A support vector machine with wavelet kernel was developed for accurate modeling of nonlinear systems. A method of providing an optimized model of a nonlinear system includes using a support vector machine (SVM) having a wavelet kernel, where support vectors include a family of multidimensional wavelets. Training the SVM allows optimization of the number of support vectors, the weights of the support vectors, and the translation factors of the support vectors. Use of a novel linear programming approach reduces computational demands required for training, allowing optimized support vectors to give an optimized model of the nonlinear system. Further, on-line retraining is possible, so that the model can be adapted to changing conditions.