Deep Entropy-Learning based virtual inertia control for VRFB regulation considering Phase-Locked loop dynamics

Abstract

The virtual inertia control (VIC) technique is often adopted as a prominent mechanism to address the inertia deficiency of modern integrated power systems with high penetration of sustainable energy units. In such control mechanisms, a phase-locked loop (PLL) is required to estimate the grid frequency. However, utilization of such frequency measurement platforms is accompanied by large frequency fluctuations due to its inherent dynamics. In particular, this paper proposes an adaptive integral recursive backstepping (I-RBSC) based on virtual inertia for the regulation of vanadium redox flow battery (VRFB) in stand-alone Micro-Grids (SMGs). Unlike previous works that neglected the effect of frequency measurement devices, this work investigates the effect of PLL dynamics in virtual inertia control. The I-RBSC controller is designed by an entropy-driven actor-critic neural networks (NNs) to adaptively adjust the tunable coefficients during its interaction with the SMG. By training the ability of deep NNs, the entropy is maximized to regulate the system output. The VRFB unit can be discharged up to 90% which can quickly respond to the randomness of loads and sustainable energy units. The transient outcomes of SMG reveal the feasibility of I-RBSC (designed by entropy learning) to address the challenges of low inertia and PLL dynamics. The real-data of wind turbine and solar radiation are utilized to simulate the SMG in a more realistic framework from a systematic point of view. The comprehensive analysis under typical scenarios confirms the supremacy of the suggested VIC controller over prevalent state-of-the-art virtual inertia controllers including conventional VIC, fuzzy logic, backstepping, and model-predictive control (MPC).