Data-Driven Elastic Fuzzy Logic System Modeling: Constructing a Concise System With Human-Like Inference Mechanism

Abstract

The construction of fuzzy logic systems (FLSs) using data-driven techniques has become the most popular modeling approach. However, this approach still faces critical challenges, including the difficulty in obtaining concise models for high-dimensional data and generating accurate fuzzy rules to simulate human inference mechanism. To tackle these issues, a new FLS modeling framework called data-driven elastic FLS (DD-EFLS) is proposed in this paper. The DD-EFLS has two key characteristics. First, the fuzzy rules in the rule base can use different feature subspaces that are extracted from the original high-dimensional space to yield simple and accurate rules in feature spaces of lower dimensionality. Second, fuzzy inferences from various views are implemented by embedding different rules in the corresponding subspaces to imitate human inference mechanism. Based on the DD-EFLS framework, an elastic Takagi–Sugeno–Kang (TSK) FLS modeling method (ETSK-FLS) is proposed to train the elastic TSK FLS using the concise rules and a more human-like inference mechanism for modeling tasks based on high-dimensional datasets. The characteristics and advantages of the proposed framework and the ETSK-FLS method are validated experimentally using both synthetic and real-world datasets.