High-Precision Rotor Position Estimation via Cascaded Model-Based and Data-Driven Estimator: An Integrated Method

Typical sensors for rotor position measurement are susceptible to environmental factors and may even cause reliability concerns during online operation. Thus, sensorless control becomes increasingly crucial in motor drives. To fulfill the precise position estimation of permanent magnet synchronous motors (PMSMs), a novel hybrid-driven position estimation method is proposed in this article. Instead of employing huge neural networks (NNs) for direct position estimation with a laborious offline training process, the preliminary position information is extracted from the designed model-based estimator in the proposed method and fed to the data-driven radial basis function (RBF) NN compensator to predict the position residuals for high-precision position estimation. For the convenience of practical application, an online training scheme based on the gradient descent method is designed for the learning of RBF compensator. Additionally, a domain-related compensation strategy is introduced to enhance the dynamic performance of the proposed method. By combining the respective strengths of model-based and data-driven methods, the proposed method significantly alleviates the parameter-tuning burden in the conventional model-based methods with enhanced estimation precision and addresses the high computational complexity in the data-driven methods with better generalization performance. Finally, real-time experiments are presented to verify the effectiveness and viability of the proposed method.