Layout Optimisation
When optimising a wind farm layout, turbines locations are changed in each iteration and the total energy of all turbines in non-neighbour farms is calculated. This energy yield is the [full yield](#Annual energy production predictions gross full and net yield) with wind farm blockage effects considered if this model is selected. Wake affects from all wind farms, including neighbours, are considered.
Random Optimisation Algorithms
Onshore irregular
The onshore irregular algorithm is based on the optimiser in WindFarmer 5.3 and performs much better than current random optimisers when there is a significant variation in the wind resource across the site.
To converge faster in these conditions the onshore irregular layout algorithm actively searches first for locations where the gross yield is higher, while retaining a random element that helps to prevent getting stuck in a local maximum.
The random element in layout acceptance is achieved using an acceptance factor: to accept a new turbine location for a proposed layout, the gross yield for the wind farm must be greater than the acceptance factor * the previous max gross yield. The default value of 0.999 is set such that the layout algorithm mostly searches for locations where the gross yield is higher, whilst retaining a random element that helps to prevent getting stuck in a local maximum. If the optimisation is failing to leave local minima consider reducing the acceptance factor slightly. Reducing the acceptance factor too much will effectively remove any check on Gross yield and make the onshore irregular algorithm behave more like the random walk algorithm (with a very large half-life).
Random walk
The random walk option is useful when you start with a layout that is already well designed.
On each layout iteration we move all turbines in a random directions with a randomly chosen jump sizes up to a maximum jump.
The new locations are based on the initial layout so you will get small variations from the start layout.
The maximum jump size decays over iterations as the model (hopefully) converges on a better layout. This is jump size evolution is defined by the initial max distance and half live parameters.
To get the best performance from the Random Walk, consider what is the largest location jump size you wish to make. At the start of the optimisation, turbines need to be able to make jumps to search the entire buildable area space to avoid getting stuck in local maxima locations. Also consider if you want turbines to jump any buildable area gaps (or not). Roughly, the initial max distance should be greater than the average distance between turbines in the optimised layout.
To set the jump size decay rate, set the half-life (in iterations). If you wish to run a longer optimisation increase this value. This setting should be considered along with the initial max jump size to ensure that late in the optimisation you are still making meaningful sized jumps.
Random independent
Turbine placement is completely random when using this algorithm. Each subsequent layout is unrelated to the previous layout. This can be useful for generating several starting layouts but it is unlikely to be an effective computational optimisation method.
Symmetrical Layout Optimisation Algorithm
For wind farm sites with low spatial variation in wind resource, symmetrical wind farm layouts may offer solutions to minimise infrastructure costs at the price of acceptable performance loss.
The symmetrical layout optimiser can also be used to rapidly design an initial layout. It generates a layout with up to a user defined number of turbines and respects geographic and separation constraints.
The symmetry used in the layout optimisation is based on geometric units that are aligned in two principal wind directions, which need not be orthogonal. The two principal axes are determined from the energy density calculated for each wind direction sector. This is based on a selected wind speed and direction frequency distribution at a measurement site, combined with the turbine power curve. The two principal axes then point towards the two direction sectors with the highest energy density.
The relative turbine spacing along the principal axes of a unit is determined by the weight of their relative corresponding energy densities. The principal axes and the corresponding weights define the basic shape of the symmetry unit, as shown below.
Basic shape of the symmetry unit
A symmetry unit can consist of either 4 turbines with 1 turbine at each corner of the unit ”2-2“ ,or of 6 turbines with two additional turbines at the centre of the longer sides of the unit “3-2”. In the end, the whole layout is composed of such repetitive units. The two unit types are shown below.
"2-2" and "3-2" unit types
A 2-2 unit has fewer turbines per unit and therefore the units need to be smaller (with shorter diagonal spacing in direction of the priority planes) to accommodate a certain number of turbines. In comparison a 3-2 unit contains more turbines per unit and they can afford to be larger (with longer diagonal spacing). On the other hand, a 3-2 unit has less angular freedom than a 2-2 unit because there are turbines at the midpoints of the longer sides of the unit, which means a certain deviation of the wind direction can cause these middle turbines to be in the wakes of the upwind turbine in the unit.
During the optimisation process, the units are expanded or compressed uniformly in order to place the number of turbines desired inside a specific area. This is accomplished by scaling the lengths of the axes whilst the aspect ratio of the unit remains the same. Additionally, variations of the layout are considered by rotation and counter-rotation of both principal units.
The symmetric layout optimisation uses a deterministic algorithm where a discrete number of cases are investigated and the layout with the maximum energy yield selected. With this approach, the global maximum may well be missed; however, the automatic analysis that is carried out allows a symmetrical layout with competitively high energy yield to be generated very quickly.
Other constraints such as exclusion zones are automatically be considered. Some results from the symmetric optimiser are presented in [28].