Wind farm design has long been an application domain for evolutionary learning and optimization due to the complexity of the design space, and the discontinuities in the search space caused by the wake effects that make it hard to optimize analytically. Now, with a need to increase the renewable energy portfolio, existing wind farm layout approaches are being tested under a variety of scenarios. Newer models to evaluate layouts and newer constraints emerge, demanding more sophistication from the algorithms.
We propose the following competition to enable a basis of comparison for the existing algorithms and to encourage new ways to solve the wind optimization problem. As part of the competition, we will provide layout evaluators, wind field conditions, and baseline performances for the different layout optimization problems. Competitors are expected to contribute to an open source library for wind farm layout optimization.
This competition aims to bring more realistic problems to algorithm developers and to create an open source library useful beyond the scope of this competition. Historically, stochastic search has been the basis of the best approaches to solving the wind farm layout optimization problem, and as the industry continues to develop new models, constraints, and situations, it is timely that we propose a method for the integration of these two communities. This competition sets the stage for that integration.
Here, we present an open source library for wind farm layout optimization with an API built to support stochastic search algorithms. We will showcase the state of the art algorithms on a platform where the We will showcase the state of the art algorithms on a platform where the industry can bring their layout evaluation models. Furthermore, by focusing the problem parameters, we will create mechanisms for the type of uniform comparison that has so far been lacking in this field.
We will provide an API in C++, Java, and MATLAB for optimization algorithms that interacts with our open source wind layout evaluation model, and a basic scenario consisting of sample wind data and wind farm geography and obstacles. Different implementations of the wake model besides the provided versions will have to be approved and validated in advance. An automated submission system will allow competitors to see their rank on a public leader board.
The APIs and examples of use can be found on the API git repository.
The competition schedule is as follows:
March 15, 2014: Provided layouts and wake model code online
April 15, 2014: Automated submission system and live leader board online
June 1, 2014: Final submission deadline => Extended to June 15!
The competition goal will be split in two tracks:
While the first problem is the most common in the wind farm layout optimization community, the latter allows for a more realistic challenge. The cost-benefit analysis of adding a turbine to a field, here simply based on energy capture, could be expanded to include local constraints such as price, cabling, and terrain.
During the competition, contestants will be given:
As the layout evaluation is usually the most computationally part of the optimization, this allotment adds the challenge of computational resource efficiency. Competitors must design effective stop criteria for their algorithms.
Competitors will be allowed to enter either or both tracks and rankings and prizes will be given for each track. They will be compared on each layout to optimize according to the energy output for the first track and the number of turbines implanted on the second track (and energy output for ex-aequo) and ranked with point system:
The winner of each track will be the one with the maximum of points.
The slides presented during GECCO 2014 can be downloaded here.
We are doing our best to find a prize for each track of the competition. If you are interested in sponsoring this competition, please contact the competition organizers
Dennis Wilson: dennisw@mit.edu
Sylvain Cussat-Blanc: sylvain.cussat-blanc@irit.fr
Kalyan Veeramachaneni: kalyan@csail.mit.edu