APIs for the Wind Farm Layout Optimization competition

These APIs are for the GECCO 2014 Wind Farm Layout Optimization Competition. Competitors are expected to use these functions to determine the fitness of the layouts they are optimizing. All APIs keep a running count of the number of evaluations made, which will be used in the comparison of competing algorithms. We hope competitors will find these APIs useful and clear and will not try to subvert them.

Between the C++, Java, and MATLAB implementations, the basic operation is the same and consists of only two main methods:

• Initialize/construct: The evaluator class must be constructed and initialized with a wind scenario number. We have provided 20 sample Scenarios, which can be found in Scenarios, the latter 10 of which are the same as the first 10 with the addition of two obstacles.

• Evaluate: This method takes a layout as an n-by-2 matrix, where n is the number of turbines, and the columns represent the x and y coordinates of each turbine, respectively. This method returns the wake free ratio in Java and C++, sets all variables, and updates the evaluation count.

Users have read access to the following variables without increasing the evaluation count. All variables are updated by the evaluator class:

• Wake free ratio: the fitness of the entire layout. It is a value between 0 and 1 representing the energy capture of the field over the theoretical energy capture of the same number of turbines without any inter-turbine wake. In C++ and Java, this value is returned by the evaluate method. If any of the turbines are within 8 radii of each other, given as R in the wind scenarios and here a set value of 38.5m, the layout will be considered invalid and the wake free ratio will be negative.

• Energy output: the energy capture of the entire field in kWh.

• Turbine Fitnesses: the wake free ratio for each turbine. This is an n-by-1 matrix with the same turbine order as the input to the evaluate method.

• Energy outputs: the energy at each turbine location in 24 directional bins. As can be seen in the wind scenario files, the wind distribution is split up over 24 directional bins. This n-by-24 matrix gives the energy capture of each turbine location, in the same order as the input to the evaluate method, for each of the 24 directional bins.

• Number of evaluations: the number of times the evaluate method has been called so far.

We expect an algorithm class or function, depending on language, that takes in an evaluator class as its only argument. We have provided an example binary GA working on the second track problem, the optimization of the number of turbines given a minimum wake free ratio per turbine, to show the expected format of submission. Note that we do not expect to receive the contents of the corresponding "main" file - we will use a similar script to run submissions and provide this only as an example.

All examples expect the scenario files in the current directory structure, and while you can change this, to make the examples work, you must

$ git clone https://github.com/d9w/WindFLO
$ cd WindFLO

The C++ interface is displayed in main.cpp and GA.cpp. To compile and run using gcc, a makefile has been provided. We are still working on the expectations for C++ submissions, but expect to build your algorithm with a makefile similar to this one. The example GA can be run using (from the WindFLO directory):

$ cd c++
$ make
$ ./eval.o

The Java interface is displayed in main.java and GA.java. To compile and run using Java v6, use (from WindFLO directory):

$ cd Java
$ javac main.java
$ java main

The MATLAB interface is displayed in main.m and GA.m. main.m can be executed in the MATLAB environment if the current directory is WindFLO/MATLAB:

>> cd MATLAB
>> main

We are continuously working on the competition software as we work on the automated submission and leaderboard system. If you would like to contact us concerning the APIs, please email us. Our emails can be found on the competition site.