Hot metal temperature is a key factor affecting the quality and energy consumption of iron and steel smelting. Accurate prediction of the temperature drop in a hot metal ladle is very important for optimizing transport, improving efficiency, and reducing energy consumption. Most of the existing studies focus on the prediction of molten iron temperature in torpedo tanks, but there is a significant research gap in the prediction of molten iron ladle temperature drop, especially as the ladle is increasingly used to replace the torpedo tank in the transportation process, this research gap has not been fully addressed in the existing literature. This paper proposes an interpretable hybrid deep learning model combining Bi-LSTM and Transformer to solve the complexity of temperature drop prediction. By leveraging Catboost-RFECV, the most influential variables are selected, and the model captures both local features with Bi-LSTM and global dependencies with Transformer. Hyperparameters are optimized automatically using Optuna, enhancing model performance. Furthermore, SHAP analysis provides valuable insights into the key factors influencing temperature drops, enabling more accurate prediction of molten iron temperature. The experimental results demonstrate that the proposed model outperforms each individual model in the ensemble in terms of R², RMSE, MAE, and other evaluation metrics. Additionally, SHAP analysis identifies the key factors contributing to the temperature drop.

Zero-calibration cross-subject steady-state visual evoked potential brain–computer interfaces (SSVEP-BCIs) support plug-and-play interaction, yet short-window decoding for unseen users remains difficult because of inter-subject variability, source-to-target shifts, and limited frequency-discriminative evidence. This study proposes DP-Net, a dual-path network for zero-calibration cross-subject SSVEP decoding under short time windows. DP-Net integrates a data-driven neural path (DDNP) with a template-guided analytical path (TGAP). DDNP learns nonlinear spatiotemporal representations through cross-electrode spatial filtering, multi-scale temporal modeling, and global self-attention. TGAP derives template-similarity scores from standard harmonic and source-domain empirical templates using QR/SVD. An analytical score preservation module (ASPM) retains three score types for fusion. The feature vectors are concatenated, batch-normalized, and classified by an MLP. DP-Net was evaluated on the Benchmark and BETA datasets under subject-level leave-one-subject-out cross-validation. Among the six reproduced baselines, DP-Net achieved higher mean accuracy–ITR performance in most tested windows, particularly from 0.2 to 0.6 s. At 0.6 s, DP-Net achieved 65.48% accuracy and 140.06 bits/min on Benchmark, and 47.01% accuracy and 83.11 bits/min on BETA. Ablation, fusion, and feature-intervention analyses indicated complementary contributions from DDNP and TGAP. The results support the use of template-derived scores with learned EEG representations for short-window zero-calibration cross-subject SSVEP decoding.