Time-varying and working-condition transition are the main problems in the industrial process.However,the static soft sensor model based on the fixed sample cannot track the current object,which leads to a poor prediction performance.In this paper,a dynamic soft sensor modeling approach based on the optimal bounding ellipsoid (OBE) and the partial least squares (PLS) algorithm was proposed.Firstly,the PLS soft sensor model based on offline data set is built.When a new query sample arrives,the statistics is established by principal component analysis (PCA) to find similar historical samples and use these similar samples to update the PLS model by OBE algorithm,so that the model achieves a good tracking effect.This method can effectively solve the problem of time-varying and working-condition transition in the process.The application results of numerical examples and the actual industrial data were given to verify the effectiveness.