Attention-Based Dual Branch Fuzzy Neural Network Control With Fixed-Time Convergence Characteristics for Active Power Filters

To address harmonic contamination, this study presents a fixed-time sliding mode controller (FTSMC) with state-dependent variable exponent coefficients for an active power filter (APF). The proposed controller utilizes an attention-based dual branch fuzzy neural network (ADBFNN) to handle unknown nonlinearities. FTSMC is adopted for its fixed-time convergence guarantee with consistent settling time independent of initial conditions. This modification enables rapid, singularity-free convergence across the entire state space (both near and far from equilibrium). Moreover, the ADBFNN, which employs parallel processing through a fuzzification branch (handling nonlinear features of the current input) and a raw-input branch (directly preserving raw information from both initial and current inputs), leverages a combination of overlapped Gaussian functions and an attention mechanism to accurately estimate the complex model of the APF system. The feasibility and effectiveness of the ADBFNN-FTSMC approach are validated through experimental evaluations on an APF prototype, demonstrating significant performance improvements over existing techniques.