Fuzzy pattern trees as an alternative to rule-based fuzzy systems: Knowledge-driven, data-driven and hybrid modeling of color yield in polyester dyeing

摘要: This paper advocates a novel approach to fuzzy systems modeling called pattern trees. is largely motivated by alleged disadvantages of rule-based system architectures that still dominate the field. Due its hierarchical, modular structure and use different types (nonlinear) aggregation operators, tree has ability represent functional dependencies in more flexible compact way, thereby offering reasonable balance between accuracy model transparency. We evaluate this new class context concrete case study, namely color yield polyester high temperature dyeing as function disperse dyes concentration, time. To end, we compare three possibilities for construction: purely knowledge-driven, data-driven hybrid combining these two. Our results show that, comparison conventional using Mamdani rules, trees are not only accurate but also therefore easily interpretable, regardless whether models constructed or manner. Moreover, can outperform knowledge-driven if expert knowledge calibration combined suitable way.