Abstract: The kernel problem in soft sensor of industrial processes is how to build the soft sensor model. However, there exist some questions to some extent in soft sensor model with conventional modeling methods such as global single model and multiple models. Using the high generalization ability of support vector machines (SVMs) and the idea of locally weighted learning (LWL) algorithm, this paper proposes a novel learning algorithm named weighted support vector machines (W_ SVMs) which is suitable for local learning. We also present a drifting modeling method based on this algorithm. The proposed modeling method is applied to the estimation of Box-Jenkins gas furnace and FCCU and the simulation results show that the proposed approach is superior to the traditional modeling methods.