数据驱动的水火弯板变形时空动态预测方法

Line heating is a critical process for forming ship hull outer plates, involving complex thermomechanical coupling effects. The deformation process is intricate, making real-time, dynamic, and accurate prediction of its spatiotemporal evolution challenging. To address this issue, this paper proposes a CNN-Double-Layer LSTM-AM model that integrates spatiotemporal features. The model uses a single heating line as a time step to dynamically predict transverse shrinkage and angular deformation. The model incorporates key influencing factors of line heating deformation: it employs a Convolutional Neural Network (CNN) to extract spatial features from deformation data, utilizes Double-Layer Long Short-Term Memory (LSTM) networks to model both temporal dependencies among multiple heating lines within a single plate and deformation correlations between multiple plates, and introduces an Attention Mechanism (AM) to enhance feature weights of critical time steps. Results demonstrate that the model achieves a Mean Absolute Error (MAE) of 0.0383 for angular deformation prediction and an R² coefficient of 0.96 for transverse shrinkage prediction, which effectively solves the spatiotemporal dynamic evolution prediction challenge in line heating deformation processes and provides technical support for realizing closed-loop intelligent control of "processing-prediction-feedback".