Optimisation of liquefied natural gas production: genetic algorithm and custom-developed method

This study comprehensively analyses various optimisation techniques applied to Liquefied Natural Gas (LNG) production. Two datasets were used to assess the performance of these techniques, with a focus on improving LNG output. The results revealed that the genetic algorithm exhibited the highest average percentage improvement in the first dataset, achieving a 12% optimisation, followed closely by a custom-developed optimisation method at 11%. Bayesian optimisation showed an average of 4%, while gradient descent demonstrated the lowest optimisation with - 2%. Notably, the second dataset displayed even more significant improvements, with the custom optimisation algorithm leading at an average of 32%, surpassing the genetic optimization method's 30%. This study underscores the efficacy of the custom algorithm and its potential for enhancing LNG production, positioning it as a promising alternative to traditional optimisation approaches.


Introduction
The world is already making the switch to natural gas as a cheaper and cleaner energy source.It is mostly replacing coal as the most eco-friendly choice because it produces fewer carbon emissions (Mofid & Fetanat, 2019;Salehi, 2018;Wang, 2017).The amount of natural gas used is expected to rise by a large 40% between 2014 and 2040.(BP, 2017).Jackson, Eiksund, and Brodal's study from 2017 found that natural gas-powered plants made up 37% of fossil fuel energy in 2030, up from 30% in 2013.
Because it burns cleaner and releases fewer greenhouse gases, liquefied natural gas (LNG) is quickly becoming the world's main energy source.This trend has sped up since the recent energy crisis (Sang et al., 2020).Pipelines or liquefaction are the main ways that natural gas is moved.Energy companies often use liquefying natural gas for longdistance transport because pipeline restrictions, fixed transit routes, and long-term contracts make it hard to get pipeline gas (Lee et al., 2020).
Gases are liquefied, which turns them into liquids.LNG is made by cooling natural gas to -162 degrees Celsius at room temperature and pressure.Natural gas is much easier to transport when it is liquefied because it takes up only one-six hundredth as much space as when it is gaseous (Khalilpour & Karima, 2009).
To get these very low temperatures, you need refrigerants, and the way they heat up and cool down must be very similar to natural gas.Refrigerants, which are often found in air conditioning and refrigeration systems, are very important for keeping LNG at the low temperatures it needs to be stored and transported.
Because of this, how well this refrigeration process works is very important, since a bad system can cause less production.The industry needs refrigerants that are good for the environment and use little energy.As time has gone on, different types and methods of refrigerants have been used to make LNG.These include the turbo-expander process, the cascade process, and single/dual mixed refrigerant (SMR/DMR) technology.The main differences between these methods are their start-up and running costs, which depend on things like how much they can produce, how much equipment they need, and how much labour costs.
Mixed-refrigerant (MR) processes, on the other hand, make design and operation more difficult because there are more thermodynamic interactions.This makes it harder to manage and improve the process (Shukri, 2004).Which refrigerant to use depends on things like the temperature range you want, how easy it is to get, how much it costs, and what you know from past experience.For example, an olefins factory might have ethylene and propylene on hand, while a natural gas processing plant might have ethane and propane on hand.To keep things clean, it's important to use the right refrigerant.Halocarbons are often preferred because they don't catch fire.
The Propane Precooled Mixed Refrigerant (C3MR) system is a common way to cool things down these days.This method uses a propane refrigeration system to cool LNG to -35°C before it goes into a mixed refrigeration system that has methane, ethane, propane, and nitrogen (Bahadori et al., 2014).

Material and methods
To optimise the liquefied natural gas production of an industry, an artificial intelligence (AI) program was used.Specifically, the python programming language was the optimum and readily available software to be used.This study collected data comprising the LNG production, refrigerants, temperature, and pressure of the refrigeration processes.These data were processed in the software using four regression analysis models.

Material
The material used in this research include: An artificial intelligent (Python programming language) software, PI Processbook software 2015 version 3.6.2.271, PI datalinks, Visual studio (VS) code editor and Microsoft Excel 365.Python is an interpreted, high-level programming language that may be used for various projects.The principle behind its design prioritizes the readability of the code by heavily indenting it.The PI processbook and PI datalink add-in were basically used for data collection from the plant site.While the VS code editor is mainly an Integrated Development Environment (IDE) source code editor used to debug, highlight syntax and for coding of the GUI script.It is a userfriendly coding environment.

Process Optimisation Description
Figure 1 shows the sequential order or steps used to achieve the aim and objectives of this research.It depicts the schematic breakdown of the optimisation process using the artificial intelligence data driven approach.

Data Collection
The data collection for this work was done using PI Processbook 2015 software version 3.6.2.271 R2 and PI datalink.PI Processbook is an OSIsoft vendor software that enable users to retrieve real-time data from the PI system which is linked to a live process plant.The software application has the capability to create dynamical graphical display, trends from historical and real time data.To retrieve the data used for the work, the PI datalink was connected to the PI server and then to the liquefaction plant via several process control schemes as shown in Figure 2. The PI datalink is a Microsoft Excel add-in feature linked to the PI software.The sample data multiple value function of the PI datalink was used to retrieve about 10 years liquefaction unit data set at an hourly interval.

Discussion of Results
Tables 1 and 2 demonstrate, respectively, the average percentage improvement in the flow of 20 observations from the first and second datasets that each of the various optimisation techniques were able to accomplish.In Table 1, it can be observed that the genetic algorithm had the greatest optimisation, which resulted in a 12 % gain on average, followed by the custom optimisation method (11 %).The Bayesian method produced an average of 4%, while the gradient descent method produced the lowest percentage, which was -2 %.Figures 3 to 12 provide a visual representation of the optimisation findings, respectively.According to Table 2, the average optimisation result achieved by the custom optimisation algorithm was 32 %, which was higher than the average optimisation result achieved by the genetic optimisation method, which was 30 %. Figure 3 to Figure 12 provide a graphical representation of the performance of each optimisation technique, respectively.

Conclusion
The Genetic algorithm achieved the best results, with an average improvement of 12 % in optimisation, followed by the custom-developed optimisation method that we produced, which achieved 11 %.The Bayesian method produced an average of 4 %, while the gradient descent method produced the lowest percentage, which was -2 %.We found that our built bespoke optimisation method had the greatest average optimisation result of 32 % for the second validation LNG data set.This was followed by genetic optimisation, which had a result of 30 % for average optimization.

Figure 1
Figure 1 Process description of Artificial Intelligence optimisation

Figure 2 PI
Figure 2 PI System data Collection scheme 2.4.Optimisation Algorithms Tested Three optimisation algorithms were used in the process and these are'  Bayesian Optimisation  Genetic optimisation  Gradient Descent Optimisation

Figure 4 Figure 5 Figure 6
Figure 4 Genetic Optimisation result on First Dataset

Figure 7
Figure 7 Bayesian Optimisation result on First dataset

Figure 8 Figure 9 Figure 10 Figure 12
Figure 8 Overall Optimisation result on Second dataset

Table 1
Average Percentage Increase on First 20 observations in the First Sample set Figure 3 Overall Optimisation Result on First Dataset

Table 2
The Average Percentage Increase on the First 20 observations in the second Sample set