Sample Selection Method Research On Non-parameter Model And Application On Condition Monitoring Of Wind Turbines

As a clean renewable energy, wind power generation is subject to extensive attention and development in worldwide these years. Condition monitoring of the wind power generation is the important component in wind power research field, where it is a important research in adopting operating data to condition monitoring. The data of wind power generation is randomness and volatility. So it is very difficult to analyze large-scale data. A direct and effective way is sample selection method to simplify training sample set under guaranteeing the classification performance. Sample selection is not only to reduce to count the cost and quicken learning rate, but also evitable to overfitting and enhance the generalization ability of sorting algorithms.In this paper, aiming at the sample selection of wind power generation data to research the sample selection method with training sample set, and research on the main issues of condition monitoring, the main content of work and the innovate achievement are as follows:1. Through analyzing current situation of sample selection method and condition monitoring of the wind power generation, there are4issues in modeling of wind power generation:mass data of sample, multiple correlations of sample, nonlinearity of sample and randomness of sample. And a conclusion is put forward to adopt sample selection method to reduce quantity of the sample set without influencing prediction accuracy and characteristic. To same prediction accuracy, modeling time is reduced; To same quantity of sample set, prediction accuracy is increased.2. Adopting fuzzy C mean cluster and nonlinear state estimate technology, modeling with the temperature of wind power generation gearbox, proceed sample selection with fuzzy C mean and make contrastive analysis between prediction and other situations.3. Sample selection method is rearched Based on uniform design principle. Modeling data is selected based on uniform design principle, and the sample which is selected is used to be modeling and analysed.