Development of a LSTM based Model for the Optimal Operation of an Ambient Air Vaporizer of LNG

Aiming to enhance the operational efficiency and safety of Ambient Air Vaporizers (AAVs), this study utilizes real operational data collected over a year from an active natural gas production facility to develop a foundational model that improves operations currently reliant on onsite experience and intuition. Specifically, the data revealed strong seasonality in production and energy costs. Through regression and variance analyses, we found that the outlet temperature of natural gas (NG) is significantly influenced by current atmospheric conditions, fan speed, and existing NG outlet temperatures, informing our model development. We tested and compared Long Short-Term Memory (LSTM) and Deep Neural Network (DNN) predictive models, with the LSTM model showing superior performance, indicating its potential as a foundation for real-time fan speed control. Despite data collection limitations due to high safety operations and restricted fan speeds, we developed highly performant .0models by testing various parameters. The AI model provides accurate predictions of discharge temperature based on fan operating conditions, serving as a valuable tool for facility operators and essential for developing automated models through further reinforcement learning.