On-line power systems security assessment using data stream random forest algorithm modification

Abstract

Voltage instability is among the main factors causing large-scale blackouts. One of the major objectives of the Control centers is a prompt assessment of voltage stability and possibly self-healing control of electric power systems. The standing alone solutions based on classical approximation methods are known to be redundant and suffer with limited efficiency. Therefore, the state-of-the-art machine learning algorithms have been adapted for security assessment problem over the last years. This chapter presents an automatic intelligent system for on-line voltage security control based on the Proximity Driven Streaming Random Forest (PDSRF) model using decision trees. The PDSRF combined with capabilities of Lindex as a target vector makes it possible to provide the functions of dispatcher warning and “critical” nodes localization. These functions enable self-healing control as part of the security automation systems. The generic classifier processes the voltage stability indices in order to detect dangerous pre-fault states and predict emergency situations. Proposed approach enjoy high efficiency for various scenarios of modified IEEE 118-Bus Test System enabling robust identification of dangerous states.