Aiming at solving the problems of the least squares support vector machine(LSSVM) occurred during on-site data modeling

such as being difficult to meet the distinct operating conditions poor robustness

a robust LSSVM model based on online adaptive revision(online-RLSSVM) was proposed. This method uses the total forecast error as the threshold value

adaptively updates the model parameters according to different working conditions

which improves the adaptability of the model to the data. At the same time

the fuzzy membership gives fuzzy dynamic weights to the square error term in the vector machine optimization

to enhance the anti-noise ability of the robust LSSVM model. Furthermore

this method was applied to predict the primary airflow in power plant and the results were compared with that of the ordinary LSSVM model. The results show that

the established model has better robustness and higher prediction accuracy. This algorithm can be used for real-time prediction and estimation of data under different operating conditions

and the research provides good data support for various online monitoring systems.