Abstract
Fuzzy inference systems provide a simple yet effective solution to complex non-linear problems, which have been applied to numerous real-world applications with great success. However, conventional fuzzy inference systems may suffer from either too sparse, too complex or imbalanced rule bases, given that the data may be unevenly distributed in the problem space regardless of its volume. Fuzzy interpolation addresses this. It enables fuzzy inferences with sparse rule bases when the sparse rule base does not cover a given input, and it simplifies very dense rule bases by approximating certain rules with their neighbouring ones. This chapter systematically reviews different types of fuzzy interpolation approaches and their variations, in terms of both the interpolation mechanism (inference engine) and sparse rule base generation. Representative applications of fuzzy interpolation in the field of control are also revisited in this chapter, which not only validate fuzzy interpolation approaches but also demonstrate its efficacy and potential for wider applications.
Introduction
Fuzzy logic serves as a robust framework for managing uncertainty and linguistic variables in computational systems. However, traditional models such as Mamdani and Takagi-Sugeno-Kang (TSK) face a significant limitation: the requirement for a dense rule base. In these systems, the antecedents (the "If" parts) of the rules must comprehensively cover the input space. When a system encounters an input that does not overlap with any defined rule—a common occurrence in sparse rule bases—the inference engine fails to produce an output.
Fuzzy Interpolation Systems (FISs) mitigate this by "filling in the blanks" between existing rules. This capability not only ensures system robustness in the face of incomplete knowledge but also allows for the intentional simplification of complex rule bases, enhancing computational efficiency and transparency.
Interpolation Methodologies
The literature categorizes fuzzy interpolation into several distinct mechanisms based on the underlying fuzzy model and the mathematical approach to bridging gaps.
Mamdani-style Interpolation
Mamdani-style FISs typically employ one of two strategies:
- Resolution Principle-based: This approach decomposes fuzzy sets into horizontal slices known as $\alpha$-cuts. Interpolation is performed on these individual slices, which are then reassembled to form a final fuzzy conclusion.
- Analogy-based (Transformation): Rather than slicing, this method utilizes a "scale and move" transformation. It determines the relative position and shape of an observation compared to a rule's antecedent and applies a corresponding transformation to the rule's consequent.
TSK-style Interpolation
Traditional TSK models utilize polynomial functions for their consequences. Extended TSK models for interpolation incorporate modified similarity measures. This ensures that every rule in the base contributes to the final output to some degree, even without direct overlap, by calculating a weighted average of the fired rules.
Rule Base Generation and Complexity Reduction
A key advantage of FIS is the ability to reduce rule base complexity. By treating the problem space as a geometric surface, researchers can identify areas of high "curvature" where rules are essential. In "flatter" regions, rules can be removed, and the system can rely on interpolation. This approach allows for the creation of highly efficient, sparse rule bases that maintain the accuracy of their denser counterparts.
Case Studies and Results
The efficacy of FIS has been demonstrated across various benchmarks and real-world scenarios:
- Truck Backer-Upper Control: In this classic non-linear control problem, FIS-based controllers achieved smoother performance than conventional systems. By leveraging symmetrical redundancies, researchers were able to reduce the rule base from 9 rules to just 6 without compromising control quality.
- Smart Home Heating Systems: Managing home heating based on GPS data and historical patterns typically requires a massive rule base. A standard fuzzy partition for this problem would involve tens of thousands of rules. Through fuzzy interpolation, this was reduced to a manageable 72 rules. The resulting system successfully optimized boiler burning time and improved preheating accuracy based on predicted arrival times.
Conclusion and Future Directions
Fuzzy interpolation provides a critical "safety net" for fuzzy inference systems, enabling them to operate effectively with sparse or incomplete data. While current research has validated these approaches in control and optimization, several avenues for future work remain. These include the integration of Type-2 fuzzy sets to better handle linguistic uncertainty, the development of rigorous mathematical proofs for convergence properties, and the creation of unified platforms that seamlessly combine traditional and interpolative fuzzy models.
Summary updated