An Improved Multivariate State Estimation Technique for Condition Monitoring of Boiled Feedwater Pumps

Condition monitoring (CM) of boiled feedwater pumps (BFPs) is an important task for thermal power plants. Considering the variable operating conditions of BFPs, it is necessary to capture their dynamic features and update CM models in real time. In this paper, an improved multivariate state estimation technique (MSET) is proposed. Firstly, the states of BFPs are serialized to incorporate temporal features. Secondly, for the memory matrix (MM) of MSET, an expansion strategy is introduced, aiming to improve the generalization capability of the method. Finally, an MM thinning strategy is proposed to ensure the feasibility and improve the efficiency of the method. The experimental results show that state serialization and MM expansion greatly improve the accuracy and sensitivity of the method for CM of BFPs, and MM thinning avoids the method failure and improves the computational efficiency.