Stacking Ensembles-based Approach for Pattern Diagnosis of Tabular Data

The introduction of an automation system in facilities has enabled the continuous flow of data streams, facilitating easier and more accessible data collection. Moreover, advancements in machine learning and deep learning techniques have made it possible to detect abnormalities in data streams, such as electrical energy and equipment diagnosis data, almost in real-time. Early detection of such anomalies allows for proactive maintenance and cost savings, ensuring stable operation of the entire system. Conventionally, the State of Charge (SOC) and State of Health (SOH) have been used as major criteria for assessing facility abnormality in electrical energy data. However, these measures are not suitable for real-time diagnostics due to their design for post-discharge battery analysis. In this paper, we propose the utilization of ensemble stacking technique from the ensemble models to enable accurate and real-time health condition monitoring of electrical equipment using open data sources. Additionally, we perform a comparative analysis among various machine learning and deep learning techniques and optimize the model using a grid search algorithm to achieve a high-performing, robust, and generalizable health condition monitoring system.