A method for automatic generation of piecewise linear models

This paper presents a method to automate generation of piecewise linear (PL) models which approximate some nonlinear multi-variable scalar functions with a given relative error. The method allows to tune the models for any given relative error and assures that any intermediate function simplification will not affect the overall accuracy. Estimation of the accuracy of the PL model and a possible trade-off between the size of the model and its accuracy is done as well. The given relative error is also used to estimate contribution of each factor in the overall function and to simplify the function neglecting the factors whose contribution is smaller than the given error. This allows for the final model to be the smallest possible, for the given accuracy