A parametric model for cavity filter using multiple outputs least‐squares fuzzy support vector regression

Abstract

The tuning of a microwave cavity filter is exceedingly difficult, and there is a lot of nonlinearity and ambiguity in the relationship between tuning elements and characteristic variables. Traditional manual tuning makes it difficult to determine the tuning direction and amplitude of tuning elements, which has a significant impact on microwave cavity filter mass manufacturing. In light of this, this paper proposes a parametric model that combines principal component analysis and multi-output least squares fuzzy support vector regression. The principal component analysis approach is first used to minimize the dimension of tuning data and retrieve important data that can reflect the tuning law. Second, the fuzzy membership function based on the distance of the class center is introduced into the multi-output fuzzy support vector regression, which solves the problem of excessive sensitivity to the outliers of the sample in the modeling process. Finally, a chaotic ant colony algorithm is used to optimize the parameters of least-squares fuzzy support vector regression. The experimental results of the eighth order cavity filter demonstrate that the proposed method has high prediction accuracy, strong generalization ability, and good robustness. It demonstrates a theoretical foundation for real-time cavity filter tuning and accurate prediction.