有混合数据输入的自适应模糊神经推理系统

An Adaptive Network-based Fuzzy Inference System with Mixed Data Inputs

现有数据建模方法大多依赖于定量的数值信息,而对于数值与分类混合输入的数据建模问题往往根据分类变量组合建立多个子模型,当有多个分类变量输入时易出现子模型数据分布不均匀、训练耗时长等问题.针对上述问题,提出一种具有混合数据输入的自适应模糊神经推理系统模型,在自适应模糊推理系统的基础上,引入激励强度转移矩阵和结论影响矩阵,采用基于高氏距离的减法聚类辨识模型结构,通过混合学习算法训练模型参数,使数值与分类混合数据对模糊规则的前后件参数同时产生作用,共同影响模型输出.仿真实验分析了分类数据对模型规则后件的作用以及结构辨识算法对模糊规则数的影响,与其他几种混合数据建模方法对比表明本文所提出的模型具有较高的预测精度和计算效率.

The available data modeling methods mostly depend on quantitative numerical information. But the data modeling with both numerical and categorical data input often has to build multiple sub-models on the basic of combination of categorical variables. It is likely to present unevenly data distribution of sub-models, time-consuming training process and other problems when the multiple categorical variables are input. For the above problems, an adaptive network-based fuzzy inference system with mixed data inputs is proposed. Based on the structure of the adaptive network-based fuzzy inference system, a firing-strength transform matrix and a consequent influence matrix are introduced. The subtractive clustering based on the Gaussian distance is adapted to identify structure of model, and a hybrid learning algorithm is used to train parameters of model. The numerical and categorical data play an important role on the antecedent and consequent parameters of fuzzy rules, and jointly affect the output of model. The simulation experiment analyzes the effect on categorical data to the consequent rules and structure identification to number of fuzzy rules. Comparing with others data modeling with mixed data inputs, the proposed model in this paper has higher prediction accuracy and computational efficiency.