A nonlinear transversal fuzzy filter with online clustering

In this paper, a nonlinear transversal fuzzy filter with online clustering is proposed. It is based on radial-basis-function networks (RBFN) and functionally is equivalent to the TSK fuzzy system. The proposed filter has the following features: (1) hierarchical structure for self-construction. The fuzzy rules, i.e., the RBF neurons are generated automatically during the training process. (2) Online clustering. Instead of selecting the centers and widths of membership functions arbitrarily, an online clustering method is applied to ensure reasonable representation of input terms associated with an input variable. It not only ensures proper feature representation, but also it optimizes the structure of the filter by reducing the number of fuzzy rules. (3) All free parameters in the premise and consequent parts are determined online by a hybrid sequential algorithm without repeated computation to facilitate real-time applications. Using the proposed hybrid learning algorithm, low computation load and less memory requirements are achieved. Simulation results show that the proposed filter can obtain better accuracy with lower system resource requirements compared with other existing approaches.