SOFIS: A New Approach to Build Fuzzy Inference Systems

Fuzzy control is a system control design technique developed to take advantage of the heuristic knowledge that an expert operator may have. While it has been demonstrated to be useful in many practical situations, the fine-tuning of a fuzzy controller is a timeconsuming process. To accelerate this process, and even automate the generation of a fuzzy controller, Adaptive Network-based Fuzzy Inference Systems (ANFISs) were proposed. ANFIS frames a fuzzy controller as a neural network, enabling the use of common training algorithms to tune the controller. In this paper, a Swarm Optimized Fuzzy Inference System (SOFIS) is proposed as an alternative to ANFIS. SOFIS not only leverages the heuristic knowledge of an expert operator, but also incorporates formal knowledge of the system. Furthermore, it simplifies experimentation with various controller and system parameters, making it a valuable control research tool. Additionally, derivative-free optimization algorithms are less prone to becoming stuck in local minima. The proposed SOFIS is evaluated using a boost converter: a nonlinear, non-minimum phase system with highly uncertain parameters typically used as a benchmark for new control algorithms. The obtained results show outstanding performance and exhibit empirical robustness.