Applications of Decision Tree and Random Forest Methods for Real-Time Voltage Stability Assessment Using Wide Area Measurements

Traditionally, voltage stability assessment (VSA) are widely investigated by model-based approaches. Several achievements have been developed along this direction, including continuation power flow methods (CPFLOW), direct methods, and optimal power flow methods. Since precise model information are required and their computations are very demanding, their applications to real-time VSA are challenging, especially when network operating conditions and/or network topology may be always changed. In recent years, with wide deployment of synchronized phasor measurement unit (PMUs), PMU-based wide area measurement system (WAMS) has already attracted lots of interests in investigating VSA in advanced artificial intelligence approaches. By collecting real-time big data from power grids and studying these historical data through statics analytics, some prediction models can be constructed for VSA of the current operation conditions. This chapter presents some recent advances in data mining framework for power system VSA under real-time environments. The proposed framework adapts a new enhanced online random forest (EORF) algorithm to update decision trees (DTs), such as tree growth and replacement. By means of weighted majority voting, one of the ensemble learning skills, DTs in the random forest are able to reach consesus to deal with power system changes. The proposed EORF framework is first tested on IEEE 57-bus power systems, and then is applied to Taiwan 1821-bus power system. Through comprehensive computer simulations, the robustness, the computation speed, as well as the assessment accuracy, of the proposed EORF framework are justified for assessing the power system voltage stability in real-time.