In order to solve the problem of frequent failures of coal mills in coal-fired power stations

a fault early warning method based on multi-state estimation(MSET) and vector similarity is proposed. Taking the mediumspeed coal mill of a 350 MW unit as the research object

the key parameters that can characterize its operating status are selected

the historical data containing fault information is collected

and the data set is divided. The process memory matrix is established on the training set data by equal-spacing sampling method

and the observation vector is input into the MSET model to obtain its corresponding estimation vector. At the same time

the similarity function of the two is defined

and the early warning threshold is determined by the sliding window method. At last

the actual operation data is used to carry out verification

and the results show that

when the coal mill runs normally

the average relative error of the input model data prediction is below 1%. When the coal jamming fault occurs in the coal mill

the model can detect the abnormal operation of the coal mill in time and issue an alarm 380 s before tripping. It is proved that the model can accurately predict the change trend of various parameters and realize the rapid and accurate intelligent early warning of coal mill failures.