Unsupervised Anomaly Detection in Electric Power Networks Using Multi-Layer Auto-Encoders

With significant advancements in electric power networks, maintaining robust health monitoring has become a critical issue. The traditional statistical approaches, while efficient, often struggle to cope with the complex, high-dimensional data provided by devices such as Phasor Measurement Units (PMUs), which offer high-resolution, real-time monitoring of the power grid. To address this, we introduce a novel, unsupervised machine learning framework for robust health monitoring of electrical power networks, which transforms the complex PMU data into insightful, actionable information. Our approach employs a multi-layer autoencoder, a type of artificial neural network known for its proficiency in handling high-dimensional data. This model comprises two stacked layers with 25 neurons and 15 neurons, respectively, which utilizes a non-linear logarithmic sigmoid function in the encoder phase and a linear function in the decoder. The autoencoder is trained under normal conditions, minimizing reconstruction errors, and any significant increase in these errors during real-time monitoring is flagged as an anomaly, indicative of potential issues in the grid's operation. The results from our study have demonstrated the significant potential of our approach. Our model successfully identified anomalies with a high precision of 0.9983, a recall of 0.9841, and an F1 score of 0.9920, highlighting its accuracy and reliability. Furthermore, via comprehensive data visualization techniques, we effectively delineated the reconstructed data and highlighted the anomalies detected by our system. This research paves the way for more robust and reliable health monitoring of electrical power networks, substituting conventional statistical methods with a sophisticated, unsupervised machine learning approach. By facilitating real-time anomaly detection, our framework enables timely intervention, which enhances the robustness and reliability of power networks, and ensuring uninterrupted service to consumers. Thus, our approach stands as a significant advancement in the mission for efficient and reliable health monitoring of modern power networks. By employing a multi-layer autoencoder for real-time monitoring and anomaly detection leveraging digital engineering methods to ensure operational integrity and resilience in power systems.