It is of great significance to carry out health condition assessment and fault early warning of auxiliary equipment for safe operation of thermal power units in new power system. By taking the forced draft fan of a supercritical 660 MW thermal power unit as the research object

a method to construct dynamic memory matrix based on multiple characteristic parameters is proposed. The application shows that the proposed method can improve calculating speed of model effectively while ensuring the accuracy of calculated results. This work also presents a calculation method of weighted coefficients to modify the multivariate state estimation technique(MSET). The global similarity and parameter similarity indexes are introduced for fault early warning and recognition. An early fault warning model based on dynamic matrix and weighted MSET is utilized to simulate faults of forced draft fan. The results indicate that the weighted MSET model can not only improve the prediction accuracy of abnormal parameters under fault conditions effectively

but also reduce the influence of abnormal parameters on the predicted results of normal parameters. Consequently

the model proposed can realize both early warning of forced draft fan faults and recognition of abnormal parameters.