Answers are grounded in this paper and cite their sources.
Accurate prediction of icing-induced comprehensive tension loads on transmission lines is critical for icing disaster warning and safe operation and maintenance. Load sequences are affected by noise, wind-induced vibration, and abrupt meteorological changes, resulting in non-stationary fluctuations, local abrupt changes, and long-term dependencies. To address these issues, this paper proposes an SSA-BiGRU-Informer fusion prediction model. First, singular spectrum analysis (SSA) is used to decompose and reconstruct the historical load series, thereby weakening low-energy disturbances. A bidirectional gated recurrent unit (BiGRU) is then employed to extract local bidirectional temporal features, while Informer is used to capture long-timescale dependencies. Experiments based on online monitoring data from a 220 kV overhead transmission line in Zhejiang Province show that the proposed model achieves RMSE, MAPE, MAE, and R 2 values of 332.46 N, 0.64%, 223.08 N, and 0.94, respectively. Overall, SSA-BiGRU-Informer achieved the lowest mean RMSE, MAPE, and MAE and the highest mean R 2 among the compared models. Paired t-tests with Holm-adjusted p-values further supported significant RMSE improvements over three of the four baseline models. Ablation experiments show that SSA, BiGRU, and Informer contribute to noise suppression, local dynamic feature extraction, and long-range dependency modeling, respectively, verifying the effectiveness of multi-module collaborative modeling. The cross-city and cross-line experiment, conducted during the same one-month period on another overhead transmission line located in a different city within Zhejiang Province, shows that the proposed model also performs well, indicating its applicability to tension-load prediction for icing-disaster warning and stress assessment.
No takes yet. Share an insight, caveat, or question.
Peng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: