A Robust Parameter Identification Strategy of Composite Load Model with a Neural Differential Algebraic Network

Recently, the ambient signal (AS) based load identification method has been favored by researchers due to its ability to capture the time-varying nature of load characteristics. However, load characteristics are not sufficiently perturbed by small disturbances, leading to the easy distortion of effective signals in AS, and inaccurate identification results that cannot reflect the actual load composition and model parameters. To address this issue, this paper proposes a real-time load composition estimator based on a neural differential-algebraic equations network (NDAE) to guide the parameter optimization process. Moreover, considering the redundancy of AS, a hierarchical strategy based on the verification and synthesis of multiple sets of identification results is designed to improve the reliability of the final conclusion. The effectiveness of the proposed strategy is verified using the WSCC 9-node simulation system.