Deep-Learning Approach Featured Problematic Parameters Selection and Calibration for Load Models

Regular validation and calibration of load models are essential for accurately analyzing the dynamic behavior of power systems. With the increasing integration of distributed energy resources (DERs), traditional load models are becoming progressively insufficient in accurately reflecting the low inertia and high stochasticity characteristics of DERs. In addition, the structure of load models has become increasingly complex, making it challenging to locate the problematic parameters (PPs) that result in the errors of model outputs. Therefore, this paper proposes a novel load model that incorporates grid-forming photovoltaics into the traditional composite load model. In addition, a novel method to select and calibrate the PPs has also been developed. First, based on the event playback, the dynamic response of the load model is compared with actual PMU data to check whether the model calibration is needed. Second, a Transformer network is employed to locate the PPs that truly cause model errors. The Transformer can automatically learn the relationship between the parameter perturbations and the mismatches of model outputs. Finally, an advanced PSO-CSA algorithm is applied to calibrate the selected PPs accurately. The case studies on the IEEE 9-Bus system verified the effectiveness of the proposed approach.