Experimental data and Simulink models for knowledge-guided DNN-based PMSM harmonic compensation

Description

This dataset provides the experimental data, simulation models, and source codes associated with the study on a knowledge-guided deep neural network (KG-DNN) with an enhanced adaptive notch filter (EANF) for harmonic compensation in permanent magnet synchronous motor (PMSM) drives. Experimental data were collected on a 1.5 kW surface-mounted PMSM testbench controlled by a dSPACE DS1202 real-time system. The control algorithm was executed at a 10 kHz interrupt rate, and signals were recorded at 8192 Hz. The dataset includes two CSV files. experimental_sample.csv contains time-series measurements for six control configurations (A–F), with columns for timestamp (s), reference speed (rpm), measured speed (rpm), d/q-axis currents (A), d/q-axis voltages (V), electromagnetic torque (N·m), average torque error (N·m), total harmonic distortion (%), and configuration label. sensitivity_scan.csv reports the parameter sensitivity scan results, including parameter name, value, average torque error, THD, convergence epochs, and stability status. Reproduction instructions are provided in the Supplementary Material of the associated paper.

Steps to reproduce

Detailed reproduction steps are provided in Supplementary Material C of the associated paper. In brief: (1) install MATLAB R2022b or later with Simulink, Deep Learning Toolbox, and DSP System Toolbox; (2) add Core_Algorithms/, Models/, and Training_Data/ to the MATLAB path; (3) run main_drive_script.m and select Train_Phase or Test_Phase; (4) open sim_model_PMSM.mdl to reproduce the closed-loop simulations; (5) run validate_model.m to reproduce Table S2. The dataset experimental_sample.csv and sensitivity_scan.csv can be loaded directly for independent verification.

Institutions

• Fujian Chuanzheng Communications College

  Fujian, Fuzhou

Categories

Licence