Power Oscillation Source Location Based on the Combination of Energy Function and Logistic Regression in a Fully Data-Driven Approach

Abstract

With the increasing emphasis on green energy transformation, power systems are evolving into a “double high” structure characterised by a high integration of renewable energy sources and extensive use of power electronics. This transformation leads to more complex system topologies, necessitating improvements in monitoring and control measures. Traditional model-based approaches for identifying power oscillation disturbance sources are increasingly inadequate for the demands of modern power systems. The rapid development of Wide Area Measurement Systems (WAMS) has heightened interest in leveraging system response data for disturbance source localisation. This paper introduces a data-driven numerical method that combines energy functions with logistic regression, enhancing localisation accuracy by utilising power oscillation mechanisms and response data—and specifically improving accuracy by 15–22% over traditional methods. The proposed method identifies potential disturbance sources, ranging from minor random load fluctuations to significant forced power oscillations. A key innovation is the application of logistic regression for the automatic classification and localisation of disturbance sources, reducing the need for manual intervention and addressing the limitations of traditional energy function methods. Validation on WSCC 9-bus, New England 39-bus and 197-bus systems demonstrates 99.5% accuracy for single-source disturbances and 84.5-96.1% for multi-source scenarios, outperforming SVM (98.6%) and LDA (95.4%) while reducing computation time to 0.03s. By quantifying disturbance source localisation in power oscillations, this approach significantly enhances both the accuracy and efficiency of the localisation process.