The invention discloses an integrated-variable-selection PLSR (partial least squares regression) soft measurement method. The method is characterized in that three different variable-selection PLSR soft measurement models are established simultaneously, and integration is carried out in a weighting manner to obtain an output estimation value; online implementation of soft measurement is not limited to a single variable-weighted PLSR model, but adopts a mode of integrating a plurality of measurement models, thereby ingeniously avoiding the problem of selecting a best variable-selection PLSR method for establishment of the soft measurement model for the current data; and besides, by carrying out appropriate weighting on the output estimation value of each model through regression coefficient vector calculated through an PLSR algorithm, applicability of some variable selection method does not need to be checked repeatedly, and precision of the soft measurement model is further improved. The method, so to speak, efficiently improves applicability of the variable selection PLSR method for soft measurement modeling by utilizing an integrated modeling idea on the basis of existing work.