Abstract: To improve the nonlinear processing capacity of the partial least-squares algorithm based on a kernel function, reduce the sensitivity of the soft-sensor model to abnormal data, and improve the generalization ability of this model, we present a local weighted mixed-kernel partial least squares algorithm for soft-sensing online modeling. The original inputs are mapped into a high-dimensional feature space via a mixed-kernel function comprising several kernel functions with different properties. In the high-dimensional feature space, the mapped data are weighted according to the weight of each sample calculated by a locally weighted learning algorithm. The kernel partial least squares algorithm is then used to establish an online local soft-sensing model. A numerical simulation and an online soft-sensor modeling simulation using data related to industrial bisphenol-A production units is used to demonstrate the effectiveness of the proposed algorithm.