Due to the limitations of process modeling engineering applications of off-line least square support vector machine(LSSVM) and no-sparse strategy online LSSVM

an online recursive LSSVM soft sensor model based on selective recursive and self-adaptive updating model parameters is presented. This method uses the fast leave-oneout cross-validation(FLOO-CV) error as the modle update threshold. In forward learning

the FLOO-CV prediction error-based threshold is used to enhance the sparse ability of the model by only introducing the samples with larger predictive error and the FLOO-CV is also utilized in backward learning only to delete redundant samples with the smallest error

which enhances the model's generalization ability. In order to verify the model by using the actual operation data of the power plant and forecast the primary air flow of the coal mill online

a set of online soft sensor platform is developed. The platform is verified on the field of a 1 000 MW unit

and the results show that

this platform has high prediction accuracy for primary air flow

it can replace the flow meter with failure and ensure the stability and reliability of primary air flow signal of the coal mill.