Scholarlink

<p>To address the problems of current AC contactor state classification methods, such as reliance on manual rules or supervised learning, high labeling costs, and limited generalization, this paper proposes an AC contactor state classification method based on Variational Long Short-Term Memory Autoencoder (VLAE) and Gaussian Mixture Model (GMM). First, VLAE is employed to learn long-term temporal dependencies and dynamic changes, while constraining the latent features to follow a continuous and smooth Gaussian distribution. Then, GMM is applied for clustering in the latent space to realize automatic state classification. Case analysis shows that the proposed method outperforms traditional methods (e.g., LSTM-AE + K-means) in terms of Davies-Bouldin Index (DBI), Calinski-Harabasz (CH) Index, and Silhouette Coefficient. Visualization via t-distributed Stochastic Neighbor Embedding (t-SNE) reveals clear separation of state clusters in the latent space, further verifying the effectiveness of the proposed method and providing a cost-effective solution for the intelligent operation and maintenance of industrial equipment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

State Classification of AC Contactors Using Variational LSTM Autoencoder and GMM

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Shuxin Liu,
  • Kai Huang,
  • Chaojian Xing,
  • Yankai Li,
  • Jing Xu

摘要

To address the problems of current AC contactor state classification methods, such as reliance on manual rules or supervised learning, high labeling costs, and limited generalization, this paper proposes an AC contactor state classification method based on Variational Long Short-Term Memory Autoencoder (VLAE) and Gaussian Mixture Model (GMM). First, VLAE is employed to learn long-term temporal dependencies and dynamic changes, while constraining the latent features to follow a continuous and smooth Gaussian distribution. Then, GMM is applied for clustering in the latent space to realize automatic state classification. Case analysis shows that the proposed method outperforms traditional methods (e.g., LSTM-AE + K-means) in terms of Davies-Bouldin Index (DBI), Calinski-Harabasz (CH) Index, and Silhouette Coefficient. Visualization via t-distributed Stochastic Neighbor Embedding (t-SNE) reveals clear separation of state clusters in the latent space, further verifying the effectiveness of the proposed method and providing a cost-effective solution for the intelligent operation and maintenance of industrial equipment.