Model predictive control arithmetic was used for wind turbine pitch control, whose nonlinear model was identified by support vector regression (SVR). But the model of wind turbine could be changed in fieldwork, so incremental learning algorithm was adopted for SVR online identification. In order to shorten the calculation time of SVR online identification, the improved sequential minimal optimization (SMO) algorithm was used to substitute for the original quadratic programming (QP). And the algorithm was further improved by the elimination of invalid break points and the model's being stored and reused. Because the differential loop was used in the electro-hydraulic proportional pitch-controlled system and the direction of load was changeless, the models of feathering and backpaddling are different. Therefore the two models were switched in the predictive control process. At last two models switched predictive pitch control algorithm based on improved incremental SVR was presented and tested in the pitch-controlled wind turbine semi-physical simulation test-bed. The results showed the power was kept more steady around the rated by the algorithm than traditional PID control one, when wind speed was above the rated.
[1]
John C. Platt,et al.
Fast training of support vector machines using sequential minimal optimization, advances in kernel methods
,
1999
.
[2]
R. Chedid,et al.
Intelligent control of a class of wind energy conversion systems
,
1999
.
[3]
Qi Miao,et al.
Nonlinear model predictive control based on support vector regression
,
2002,
Proceedings. International Conference on Machine Learning and Cybernetics.
[4]
S. Sathiya Keerthi,et al.
Improvements to the SMO algorithm for SVM regression
,
2000,
IEEE Trans. Neural Networks Learn. Syst..
[5]
A. Gretton,et al.
Support vector regression for black-box system identification
,
2001,
Proceedings of the 11th IEEE Signal Processing Workshop on Statistical Signal Processing (Cat. No.01TH8563).