In the field of sensorless control for Permanent Magnet Synchronous Motors (PMSM), significant progress has been achieved by researchers, yet challenges and difficulties persist under low-speed operating conditions.This paper introduces an innovative sensorless control strategy for PMSM, focusing on improving the precision and stability of speed control at low speeds. The approach integrates an Active Disturbance Rejection Control (ADRC) speed loop optimized through BP neural network-based parameter adjustments and an adaptive phase-locked loop (PLL). The disturbance observed by the active disturbance rejection velocity loop extended state observer (ESO) is used as the input of the enhanced neural network, and its dynamic output is used as the control parameter of the system, which can effectively solve the complexity and unknown interference in different operating environments. The bandwidth of PLL is adjusted according to the disturbance in real time to enhance the estimation accuracy of the speed under non-inductive control. This strategy dynamically adjusts the parameters of the ADRS controller and the accuracy of the speed estimation in real time, enhances the robustness of the system, and effectively inhibits the speed fluctuation at low speed.