PTQ Q3 2024 Issue

Model development approach After completing the necessary tests concerning the S/C ratio, the initial objective was to construct an analytical model capable of forecasting the required natural gas consumption at a given S/C value to meet H 2 demand. Subsequent efforts focused on developing analytical models for fuel gas consumption, steam production, and methane slip. In the data analysis phase, the primary focus was on understanding the influence of the S/C parameter on energy consumption and production levels. To capture this relationship, a search for suitable general models com- menced, employing tree-based and regression algorithms to ensure a comprehensive understanding of the S/C ratio’s impact (see Figure 2 ). Natural gas consumption model The constructed natural gas consumption model was intended to estimate the consumption of natural gas contin- gent upon the H 2 requirements and the S/C ratio. While the S/C ratio was initially posited as an independent variable in the model, its direct correlation with the target variable could have been more robust in the dataset than theoretically expected. Consequently, the model was refined to reflect the S/C ratio’s effect through the intermediary of methane slip – a variable that showed a more pronounced relationship. All of the analytical model’s development involved scru- tinising data from various periods across 2022 and 2023, with validations of different model iterations ensuring robustness. Key parameters incorporated included the hydrogen requirement, the molecular weight of natural gas, reformer inlet temperature, and S/C ratio. Collaborative reviews with the process department yielded a consensus on the logical coherence of the variables exhibiting high correlations, indicating their relevance and impact within the model framework. The detailed heatmap in Figure 3 visually represents these relationships while maintaining the confidentiality of the specific model dynamics. During the model refinement phase, a nuanced

Steam export

Natural gas

Hydrogen

MTS reactor

Steam reformer

PSA

Steam

Fuel gas

Purge gas

Utilities

Figure 1 Hydrogen generation unit boundaries

of steam so that carbon will eventually be removed by the reverse CO reduction reaction (1.4). A low steam-to-carbon ratio can lead to carbon deposition and, thus, long-term cat- alyst damage. Reformer temperature and steam-to-carbon ratio will affect the final gas composition of the reformer effluent. Project objectives The cost optimisation formula for this project incorporates the sum of natural gas and fuel gas consumption costs, deducts the profit from steam production, and divides the result by the hydrogen demand (see Figure 1 ). To achieve this, the development of analytic models was essential to accurately predict the consumption and production of nat- ural gas, fuel gas, and steam, the key variables influencing the operational cost in the HGU. A controlled test was initiated within the HGU to eluci- date the impact of varying steam-to-carbon (S/C) ratios. The S/C value was systematically altered to several prede- fined levels over designated periods during this test. The primary objective was to observe the consequent effects on the unit’s performance. Additionally, this exercise was instrumental in generating a clean dataset that would later be utilised in the modelling phase to ensure robustness and accuracy of the predictive analytics.

Prediction models

Optimisation layer

Outputs

Infrastructure

Reporting An e-mail is shared with stakeholders hourly, containing current feeds, capacities, optimum S/C value and optimum consumptions. Hourly results is also written to several PHD tags to track easily. Monitoring the results and taking proactive actions.

50+ PHD tags

Natural gas feed model

Iterate models on the range of S/C values with a chosen step size

PostgreSQL

Methane slip model

Fuel gas consumption model

Airow

Steam export model

Finding optimum S/C which gives minimum operational cost

AI based prediction models built to get predictions while iteratin S/C values

Big data from sensors and user inputs are consolidated

Figure 2 Analytical infrastructure

PTQ Q3 2024