Abstract: Model predictive control arithmetic is used for wind turbine pitch control, whose nonlinear model is identified by support vector regression (SVR). But wind turbine’s model may be changed under fieldwork, so incremental learning algorithm is adopted for SVR online identification. The improved sequential minimal optimization (SMO) algorithm is used to substitute the original quadratic programming (QP). And the algorithm is further improved by the method that the invalid break points are eliminated and the model is stored and reused. So the cal-culation time of SVR online identification is greatly shorted. Because the differential circuit is used in the electro-hydraulic proportional pitch-controlled system and the direction of load is changeless, the model is different between feathering and backpaddling. Therefore the two models are switched in the predictive control course. Then when wind speed is above the rated, the generator power is kept more steadily around the rated and the pitch load fluctuation is greatly reduced by the algorithm which is used in the pitch-controlled wind turbine semi-physical simulation test-bed than traditional PID control one.

Key words: Model predictive control, Pitch-controlled SVR, Semi-physical, SMO