METHOD FOR COMPUTER-AIDED GENERATION OF DATA-CONTROLLED MODEL OF ENGINEERING SYSTEM, PARTICULARLY GAS TURBINE OR WIND TURBINE Russian patent published in 2016 - IPC G05B17/02 G05B13/04 G06N7/00 

FIELD: physics; control.

SUBSTANCE: invention relates to a method for computer-aided generation of a data-controlled model of an engineering system, particularly a gas turbine or wind turbine. A data-controlled model is trained preferably in low-density training data regions. A density estimator issues, for data sets from the training data respectively, a confidence level which the greater, the higher the similarity of the corresponding data set with other data sets from the training data, wherein the data-controlled model enables to reproduce training data sets with a model error, respectively. The density estimator and the data-controlled model, trained at the corresponding iteration step, enable to select or weigh data sets from the training data for training at the next iteration step, wherein data sets from the training data with low confidence levels and high model errors are selected faster or weighed higher. The generated data model is trained faster and with fewer computational resources.

EFFECT: by establishing optimisation criteria, for example, low toxic emissions or low combustion dynamics in the gas turbine, the service life of the engineering system can be prolonged.

24 cl, 2 dwg