Phase transition prediction of dynamical systems based on deep learning

Many dynamical systems sometimes occur phase transitions, abruptly shifting from one state to another. Predicting potential phase transitions from externally observed data is a significant challenge. In order to make accurate prediction of phase transitions, an AT-TCN-LSTM model is proposed in this paper. The AT-TCN-LSTM model combines Temporal Convolutional Neural Network (TCN) and Long Short-Term Memory (LSTM) to effectively capture temporal features in observed system data. The TCN block captures both short-term and long-term dependencies, enabling the model to understand causality in time series data. Additionally, an attention mechanism is introduced to enhance the utilization of the captured features. Experimental results confirm the superiority of the proposed model, demonstrating better performance compared to existing methods in predicting phase transitions.