freestream conditions. Coquelet et al (2022) investigated the case of waked turbines and concluded that it provided challenges, calling for new IPC schemes accounting for partial or total wake impingement. When it comes to wake mixing, the controllers act in a dynamic way.…”

Model-Free Based Pitch Control of a Wind Turbine Blade Section: Aerodynamic Simulation

“…Strategies based on learning methods such as neural networks and reinforcement have been proposed e.g. in [6], [7], iterative learning control in [8], neural based PID controllers in [9], feed-forward model predictive control, in [10], and adaptive high order sliding-mode controller [11]. The direct control of the blade using local inflow measurements has been studied in [12] and cascaded controllers has been proposed in [13].…”

Handling Individual Pitch Control within an Actuator Disk framework: verification against the Actuator Line method and application to wake interaction problems

“…Numerical approaches offer a convenient framework for investigating IPC strategies. The highest fidelity tools that also capture the wake response usually consider Large Eddy Simulation (LES), combined to discrete type line approaches [5,6], such as the Actuator Line (AL) or Lifting Line methods. The use of these wind turbine models remains computationally affordable at the scale of a small wind farm.…”