Scholarlink

<p>The study aims to present the effectiveness of anomaly detection algorithms in lighting systems based on analyzing records from electricity meters. The road lighting management system operates continuously and in real-time, requiring online anomaly detection algorithms. The paper examines two machine learning-based algorithms: Autoencoder with LSTM-type recurrent neural network and Transformer. The results obtained for these algorithms are compared with a simple mechanism for comparing energy consumption in consecutive periods. Classification metrics such as error matrix, sensitivity, precision, and F1-score were used to evaluate the performance of the algorithms. The analysis showed that the Autoencoder algorithm achieves better accuracy (F1-score = 0.9565) and requires significantly fewer computing resources than the Transformer algorithm. Although less efficient (F1-score = 0.8125), the Transformer algorithm also demonstrates the ability to detect anomalies in the road lighting system effectively. Implementing the Autoencoder algorithm on an actual ILED platform allows anomaly detection with a delay of 15&#xa0;min, which is sufficient to take corrective action. The conclusions of this study indicate the significant advantage of machine learning-based algorithms in anomaly detection in lighting systems, which can significantly improve the reliability and efficiency of urban lighting management.</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

Anomaly detection in urban lighting systems using autoencoder and transformer algorithms

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月15日

查看全文
  • Tomasz Śmiałkowski,
  • Andrzej Czyżewski

摘要

The study aims to present the effectiveness of anomaly detection algorithms in lighting systems based on analyzing records from electricity meters. The road lighting management system operates continuously and in real-time, requiring online anomaly detection algorithms. The paper examines two machine learning-based algorithms: Autoencoder with LSTM-type recurrent neural network and Transformer. The results obtained for these algorithms are compared with a simple mechanism for comparing energy consumption in consecutive periods. Classification metrics such as error matrix, sensitivity, precision, and F1-score were used to evaluate the performance of the algorithms. The analysis showed that the Autoencoder algorithm achieves better accuracy (F1-score = 0.9565) and requires significantly fewer computing resources than the Transformer algorithm. Although less efficient (F1-score = 0.8125), the Transformer algorithm also demonstrates the ability to detect anomalies in the road lighting system effectively. Implementing the Autoencoder algorithm on an actual ILED platform allows anomaly detection with a delay of 15 min, which is sufficient to take corrective action. The conclusions of this study indicate the significant advantage of machine learning-based algorithms in anomaly detection in lighting systems, which can significantly improve the reliability and efficiency of urban lighting management.