Hybrid Fuzzy-AI Control

Abstract

This chapter explores hybrid fuzzy–AI control architectures that combine interpretable rule-based control with learning-based adaptation. It begins by reviewing the motivation for hybrid approaches, particularly in situations where fuzzy controllers provide interpretable baseline behavior but may have limitations under disturbances or unmodeled dynamics. The chapter surveys several integration patterns, including fuzzy-LSTM, fuzzy-DNN, fuzzy-Transformer, and fuzzy-reinforcement learning architectures, before developing a practical hybrid control strategy in which a neural component supplements, rather than replaces, the fuzzy controller. The chapter then implements a practical hybrid Fuzzy Logic Controller with Online Neural Residual Compensation (FLC-ONRC). In this architecture, a lightweight neural compensator learns residual corrections to the control signal while the fuzzy rule base remains fixed. The chapter demonstrates how this approach improves disturbance rejection and tracking performance in the presented simulation without sacrificing the interpretability of the fuzzy controller. Through this design, the chapter illustrates how hybrid architectures can combine the strengths of rule-based reasoning and adaptive learning in control systems.