Abstract
Polyimide exhibits outstanding thermal stability, electrical insulation, and mechanical strength, making it widely used as an insulating material in high-power-density motor windings for aerospace applications. The glass transition temperature, as a key parameter characterizing thermal stability of polyimide, provides crucial guidance for structural design and performance optimization of motor insulation materials under high-temperature service conditions. Machine learning methods, as an emerging technology, have been widely applied in the prediction of polyimide glass transition temperatures. Prediction models typically utilize molecular descriptors as input features to characterize molecular structures and their physicochemical properties. However, traditional molecular descriptors often exhibit high dimensionality and strong redundancy, leading to reduced prediction accuracy and limited generalization capabilities when directly used as model inputs. To address this, this paper proposes a two-stage feature selection method based on PCA-LASSO. First, PCA is employed to reduce redundancy and noise in high-dimensional features. Then, LASSO is used for feature selection to identify key features strongly correlated with the glass transition temperature of polyimides. The reduced-dimension features are then fed into a DNN model for glass transition temperature prediction. Results demonstrate that the R2 value of the prediction results obtained using this method reached 0.84. Compared to feature selection methods without dimensionality reduction, or those using LASSO and PCA alone, the R2 improved by 71.43%, 29.23%, and 10.53%, respectively. The model demonstrates high accuracy and excellent generalization capability. This method provides a reliable theoretical basis for the design and optimization of highly heat-resistant polyimide materials.
| Original language | English |
|---|---|
| Title of host publication | Proceedings of 2026 IEEE 9th International Electrical and Energy Conference, CIEEC 2026 |
| Publisher | Institute of Electrical and Electronics Engineers Inc. |
| Pages | 728-733 |
| Number of pages | 6 |
| ISBN (Electronic) | 9798331549558 |
| DOIs | |
| State | Published - 2026 |
| Event | 9th International Electrical and Energy Conference, CIEEC 2026 - Tianjin, China Duration: 15 May 2026 → 17 May 2026 |
Publication series
| Name | Proceedings of 2026 IEEE 9th International Electrical and Energy Conference, CIEEC 2026 |
|---|
Conference
| Conference | 9th International Electrical and Energy Conference, CIEEC 2026 |
|---|---|
| Country/Territory | China |
| City | Tianjin |
| Period | 15/05/26 → 17/05/26 |
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 7 Affordable and Clean Energy
Keywords
- PCA-LASSO
- feature selection
- glass transition temperature
- machine learning
- polyimide
Fingerprint
Dive into the research topics of 'Prediction of Polyimide Glass Transition Temperature Based on Two-Stage Feature Selection Using PCA-LASSO'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver