The flow characteristic of the steam turbine valve is very important for the load regulation ability of the unit. A clustering algorithm based on information entropy is proposed to extract standard parameter vectors under different working conditions from massive historical data. Furthermore, the characteristic relationship between valve opening and flow rate can be obtained by regression. The multivariate state estimation method is introduced to estimate and judge the parameters in real time using the cluster center state vector matrix obtained by clustering. The class center state vector matrix is updated according to the distance between vectors, and then the adaptive update of the valve flow characteristic curve is realized. The actual data of a 300 MW unit is calculated, and results show that the proposed method is accurate and effective and it is suitable for engineering applications.