Abstract:
High and medium-voltage transmission lines play a critical role in supplying electricity to urban areas. Manual inspection of these extensive networks is impractical due to their vast reach, leading to potential damage and inefficiency. Additionally, partial discharges (PD), resulting from insulator malfunctions and external impacts, pose a significant threat to the integrity of power lines. Detecting PD promptly is essential to prevent equipment damage and ensure reliable power transmission. In this study, we propose a Bidirectional Long Short-Term Memory (BiLSTM) deep neural network approach to detect PDs. Article BiLSTM model is pre-trained using a unique dataset collected from real power lines with a custom-designed meter by the VSB-Technical University of Ostrava. Notably, the dataset includes real-world signal data with inherent background noise, necessitating preprocessing techniques such as central tendency, statistical dispersion, entropy, and fractality analysis. To enhance classification accuracy, we introduce a novel approach that merges two neural network models, Model160 and Model_320, each utilizing distinct input data. The hybrid model, created by combining these models, achieves remarkable performance, with an accuracy of 97.5% and a Matthews Correlation Coefficient (MCC) of 0.79. Importantly, our hybrid model surpasses Kaggle-winning solutions and other state-of-the-art models trained on the same data, demonstrating its superior capability for PD detection. This research showcases the effectiveness of combining two distinct neural network systems to achieve a synergistic enhancement in performance, setting a new standard for PD detection in high-voltage transmission lines.
|