The recognition method of external force damage sources vibration signals based on LSTM-CNN-CatBoost-GSSSA

Area of Science:

• Electrical Engineering
• Signal Processing
• Machine Learning

Background:

• Underground All-Dielectric Self-Supporting (ADSS) optical cables in urban environments face threats from construction and excavation, impacting power grid security.
• Existing single-algorithm methods lack accuracy in recognizing complex urban disturbances affecting buried cables.

Purpose of the Study:

• To develop a robust hybrid recognition model for accurately diagnosing external force damage to urban buried ADSS optical fiber cables.
• To improve the reliability and operational safety of power communication networks through advanced monitoring techniques.

Main Methods:

• Introduced a hybrid model integrating Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNN) for temporal and spatial-frequency feature extraction.
• Employed an improved golden sine-enhanced Sparrow Search Algorithm (GSSSA) for hyperparameter optimization of CatBoost and adaptive feature weighting.
• Utilized differentiable learning for dynamic weighted fusion of time series and spatial-frequency domain features.

Main Results:

• The proposed LSTM-CNN-CatBoost-GSSSA model achieved a recognition accuracy of 97.6%.
• Demonstrated superior performance compared to SVM, CNN, LSTM-CatBoost, CNN-CatBoost, LSTM-CNN-GBDT, and LSTM-CNN-CatBoost models.
• The model offers high recognition performance and computational efficiency for real-time applications.

Conclusions:

• The LSTM-CNN-CatBoost-GSSSA model provides an effective technical solution for real-time monitoring and intelligent diagnosis of damage to urban buried ADSS optical fiber cables.
• This advancement significantly contributes to enhancing the operational safety and reliability of power communication networks.