To address the challenges in partial discharge (PD) monitoring of ultra-high voltage (UHV) equipment, such as the reliance on empirical parameter selection for variational mode decomposition (VMD) and the high misidentification rate of noise-dominated modes, this paper proposes a sparrow search algorithm (SSA)-driven VMD parameter optimization and joint screening mechanism. Firstly, an SSA optimization model is constructed with the minimum average envelope entropy (MAEE) as the fitness function, enabling global adaptive matching of the VMD modal number (K) and penalty factor (α). This approach overcomes the limitations of traditional grid search methods, such as low efficiency and susceptibility to local optima. Secondly, a joint screening mechanism based on kurtosis, correlation coefficient, and energy ratio is designed. By integrating pulse characteristics, signal relevance, and energy distribution, a multi-dimensional criterion is established to accurately distinguish noise-dominated modes from effective modes. Simulation experiments demonstrate that under 3 dB Gaussian white noise interference, the proposed method achieves higher parameter optimization efficiency and reduces average iteration counts compared to particle swarm optimization (PSO) and genetic algorithm (GA). Furthermore, the joint screening mechanism exhibits a lower false screening rate than traditional energy entropy-based methods. This approach provides a high-precision and robust modal decomposition solution for PD signal preprocessing in complex noise environments, laying a reliable foundation for subsequent feature extraction.