# Optimization Programming - Which Algorithm?

Tags:
1. Aug 18, 2014

### grantwilliams

Hey guys! I need a little help in continuing my research i did this summer.

So a little background information:

This summer i conducted research at a REU where i focused on optimizing the power output of a wind farm by modeling wind turbine locations within a constrained space. I wrote a genetic algorithm that was pretty efficient at finding wind turbine layouts that produced energy more effectively than an evenly spaced grid with a wind direction modeled based on wind direction probabilities found on a real wind farm. I really enjoyed the research and the math, physics, and programming that went along with it and i would like to improve the model. I have a couple of ways I'm considering doing so.

The model i wrote is a vectorized, and parallelized genetic algorithm with a relatively simple wake model based on the distance and angle between two turbines (not a computation fluid dynamics wake). My research mentor allowed to me to come up with my own project idea and implement it fully by myself, but she also believes the work i did is good enough to publish so im currently going through the steps of editing my paper for publication. I would like to conduct more research in the same field and attempt to ready another paper for publication by the end of this school year.

These are some of the ideas i currently have:

• Implement a more accurate wake model and write the function in a way that it can be used on graphics cards
• Add a local serach function within the genetic algorithm
• Try a Articifial Bee Colony Algorithm (ABC)
• Write a model that can import GIS maps and attempt to intelligently avoid locations where trubines couldnt be placed based on geographical features
• Code the Genetic Algorithm to adaptively change its mutation rate based on effectiveness of solutions
• Find a way to model the upper boundary of a wind turbine farm's power output assuming there are wake interactions

The idea im most interested in is finding a way to model the upper bound of the wind farm's power output. No one has published a way to do so yet, and it would allow researchers to test their model against a perfectly optimized layout. For an example of what im talking about see theThe lower boundary of the traveling salesman problem

2. Aug 19, 2014

### MrAnchovy

Interesting project, well done.

Are you currently using the wake model used in WAsP described here? Moving to a CFD solution would introduce many significant problems including (i) much of the software and algorithms used are proprietory - consultancies make a lot of money advising on wind farm design; (ii) any CFD calculation is going to increase the cost of your objective function by many orders of magnitude, even if offloaded to GPUs.

Besides, "if it ain't broke, don't fix it" - have you looked at the predicted results and compared them with field data?
Terrain has a huge effect on optimal turbine siting - indeed this is one area where CFD is routinely used. I'm not sure what you mean by "couldn't be placed based on geographical features" though?
According to Betz's law this is approximately 59% of the kinetic energy of the wind entering the farm. Wake interactions don't reduce this further. It would also be useful to compare your optimised layouts with other concepts, such as a regualr grid oriented to maximise power from the prevailing wind direction and a Poisson grid.