In order to optimise the interference immunity and starting performance of the classical sliding mode controller for permanent magnet synchronous motor (PMSM), a radial basis neural network-based optimised dual adaptive sliding mode control algorithm was designed. The weights of the radial basis function(RBF) neural network were optimised in real time by the gradient descent method to dynamically control the dual adaptive function of the adaptive convergence law in the sliding mode. The convergence speed of the novel convergence function was improved by combining with the higher-order sliding film surface. The adaptive learning rate optimisation algorithm was improved to dynamically adjust the learning rate to further improve the motor control accuracy, response speed and anti-interference ability, and a sliding mode disturbance observer was used to feed forward the external disturbance to the controller in time. The model was built on MATLAB/Simulink and compared with PI, SMC, NeSMC, RBNSMC, RBNSMC + SMOD control. The experimental results show that the control algorithm has better tracking performance, can effectively enhance the system's anti-disturbance ability, and the start-up performance has significant improvement.