On the complexity and interpretability of support vector machines for process modeling

The design of a support vector machine with Gaussian kernels is considered for modeling nonlinear processes. The structure is equivalent to a neuro-fuzzy system based on a radial basis function network considering some restrictions. To improve the interpretability and reduce the complexity of the structure a hybrid learning scheme is proposed. First, the input-output data is supervised and clustered according to a modified form of the Mountain Method for cluster estimation, the subtractive clustering. Then, support vector learning finds the number of centers, its positions and output layer weights of the structure. The proposed learning scheme is applied for modeling the Box-Jenkins furnace benchmark and the distributed collector field of a solar power plant