An Evolutionary Objective Cluster Analysis-Based Interpretable Fuzzy Identification Method - Intelligent Computing for Sustainable Energy and Environment - page 169

An Evolutionary Objective Cluster
Analysis-Based Interpretable Fuzzy
Identification Method
Na Wang 1 , Chaofang Hu 2 , and Wuxi Shi 1
School of Electrical Engineering and Automation, Tianjin Polytechnic University,
399 Binshuixi Road, Tainjin 300387, China
[e-mail]
School of Electrical and Automation Engineering, Tianjin University,
92 Weijin Road, Tianjin 300372, China
[e-mail]
Abstract. In this paper, an Evolutionary Objective Cluster Analysis-
based Interpretable Fuzzy Identification Method (EOCA-IFIM) is pro-
posed for constructing Mamdani fuzzy model. Firstly, the Enhanced Ob-
jective Cluster Analysis (EOCA) algorithm is presented to obtain the
robust and the moderate compact initial fuzzy partition. Following, the
(1+1) Evolutionary Strategy (ES) is introduced to improve the semantics
of the initial parameters. Based on that, a complexity-accuracy trade-off
is well realized. The simulation results of the Box-Jenkins and the elec-
trical application example show the superiority of the presented method.
Keywords: Fuzzy modeling, Mamdani model, Interpretable, Fuzzy
identification, Objective Cluster Analysis.
1 Introduction
Maintain the interpretability of the models is one of the main objectives in
fuzzy modeling for complex systems [1]. Compared with the T-S model [1-2],
the Mamdani model is more approved than the T-S model for its full fuzzy sets
in structure. Thus it gets wide focus in interpretable fuzzy modeling [3-4].
The identification of the Mamdani model relies on two factors: the structure
identification and the estimation of parameters [5]. However, in the data-driven
methology, due to the effect of noise, the redundant, inconsistent rules, or the
undistinguishable fuzzy sets are usually generated. Thus the interpretability of
the model is decreased.
As for that, the Ad-Hoc methods, such as WM, WCA and IRL are presented
[6]. Even though, the identification of single rule still relies on the samples and
the initial fuzzy partition. As a result, it is dicult to fit the local dynamics.
Additionally, it possibly leads to redundancy and inconsistency. In this paper,
an Evolutionary Objective Cluster Analysis-based Interpretable Fuzzy Identi-
fication Method (EOCA-IFIM) is proposed to construct the Mamdani model.