Dynamic Line Rating Forecasting under False Data Injection Attacks Using Hybrid GRU–ELM Model with Hyperband Optimization

Dynamic Line Rating (DLR) forecasting enhances the operational flexibility and safety of overhead transmission lines by adapting ampacity to real-time environmental conditions. However, DLR prediction can be degraded by False Data Injection Attacks (FDIA) that corrupt inputs and threaten secure grid operation. This study presents a cyber-resilient DLR forecasting framework that combines a hybrid Gated Recurrent Unit–Extreme Learning Machine (GRU–ELM) architecture with Hyperband (HB)-based hyperparameter optimization. The proposed GRU–ELM/w HB model is evaluated under clean data and FDIA meteorological input conditions. Its performance is compared against SVM, ELM, LSTM, BiLSTM and GRU using MAE, RMSE, and R2. The proposed GRU-ELM/w HB model achieved substantial performance gains over all individual baseline methods. Although R2 drops sharply for the competing baselines under adversarial perturbations, the proposed model remains highly stable, sustaining R2 above 0.99 in the FDIA case. Compared with the standalone baselines methods, GRU– ELM w/HB reduces the RMSE by 21.57–82.34% under clean conditions and by 70.95–82.42% under FDIA. These results demonstrate that Hyperband-optimized GRU–ELM delivers accurate and FDIA-resilient DLR forecasts, maintaining stable performance under adversarial data conditions.