Wind Farm Layout Optimization for Non-Convex, Disconnected Zones

The patent segments the optimization process into two distinct phases: a global optimization phase using a genetic algorithm to obtain an initial arrangement, and a local optimization phase using a local search algorithm to refine specific turbine positions. This segmentation allows the computationally intensive genetic algorithm to run once for overall positioning, while the faster local search algorithm handles detailed refinements, thereby reducing total computing time while maintaining positioning accuracy.

Instead of applying the computationally expensive genetic algorithm to every iteration of optimization, the patent applies it only once to obtain an initial arrangement. Subsequent optimizations use the more efficient local search algorithm that only evaluates nearby positions for each turbine. This partial application of the powerful but slow genetic algorithm, combined with the efficient local search, achieves near-optimal positioning with significantly reduced computing time.

The patent segments the optimization process into a global phase using genetic algorithms and a local phase using local search. The genetic algorithm population (requiring significant memory) is maintained only during the initial global optimization, while the local search phase uses minimal memory to evaluate nearby positions. This segmentation reduces peak memory requirements compared to running the genetic algorithm throughout the entire optimization process.

The patent applies the memory-intensive genetic algorithm only partially - once for initial arrangement - rather than continuously. The subsequent local optimization uses minimal memory to evaluate discrete nearby positions for each turbine. This partial application of the resource-intensive algorithm significantly reduces overall computer memory requirements while preserving positioning accuracy.

The patent applies local search optimization individually to each wind turbine based on its specific local environment and constraints. For each turbine, the algorithm evaluates nearby discrete positions considering local wake effects, terrain features, and boundary conditions. This localized approach allows the method to adapt to complex non-convex and non-connex zones by making position decisions based on local characteristics rather than requiring global geometric simplicity.

The patent uses an iterative optimization process where turbine positions are dynamically adjusted based on wake effects and energy production calculations. The sequential nature of the local search allows the system to adapt dynamically to complex spatial constraints by evaluating and accepting position changes that improve overall farm performance while respecting local geometric constraints of non-convex and non-connex zones.