This repository contains the implementation and results of a LPTN-informed LSTM for multi-node temperature estimation in PMSMs.
The simulation for PE condition validation and the overall 23.8 hours dataset is available in this repository.
lptn_informed_LSTM.ipynb file will be uploaded once the manuscript is available in early access. In the interim, you can request a personal copy of the code file by emailing me directly or sending me a direct message.
\dataset: Temperature and operation states Dataset for the tested IPMSM.\matlab_PEcondition: A simulation file using MATLAB\simulink to demonstrate the potentional limitation in third or high order LPTN global parameter identification\python_LPTN_informed_LSTM: Script for training the LSTM model.lstm_training.py: Script for training the LPTN-informed LSTM model.
The project results include temperature estimations and errors for various components of the PMSM, along with operational parameters such as speed and torque over time. The results are visualized in two separate figures, each showing 10 subplots of temperature estimation results and errors with random operation and the Federal-Test-Procedure-75_driving_cycle (the driving cycle is not put in the manuscript).
*Fig.1 Temperature Estimation results under random operation.
*Fig.2 Temperature Estimation results under Federal-Test-Procedure-75_driving cycle.
The first row of the subplot matrix represents the speed and torque of the PMSM over time. These operational parameters are crucial as they impact the thermal behavior of the motor.
Subsequent rows show the actual and estimated temperatures for different components of the PMSM (bearing, rotor, active wind, etc.). The estimations are made using an LSTM network that has been trained and validated using K-fold cross-validation.
The second column of the subplot matrix highlights the estimation error, which is the difference between the actual data and LSTM predictions. This error analysis is critical for understanding the model's accuracy and identifying areas for improvement.
To reproduce the results and visualizations, follow these steps:
- Every code is in
lptn_informed_LSTM.ipynbincluding the third-order LPTN, definition and training of LPTN-informed LSTM, hyperparameter searching .- Ensure you have a suitable GPU (tested with RTX2060) to run the hyperparameter searching which at the end of this nootboke, and that the estimated runtime of approximately 11.5 hours is acceptable for your setup.
- The
lptn_informed_LSTM.ipynbfile will be uploaded once the manuscript is available in early access.
Contributions to the project are welcome. Please ensure to follow the existing code structure and document any changes or additions clearly.
The lptn-informed LSTM is first introducted at (Early Access)
@ARTICLE{10547453,
author={Liu, Zirui and Kong, Wubin and Fan, Xinggang and Li, Zimin and Peng, Kai and Qu, Ronghai},
journal={IEEE Transactions on Power Electronics},
title={Hybrid Thermal Modeling with LPTNInformed Neural Network for Multi-Node Temperature Estimation in PMSM},
year={2024},
volume={},
number={},
pages={1-12},
keywords={Estimation;Parameter estimation;Temperature sensors;Mathematical models;Motors;Uncertainty;Long short term memory;Lumped-Parameter Thermal Network (LPTN);Long Short-Term Memory (LSTM);temperature estimation;parameter identification;PMSM},
doi={10.1109/TPEL.2024.3409388}}
For any queries or further discussion regarding the project, please open an issue in this repository or direct connect ziruiliu@hust.edu.cn.