Modeling of nonlinear systems using relevance vector machines

A modeling method of nonlinear dynamic systems based on relevance vector machine (RVM) is presented. The RVM has a high modeling accuracy as well as a simpler model structure with a fewer control parameter in training phase. Due to the low computational complexity in testing phase, the RVM is more suitable than SVM for the real-time applications. In addition, the kernel function in RVM must not necessarily fulfill Mercer´s conditions. Several simulation examples have been used to evaluate the performance of RVM method. The results verify the effectiveness of the proposed method in the modeling of nonlinear dynamic systems.