Thyristor State Evaluation Method Based on Kernel Principal Component Analysis

Zhaoyu Lei, Jianyi Guo, Feng Zheng, Jiayang Li, Lei Wang, Liangshou Hao, Youping FanIEEE Access 2022 · DOI

Introduction

Thyristors serve as the critical "heart" of Ultra-High Voltage Direct Current (UHVDC) transmission systems. Given their central role in power conversion, any degradation or failure can lead to significant grid instability. However, accurately evaluating the aging state of these components is a complex task due to the high dimensionality of operating data and the non-linear nature of electronic degradation.

The Challenge of State Evaluation

Traditional evaluation methods struggle with two primary hurdles:

  1. High Dimensionality: Monitoring a thyristor involves tracking 42 interrelated parameters, including field test data, maintenance logs, and environmental conditions.
  2. Non-linearity: The relationship between these parameters and the actual health of the device is rarely linear. Standard models often fail to capture the subtle, complex interactions that signal early-stage aging.

Existing physics-based models are frequently oversimplified, while standard data-driven approaches often fail to filter out the noise and redundancy inherent in large industrial datasets.

Proposed Methodology

To address these challenges, this research proposes a multi-stage evaluation pipeline that combines advanced dimensionality reduction with fuzzy logic.

Handling Non-linearity with KPCA

The core of the framework is an optimized version of Kernel Principal Component Analysis (KPCA). Unlike standard PCA, which is limited to linear transformations, KPCA uses a kernel function to map data into a high-dimensional feature space where non-linear relationships can be more easily separated and analyzed.

Conceptual mapping of input space to a high-dimensional feature space via a kernel function.

Optimization via IPSO-FDA

Standard KPCA requires manual tuning of kernel parameters, which is impractical for real-time engineering. The authors introduced an Improved Particle Swarm Optimization (IPSO) algorithm, guided by Fisher Discriminant Analysis (FDA) criteria, to automatically determine the optimal settings. This ensures the model maximizes the separation between different aging states.

Integrated Evaluation Framework

Once the data is reduced from 42 dimensions to a few key "Principal Components," these factors are processed through a Fuzzy Comprehensive Evaluation (FCE) model. This stage assigns weights to the components to produce a final health score (G-value), classifying the thyristor into one of five health states.

Flowchart of the integrated thyristor state evaluation framework, illustrating the synergy between KPCA, IPSO, and Fuzzy Evaluation.

Key Insights and Results

The researchers validated their approach using both a benchmark dataset (Spoken Arabic Digits) and real-world field data from 20 thyristors, including units with 10 years of service and those that had failed technical specifications.

  • Information Retention: The model demonstrated that the first three principal components alone could capture 93.76% of the total variance. This allowed the system to discard 39 redundant parameters with negligible loss of critical information.
  • Superior Accuracy: The proposed method achieved 90% accuracy in real-world testing. It successfully identified non-compliant units that traditional methods, such as the Analytic Hierarchy Process (AHP), often flagged as healthy.
  • Identification of Critical Indicators: The analysis revealed that specific factors, namely positive breakdown voltage and reverse blocking leakage current, are the most significant predictors of a thyristor's true aging state.

Conclusion

By combining IPSO-optimized KPCA with fuzzy logic, this framework provides a robust, automated solution for power grid health monitoring. While future work will focus on scaling the system to larger datasets and reducing reliance on subjective maintenance records, this method represents a significant step toward the implementation of "digital twins" for real-time converter station management.

Summary updated

BibTeX
@article{lei2022thyristor,
  author = {Zhaoyu Lei and Jianyi Guo and Feng Zheng and Jiayang Li and Lei Wang and Liangshou Hao and Youping Fan},
  title = {{Thyristor State Evaluation Method Based on Kernel Principal Component Analysis}},
  journal = {IEEE Access},
  year = {2022},
  doi = {10.1109/access.2022.3159711},
}
Figure