Frequency-aware ICEEMDAN-ARIMA-LSTM for real-time wind speed forecasting in dynamic line rating

Feng Yang , Xiao Qin , Zhengkang Li , Kai Ji

Abstract

Short-term wind speed forecasting is critical for dynamic line rating (DLR) to maximize grid capacity, yet existing methods face challenges in handling nonlinearity and stochasticity. We propose a novel hybrid model integrating Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), ARIMA, and LSTM, with three key contributions: (1) An enhanced ICEEMDAN algorithm reducing residual noise energy by 5-8% and mode mixing by 31% compared to CEEMDAN; (2) A frequency-aware modeling strategy that dynamically assigns linear (ARIMA) and nonlinear (LSTM) sub-models based on Hurst exponent analysis; (3) A GPU-accelerated implementation achieving real-time prediction with 28-second latency. Validated on China's transmission corridors, the model reduces RMSE by 65.2% over standalone LSTM and increases line ampacity by 21.2% compared to static ratings. Its robustness (99.2% availability during typhoons) and computational efficiency (150 x faster than conventional systems) demonstrate significant potential for smart grid applications.

Key Words

dynamic line rating; empirical mode decomposition; hybrid model; real-time forecasting; wind speed prediction

Address

Feng Yang — College of Energy and Mechanical Engineering, Shanghai University of Electric Power, Shanghai 201306, China
Xiao Qin — College of Energy and Mechanical Engineering, Shanghai University of Electric Power, Shanghai 201306, China
Zhengkang Li — College of Energy and Mechanical Engineering, Shanghai University of Electric Power, Shanghai 201306, China
Kai Ji — College of Energy and Mechanical Engineering, Shanghai University of Electric Power, Shanghai 201306, China

PDF Viewer

Preview is limited to the first 3 pages. Sign in to access the full PDF.

← Back to Volume 42, No. 4