(182a) New Generation LNG Production Improvement Using AI Technology

Over the past few years, there have been many application references related to the utilization of Artificial Intelligence (AI) technology in various industries. However, the Oil & Gas industry is behind in this movement towards “AI” technology because of the shortage of published best practices. Nevertheless, JGC foresees that the following trends in the industry enables us to adapt the technology into the LNG facilities.

• Availability of Big Data in historian
• Spreading of “Cloud Platform” technology in Liquefaction facilities for easily accessible historian data
• Recent maturity of AI models available in the industry
• Revolutionary improvement of computing hardware

This paper covers JGC’s methodology to increase annual LNG production for the operating liquefaction facilities with the implementation of the AI technology.

2. Development and Method

To increase annual LNG production by optimizing the operating parameters of the existing LNG facilities, following AI technologies have been applied to develop the models:

• Recurrent Neural Network to develop a predictive model for a future condition prediction
• Partial Least Square Method to develop plant soft-sensors used as control inputs
• Deep Stacking Network to develop a mathematical plant model as an environment of reinforcement learning
• Deep Q Network to develop an optimal policy for liquefaction process control

Two methods are considered for LNG production increase:

• Online Method: Real-time Predictive model input feedback to DCS and APC (Advanced Process Control) for an automatic production increase control
• Offline Method: Regular diagnosis and improvement recommendations (digital twin process simulation model/AI model, and plant big data analysis) for an overall plant capacity enhancement

3. Results

The results of the development are as follows:

• Methodology of initial operational diagnosis has been established to quantify the benefit of AI technology for LNG production improvement, which covers both 1) hot air recirculation issues for air-cooled type liquefaction plant and 2) mixed refrigerant circuit operational optimization. The paper will introduce the application examples of the AI technology for LNG production improvement with potential benefit (i.e. 2～5 % of annual LNG production increase). The paper also explains why these two issues were focused for LNG production improvement from our process engineer’s observations.
• Appropriate models have been selected from varieties of the AI models in the industry. The paper will include how the models are analyzed and selected, together with who actually took the analysis (how to collaborate data scientist and process engineers for LNG plant design).

4. Observations

The industry should keep in mind that the use of AI is just one methodology to improve the core competence of the company and to improve its corporate value. However, most company often makes the integration of AI into an objective without considering how it can benefit its business. Without defining a clear issue, the use of AI will most likely to result unsuccessfully.

JGC believes that the main reason why the development described in this paper was successful is because more efforts were made to define the issues faced in the LNG plant operation than actually developing the AI model.

Furthermore, our past experiences reveal that the involvement of process engineers as the applicants and/or the developers is necessary because process engineers can

• understand important operating data/parameters
• define target and explanatory parameters for model development
• evaluate the model
• learn AI model development as AI technology is becoming easily accessible

With these lessons learnt, JGC believes that the incorporation of domain knowledge owner into the development team is a major factor to make AI development project into a success. The paper will summarize lessons learnt from the model development how the insights of process engineers are collaborated with examples.