Reinforcement learning could help airborne wind energy take off

Offtopic
1

@ufechner7 Are you aware of this paper on using reinforcement learning for wind kites?

Is Julia being used here. There is not an obvious link to the code.

1 Like
2

Paper here:

1 Like
3

No, I was not aware of this paper. It is very new.

Thanks for sharing! This year I will be working on airborne wind energy again, last year I was working on offshore wind.

4

Instead of reinforcement learning, why not use (stochastic) model predictive control? Wind turbine dynamics are well-defined, I’m not sure what part requires reinforcement learning.

5

Airborne wind energy systems differ from wind turbines. They use one or more wings, connected to the ground by one or more tethers. The dynamics are not well-defined at all, but change constantly depending on the tether tension, the angle of attack, the height, the speed, the position in the wind window etc.

So reinforcement learning might make sense.

And yes, it is not needed. MPC also works, but lacks strong stability guarantees. Simple single- or dual-loop PID controllers also work when combined with nonlinear dynamic inversion. Many ways to achieve the goal.

But reinforcement learning is the most fancy approach these days.

See also: Open Source AWE Simulation, Optimization and Control · GitHub

1 Like