A method based on a genetic algorithm to find PWL approximations of multivariate nonlinear functions

In this paper we present a systematic approach to find piecewise-linear approximations of multivariate continuous nonlinear functions, by ensuring a good trade-off between approximation accuracy and model complexity. The proposed (suboptimal) method is based on genetic programming and takes into account the circuit constraints concerning the lower bounds for the size of each domain region (called simplex) where a given nonlinear function is approximated linearly. As a benchmark example, we approximate the well-known Hodgkin- Huxley neuron model.