Lazy deep Q networks for unified rotor angle stability framework with unified time-scale of power systems with mass distributed energy storage

Abstract

The addition of increasing numbers of new energy sources can lead to challenges to the security and stability of power systems. The combined rotor angle stability (RAS) framework of power systems with multiple time-scales could lead to uncoordinated stability control commands for numerous stability control devices of power systems. To mitigate this deficiency, this study proposes a unified RAS framework to replace the RAS with three time-scales, which consist of small-disturbance, large-disturbance, and long-term RASs. This study proposes lazy deep Q networks for the unified RAS framework of power systems. The lazy deep Q networks approach, which is based on the Markov decision process, contains lazy learning, deep Q networks, a state-buffer, a selector operation, and a limiter operation. The unified stability framework and the conventional combined stability framework of power systems are compared under four cases, i.e., the power system with four generators, the power system based on IEEE 39-bus system, the European high voltage 89-bus system, and the European high voltage 1354-bus system. These four case studies verify that the unified stability framework based on the lazy deep Q networks approach is more stable than the conventional combined stability framework based on the combination of optimization algorithm of long-term dynamic stability, control method of middle-term transient stability, and control method of short-term static stability. The error indices for the proposed lazy deep Q networks algorithm are as follows: the AAE is at least 4.7% lower than the other compared algorithms; the IAE is at least 4.8% lower than the other compared algorithms; the ISE is at least 3.9% lower than the other compared algorithms; the ITAE is 2.3% lower than the other compared algorithms.