Abstract
The paper introduces an outlier detection method based on MLS (moving least squares) and healthy sample model for hydropower unit. It uses the unit's monitoring data normal state to build base value and limit value of healthy sample model, and make condition partitions by dividing the scale of active power and water head into several sections. Then, MLS fitting method is used to fit the three-dimension surface of healthy sample model in the grid of unit operating conditions. It is proved by the analysis and experiment that the application of MLS fitting method can implement the dynamic threshold comparison and outlier detection in full condition for hydropower, comparing with the traditional LS (least squares) method, it is more reasonable and accurate for evaluating the healthy state of unit.
