Aiming at the problem that the membrane bio-reactor (MBR) wastewater treatment process is susceptible to the influence of external disturbances such as influent water quantity and water quality fluctuations, which makes it difficult to accurately measure the membrane permeability, a soft-sensor method for permeability of membrane based on an adaptive robust fuzzy neural network (ARFNN) is proposed. Firstly, a symmetric anti-noise loss function is constructed to reduce the sensitivity of the model to external interference and improve the robustness of the soft measurement model. Then, an adaptive gradient descent algorithm is designed to dynamically optimize the model parameters and improve the detection accuracy of the soft measurement model. Finally, the convergence of the ARFNN is verified using the Lyapunov function to analyze the model's robustness, which ensures the model's convergence speed and anti-interference ability. The designed ARFNN-based membrane permeability soft-sensor method is applied to the actual wastewater treatment process, and the experimental results show that the model can not only realize the online detection of membrane permeability, but also obtain high detection accuracy under the condition of external interference.