Clustering-Enhanced MCNN-LSTM-Based Control Method for PMSM Under Multi-Operating Conditions

<p dir="ltr">To address the issues of deteriorating dynamic response, disturbance rejection performance, and steady-state accuracy of permanent magnet synchronous motors (PMSMs) under multi-operating conditions, as well as the difficulty of accurately characterizing operating-condition differences using a unified data-driven compensation model, this paper proposes a clustering-enhanced multi-scale convolutional neural network-long short-term memory (CE-MCNN-LSTM) control method.

The method constructs operational state features and incorporates K-means clustering to embed operating-condition distance information into the control compensation model, thereby enhancing the model's ability to perceive and characterize different operating conditions. Based on this, a multi-scale convolutional neural network (MCNN) is used to extract local dynamic features, which are combined with a long short-term memory network (LSTM) to establish a temporal mapping relationship for control compensation.