Optimal power dispatch within wind farm based on two approaches to wind turbine classification

Abstract

A priority classification model of wind turbine units has been established using both a self-organizing feature map (SOFM) neural network algorithm and a fuzzy C-means clustering algorithm based on a simulated annealing genetic algorithm. Ten minute average wind turbine power output, wind speed and their root-mean-square deviations (RMSD) are taken as the measured parameters. In this model, which also takes into account line losses of collection system in wind farm, wind turbine units with the highest performance are allocated to a priority group, while others within the wind farm were divided across two further classes (making 3 classes in total), thus achieving power distribution meeting the dispatching need of the grid while decreasing the power loss of the wind farm. The two approaches to clustering are compared. The results of the simulation show that the fatigue damage resulting from application of the fuzzy C-means clustering algorithm based on the simulated annealing genetic algorithm (SAGA-FCM) is reduced relative to the results from the SOFM, the number of turbine units stop is more relative to the number from the SOFM, which proves that this approach to the classification of wind turbine units before optimization and dispatching is superior and is beneficial to the operation of wind turbine units and to the improvement of the power quality.