Conventional droop control in multi-parallel grid-forming inverters exhibits poor reactive power sharing accuracy due to line impedance mismatches.In this study

we proposed a coordination control strategy integrating adaptive virtual impedance with dynamic Q-V droop regulation to overcome this issue.We established a coupling model between the line impedance and power allocation to determine the quantitative relationship between reactive power deviation and impedance difference and to analyze the mechanism of reactive power deviation formation.Based on this

we proposed a transformer neural network-based online identification method for adaptive virtual impedance and dynamic droop coefficients.The self-attention mechanism dynamically characterizes the spatial distribution features of the impedance parameters considering the real-time voltage/reactive power time-series data as inputs to realize the dynamic impedance compensation without communication interaction.The contradiction constraint between the voltage drop and distribution accuracy caused by the introduction of conventional virtual impedance is improved by dynamic droop coefficient reconstruction.Lastly

we established a hardware-in-the-loop simulation platform to experimentally validate the operational efficacy and dynamic performance of the proposed control strategy under various grid scenarios.