Using genetic algorithms to determine nonnegative monotone set functions for information fusion in environments with random perturbation

Abstract

Due to some inherent interactions among diverse information sources, the classical weighted average method is not adequate for information fusion in many real problems. To describe the interactions, an intuitive and effective way is to use an appropriate nonadditive set function. Instead of the weighted average method, which is essentially the Lebesgue integral, we should thus use the Choquet integral or some other nonlinear integrals. To apply this alternative, more realistic approach to information fusion, we need to determine the nonadditive set function from given input-output data, viewing the nonlinear integral as a multi-input one-output system. In this paper, we employ an adaptive genetic algorithm to construct an approximate optimal nonnegative monotone set function from given input-output data in an environment with random perturbation. An example for diverse strengths of random perturbation is shown to demonstrate the efficiency of this algorithm.