Multi-level interactive self-balancing optimization strategy of source-grid-load-storage considering cluster security constraints

As the global energy structure undergoes transformation, large-scale access to renewable energy presents the power system with unprecedented dynamic balance challenges. Traditional centralized power supply architecture is challenging to adapt to complex scenarios where high proportions of new energy, high-density power electronic devices, and diversified load demands are intertwined. There is an urgent need to build a new balancing mechanism for collaborative interaction between source, grid, load and storage. Aiming at the scientific problem of deep integration of cluster security constraints and multi-level interaction, this study proposes an integrated optimization strategy. By quantitatively characterizing the CIA (Confidentiality, Integrity, Availability) triple security criterion, it establishes three types of constraint models, including extreme weather equipment current carrying capacity correction coefficient, node health index and adjustment instruction convergence time threshold. Experimental verification demonstrates that this strategy effectively controls the system frequency deviation within 0.010 Hz and stabilizes the voltage deviation to below 1.50% during the 16-period scheduling cycle. At the same time, it improves energy utilization efficiency to 92%, with clean energy accounting for 61%. Carbon emissions were reduced to 10,200 tons, and pollutant emissions were reduced to 5,100 tons. The direct trust of the high-precision recommendation module in the system is positively correlated with its precision trust value, and the source-grid-load-storage (SGLS) samples exhibit significant differences in aggregation characteristics under different feature representation methods. In addition, the short-circuit current capacity of the system is stable at 31 kA, and the transient stability index reaches 0.94, which verifies the robustness of the strategy under extreme working conditions. By analyzing the dynamic influence of the ψ parameter on the detection results of each cluster, the effectiveness of the security constraint embedding method is further confirmed.