Application of data mining to optimize settings for generator tripping and load shedding system in emergency control at Hydro‐Québec

Abstract

This paper describes the on‐going work done by Hydro‐Québec to optimize the settings of automatic devices installed in its main power plants to maintain secure operation under extreme contingencies. The automatic generator tripping and load shedding system (RPTC) described in this paper is installed at the Churchill Falls hydroelectric power plant (5,500 MW) in Labrador. Data mining techniques such as decision trees and regression trees have been used. Real time snapshots of the Hydro‐Québec power system collected over a 5 year period have been used to generate large amounts of results by transient stability simulations. The processing of these data has been done using software developed by the University of Liege. This approach gives the most relevant parameters and finds optimal settings for the RPTC system, minimizing the number of tripped generator units while maintaining the same performance in terms of security coverage. New operation rules can thus be established.

Role of Statistical and Machine Learning Methods in Special Protection Scheme Logic Design and Failure Assessment

“…[15,23,24], has enhanced the process of deriving SPS operational logic from off-line computations. Typically, the operational logics derived from DTs help to differentiate secure operating regions from insecure in the space of operating parameters, such as line flows, generation, and loading conditions.…”

“…However, in current power systems faced with many uncertainties, deterministic techniques may not help in optimal design of SPS logic (such as ideal settings for generator rejection or load-shedding schemes) [15]. In this case, Monte Carlo simulation based operational planning studies [16] are promising.…”

Automatic learning of sequential decision strategies for dynamic security assessment and control

“…During the subsequent decades the progress in computational means (both hardware and software) has been tremendous, and automatic learning has matured as a rich and highly productive research field, today one of the most active areas within computer science, systems theory, computational mathematics and statistics. Nevertheless, real-world power systems security related applications of automatic learning are almost unexisting, with the notable exception of a few European TSOs (RTE, NGT, ELIA) in the context of planning and operational planning [4] and the use at Hydro-Québec for pre-setting of special protection systems [5].…”