Functional-link nets with genetic-algorithm-based learning for robust nonlinear interval regression analysis

摘要

Interval regression analysis has been a useful tool for dealing with uncertain and imprecise data. Since the available data often contain outliers, robust methods for interval regression analysis are necessary. This paper proposes a genetic-algorithm-based method for determining two functional-link nets for the robust nonlinear interval regression model: one for identifying the upper bound of data interval, and the other for identifying the lower bound of data interval. To facilitate the inclusion of regular data in the robust nonlinear interval regression model, in the fitness function, not only the cost function with different weighting schemes but also the number of training data included in the interval model is taken into account. As for resisting outliers, the effects of training data beyond or beneath the estimated data interval on the determination of upper and lower bounds can be greatly reduced during the training phase when these data are located in the rejection region. Simulation results demonstrate that the proposed method performs well for contaminated data sets by resisting outliers and including all regular data in the data intervals.