Simulation-based multihorizon LSTM-transformer prediction for live-line crossing platform stability
Abstract
Live-line crossing platforms are exposed to coupled wind, operational, and impact loads while electrical clearance must be maintained. This paper presents a multi-scale LSTM-Transformer for forecasting the continuous overturning stability margin of an insulated telescopic-boom platform at 5s and 30s horizons. Because field measurements were unavailable, a disclosed rigid-flexible simulation generated strain, tri-axial acceleration, electromagnetic force, wind speed, tilt, and rope tension signals for normal operation, sudden gust, line-drop impact, and worker imbalance. The model combines a short scale LSTM branch with downsampled Transformer attention and uses a risk-weighted loss. Against a fixed threshold, GRU, LSTM, and Transformer, the proposed method achieved the lowest known-scenario error (MAE 0.0609, RMSE 0.1291) and the strongest unseen-scenario regression (MAE 0.0489, RMSE 0.1094). Its known-scenario 5s warning recall was 84.6%; no learned model recalled the sparse unseen threshold crossings. No model anticipated abrupt 30s crossings, identifying a practical observability limit rather than supporting an unsafe long-horizon claim.