A Generating Method for Simulation Sample Sets of Dynamic Frequency Response in Power Systems for Deep Learning Models

The integration of renewable energy sources in the process of upgrading and transitioning to new-type power systems has introduced challenges related to frequency stability. Deep learning models, with their robust capability in mapping nonlinear relationships, offer a powerful approach for solving the complex nonlinear differential-algebraic equations of power systems to conduct frequency stability analysis. However, the quality and scale of the sample set are highly dependent on the quality and scale of the sample set utilized. This paper proposes a method for generating simulation sample sets that accounts for the uncertainties and probability distribution characteristics of renewable energy sources and loads through random probability sampling is proposed. Utilizing simulation software, probabilistic values are assigned to the output of all renewable energy sources and the power demand of all loads in the system, thereby generating various operating modes. Subsequently, active power disturbance faults are introduced to rapidly generate a large number of samples. Case studies demonstrate the superiority of the generated sample set compared to those in existing studies, as evaluated by multi-criteria assessment.