According to the state monitoring problem of power plant equipment

an improved auto-association kernel regression method is applied to establish the state early warning model for power plant equipments. The Mahalanobis distance operator is replaced with robust distance operator to improve the robustness and anti-pollution ability of the autoregressive model. By applying clustering ideas to distance partitioning

a variable interval state matrix extraction method based on clustering is proposed

and the specific implementation steps are also given. During training period

the model parameters are optimized by using quadruple cross-validation learning mechanism

and an optimal state monitoring model is obtained. Moreover

by using the primary air fan in a 600 MW unit as the example

the state parameter estimation model is established. The actual operational data in DCS and the simulation result shows that

the improved robust state estimation method can greatly improve the robustness of the model while maintaining high accuracy

which has strong anti-pollution ability. Thus

in the early stages of the failure

it is able to give early warning signals in time. This method provides a theoretical basis for state maintenance for power plants.