A Model-Based and Data-Driven Integrated Temperature Estimation Method for PMSM

To fulfill accurate online temperature estimation of permanent-magnet synchronous motor (PMSM), an integrated model-based and data-driven method is proposed in this paper. First, a simplified lumped parameter thermal network (LPTN) model is developed to learn the tendency of temperature variations. Meanwhile, a small-scale artificial neural network (ANN) is specifically designed to compensate the unmodeled characteristics. The parameters of LPTN model in the proposed method is identified purely from the common variables and no material information is required. With the knowledge learned by the LPTN model and powerful fitting capability of ANN, accurate estimation for both stator and rotor temperatures can be achieved with low computational burden and reduced parameter dependency. Both offline and online experimental results are presented to prove the excellent performances of the proposed method.