Probabilistic Weighted Extreme Learning Machine for Robust Modeling

Abstract

Radar performance prediction is becoming more and more important in Product-Life-Cycle-Management of radar. However, modeling the performance is difficult because of the strong nonlinearity from radar’s complex system as well as immeasurable effects like electromagnetic interference, instrument accuracy and so on. This paper proposes a probability weighted extreme learning machine to model the performance of radar under noise. First, a distributed extreme learning machine modeling is developed, upon which the probability distribution function (PDF) of multiple local models is estimated by the Parzen window method. This distribution function is further used as weights to integrate all local models to construct a global robust ELM model. The successful application of this robust probabilistic weighted ELM method to both artificial case and real life case demonstrates its great advantages in the modeling of an unknown system with various kinds of random noise.