Short-circuit current amplitude prediction method for power grids with high penetration of new energy

After the integration of high-penetration inverter-based renewable sources such as wind and photovoltaic power, grid short-circuit currents exhibit rapid transient decay, strong amplitude randomness, and complex multi-source coupling. Accurate prediction of short-circuit current amplitude is, therefore, essential for fault management and ensuring the safe and stable operation of new-energy power grids. This paper proposes an intelligent prediction framework that integrates physical priors with data-driven learning—an improved particle swarm optimization–incremental extreme learning machine (PSO-IELM) algorithm—for high-precision prediction of short-circuit current amplitudes following grid faults. The Incremental ELM adaptively determines the optimal number of hidden-layer nodes to balance model complexity and generalization, while particle swarm optimization jointly optimizes input weights, hidden biases, and harmonic suppression weights. The fitness function is constructed using the gray absolute correlation degree to better capture the dynamic trend consistency between predicted and measured sequences. Multiple sinusoidal activation functions enhance the nonlinear fitting capability of the extreme learning machine (ELM). A simulated short-circuit fault model is established to compare the predictive performance of various algorithms. Experimental verification demonstrates that the proposed PSO-IELM-GAC method achieves an average relative error as low as 1.22% under full-line and full-phase fault conditions, significantly outperforming traditional BP neural networks, standard ELM, and its variants. The single prediction time of 2.9 ms meets the requirements of real-time protection in high renewable-penetration power grids.