Gradient Boosting Decision Tree for Rotor Temperature Estimation in Permanent Magnet Synchronous Motors
Hao Jing, Zifeng Chen, Xinghao Wang, Xueqing Wang, Lefei Ge, Gaoliang Fang, Dianxun Xiao
Abstract
The increasing power density of permanent magnet synchronous motors has led to a severe motor heating problem that demands precise rotor temperature information to avoid demagnetization. Traditional temperature estimation techniques rely on thermal models that require specialized knowledge in motor design, thermodynamics, and material science. However, thermal parameters are often hard to obtain in real applications and contain significant mismatches when the working environment of motors varies. To overcome this challenge, this letter proposes an ensemble data-driven approach using the gradient boosting decision tree (GBDT) to estimate the temperature of the permanent magnet. The proposed scheme uses the temperature data at the stator tooth, winding, and yoke to predict the rotor temperature. The GBDT technique offers advantages in terms of accuracy and versatility due to its strong capability in handling complex data in various motor operating conditions, making it well-suited for industrial applications. The experimental results of a high-power machine validate the greatly improved accuracy in rotor temperature estimation over other approaches.