Coupled Hybrid Neural Network Models with Alarm Threshold Optimization for Early Warning of Abnormal Load Variations in Green Ammonia Production

Background

Green ammonia, as an excellent clean energy and hydrogen carrier, is crucial for integrating and utilizing renewable energy. However, frequent switching of production loads in the ammonia synthesis poses a significant increase in chemical safety risks, becoming a key technical challenge restricting large-scale industrialization of green ammonia. As a crucial component of safety management systems, accurately predicting and identifying abnormal conditions during load variation processes has become an urgent and critical issue to address.

Method

This study proposes an early warning method for abnormal conditions in green ammonia processes, combining deep learning with optimized alarm threshold strategies.Initially, based on process mechanisms and analysis of complex network relationship, determine the key variable with highest importance as the early warning variable. Subsequently, a hybrid LSTM-GRU model operating in parallel for time series prediction is proposed, with output results obtained through weighted allocation. Finally, the change rate in process parameters within the green ammonia process is introduced as a warning variable, optimized using ROC curves for threshold setting.

Significant findings

The proposed method was validated by applying it to a dynamic model of the green ammonia process constructed on the Honeywell UniSim^@ simulation platform. The results demonstrate that the model accurately captures the dynamic variations of the production process. The average R² value of the model prediction accuracy for load-varying processes exceeds 0.995, and the model exhibits the capability to extend the prediction horizon to 5 minutes. Furthermore, the optimized threshold enables precise diagnosis of abnormal conditions during load variations in the green ammonia process, achieving a detection rate of 99.36%. This provides an effective research tool and methodology for early safety warning in green ammonia production processes.