Aiming to address the issue that traditional steel leakage forecasting models lack sufficient time-sequence modeling capability and most of them do not process the thermocouple temperature data,which is directly used as model input,resulting in high computational complexity.By analyzing the formation mechanism and time-sequence characteristics of adhesive steel leakage and leveraging the Long Short-Term Memory network(LSTM),adding the Attention Mechanism that can focus on the key information,a deep learning-based Attention-LSTM forecasting model was proposed for predicting steel leakage.Firstly,the thermocouple temperature data is processed through data augmentation,inter-frame differencing,threshold segmentation,etc.,and the common features of bonded steel leakage are extracted as model inputs.Secondly,the model is constructed by traversing the selected parameters for training,using the mean square error as the model's loss function to find the optimal forecasting model.Lastly,to enhance the model's prediction accuracy,the Attention Mechanism module is added in front of the Dense output layer to complete the model optimization.The model is tested by applying the continuous casting field data,and the test results show that the proposed Attention-LSTM steel leakage forecasting model verifies the feasibility and effectiveness of the model with a 97.3％forecasting rate and 100％reporting rate.
NETL
NSTL
万方数据
