基于期望值简约的模糊非线性回归

Expectation reduction-based fuzzy nonlinear regression

为解决不确定性数据的学习问题,借鉴2-型模糊理论中的简约思想,提出了一种利用期望值简约技术处理三角型模糊输入输出数据的模糊非线性回归模型.首先将三角型模糊输入输出利用其期望值简约为清晰输入输出,然后利用经典随机赋权网络对其进行学习,最后再通过新定义的三角型模糊输出变量的宽度矩阵将网络清晰值输出还原为三角型模糊输出.仿真实验结果表明,与已有模型相比,该模型具有更高的学习准确度和更好的扩展能力.

Inspired by the idea of reduction in 2-fuzzy theory,an expectation reduction-based fuzzy nonlinear regression model is proposed to deal with triangular fuzzy number inputs and outputs.In this model the triangular fuzzy inputs and outputs are firstly replaced by their expectations,then the data can be trained with classical random weight network.Finally the crisp real outputs are recover to triangular fuzzy outputs by using a width matrix of original target triangular fuzzy outputs.The experiment results show that the proposed model gives higher learning accuracy and better generalization.