With the rapid integration of wind and solar energy into distribution grids, the high-frequency fluctuations of renewable energy output pose significant challenges to the Software-Defined Power Communication Converged Network (SPN), particularly in terms of dynamic communication load surges and inefficient topology adaptation. This study addresses these critical issues by leveraging the State Grid Wind-Solar Competition Dataset to develop a data-driven SPN monitoring and optimization framework. Key innovations include three core components: (1) an attention-enhanced long short-term memory (LSTM) model fused with Variational Mode Decomposition (VMD) for SPN communication delay prediction, which prioritizes high-correlation features from the dataset and achieves a mean absolute error (MAE) of 8.2 ± 0.3 ms for 1-hour-ahead prediction (the predicted target is explicitly defined as SPN communication delay in milliseconds, and the communication load index L t (in Mbps) is a key influencing feature); (2) a wind-solar fluctuation grade-driven SPN topology self-optimization method, which classifies renewable fluctuations into five grades via K-Means clustering and dynamically adjusts routing weights considering Flexible Ethernet (FlexE) hard slicing, reducing average communication latency by 62.0 ± 3.2% and packet loss rate from 5.8 ± 0.4% to 0.3 ± 0.1%; (3) a cross-climate SPN communication fingerprint library integrating seasonal features, which integrates statistical characteristics of “wind-solar features-communication load-seasonal fluctuations” from multi-climate sites, shortening new site debugging time by 97.9 ± 0.5% (from 24 to 0.5 h) and reducing post-matching latency by 26.8 ± 1.3%. Comprehensive experimental validation (10 random seed runs) demonstrates that the proposed framework effectively bridges the gap between renewable energy dynamics and power communication network operations.
Wei et al. (Fri,) studied this question.