<p>In this paper, a self-structured Chebyshev recurrent fuzzy neural network sliding mode control method (SSCRFNNSMC) based on a linear extended state observer (LESO) is proposed and applied to a single-phase active power filter (APF). The LESO is used to approximate the actual system model, estimate the uncertainties in the system, and improve the robustness of the system. At the same time, Chebyshev recurrent fuzzy neural network is used to approximate the optimal sliding mode switching gain. To solve the problem that the number of neural network nodes is too dependent on the design experience, a structure self-learning algorithm based on Chebyshev recurrent fuzzy neural network is designed, which can realize the dynamic adjustment of the number of hidden layer nodes and the number of fuzzy rules, and optimize the network structure in real time. The results of simulation and hardware experiments show that the proposed control method can compensate the harmonic current well and has good control performance, and the number of neurons in the neural network can be adjusted online, which has certain practicability and reliability.</p>