Anomaly Detection, Time Series Forecasting, CNN, LSTM, High-Voltage Ground Current Abstract: Timely anomaly detection is critical for maintaining stability in industrial systems. However, most existing anomaly detection methods for industrial data focus only on single-variable abrupt changes. As system dimensionality increases, these models struggle to capture intrinsic relationships between multiple variables, compromising detection accuracy and precision. To overcome these constraints, we present AACnL, an innovative anomaly detection algorithm that combines CNNLSTM architecture with a multivariate self-attention mechanism. AACnL first extracts spatiotemporal features from data blocks using an Attention-CNN-LSTM network and reconstructs feature values through a decoder. It then enhances anomaly detection accuracy by explicitly models variable interdependencies through self-attention. Evaluated on real-world high-voltage cable grounding current and line temperature data, AACnL achieved 95% accuracy and precision, with an F1-score exceeding 96%. These experimental results show that AACnL is noticeably more accurately than conventional techniques. This performance validates AACnL's ability to effectively capture and integrate complex multivariate relationships, offering valuable insights for further research in industrial anomaly monitoring. https://mc.manuscriptcentral.com/jcmse Journal of Computational Methods in Science and Engineering For Peer Review Page 1 of 18 https://mc.manuscriptcentral.com/jcmse Journal of Computational Methods in Science and Engineering 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 For Peer Review Anomaly Detection for Multivariate High-Voltage Cable Data Using a CNN-LSTM-Based self-attention Encoding Model Liu Qing, Jiao Yuyang*, Zhang Lantao, Li Jintao, Liu Jinlong, Chen Guoqing, Zhao Chunming, Li Yanguo1, State Grid Beijing Power Cable Company, Beijing, 100022, China Email：ncepu_ashur0925@sina.com Abstract: Timely anomaly detection is critical for maintaining stability in industrial systems. However, most existing anomaly detection methods for industrial data focus only on single-variable abrupt changes. As system dimensionality increases, these models struggle to capture intrinsic relationships between multiple variables, compromising detection accuracy and precision. To overcome these constraints, we present AACnL, an innovative anomaly detection algorithm that combines CNN-LSTM architecture with a multivariate selfattention mechanism. AACnL first extracts spatiotemporal features from data blocks using an Attention-CNNLSTM network and reconstructs feature values through a decoder. It then enhances anomaly detection accuracy by explicitly models variable interdependencies through self-attention. Evaluated on real-world high-voltage cable grounding current and line temperature data, AACnL achieved 95% accuracy and precision, with an F1score exceeding 96%. These experimental results show that AACnL noticeably more accurately than conventional techniques. This performance validates AACnL's ability to effectively capture and integrate complex multivariate relationships, offering valuable insights for further research in industrial anomaly monitoring. Keywords: Anomaly Detection, Time Series Forecasting, CNN, LSTM, High-Voltage Ground Current 1.