A Data-mining Method to Assess Automatic Generation Control Performance of Power Generation Units

Abstract

Automatic generation control (AGC) of power generation units aims at providing a satisfactory response of generated active power to desired active power dispatched from a power grid center. This paper proposes a data-mining method to estimate three metrics assessing the AGC performance of power generation units in terms of response latency, rapidity and accuracy. The proposed method is composed by two parts. The first part is to select data segments via a matrix profile technique from long-term data samples of the desired and generated active powers. The second part is to estimate performance metrics from a dynamic model between the desired and generated active powers, where the model is built by a system identification technique from the selected data segments. The proposed method resolves a major challenge that the performance metrics are defined for step responses, but the desired active power in practice changes in various forms, many of which are not suitable for performance assessments. An industrial example is provided to support the proposed method.