(371u) AI-Based Decision Making of Steam Methane Reforming Operation Strategies from Fluctuated Biogas Production.

Hydrogen production using biogas is a representative process for achieving carbon neutrality and sustainability. As efforts to reduce dependency on fossil fuels and decrease greenhouse gas emissions accelerate globally, technologies for producing hydrogen from renewable energy sources are gaining significant attention. However, biogas has various compositions that cause uncertainty within the follow-up process due to the features of the biochemical reactions. We propose a biomass-to-hydrogen production model based on data from real biogas plant to minimize the uncertainty. This study involves a series of process that utilizes biogas generated from waste to produce hydrogen through steam methane reforming process. First, we generated prediction model of biogas digestion process by analyzing the relationships between plant variables and biogas composition. Then, the data-driven model of hydrogen production process from biogas was developed to evaluate the process in terms of economics and environment. Here, we performed sensitivity analysis under four different techno-economic and environmental evaluation criteria: maximum hydrogen production (MHP), unit production cost (UPC), net CO₂ equivalent emissions (NCE), and energy efficiency (EE). Finally, we can suggest the optimal operating strategies of the process which can respond the vibration of biogas composition and market disturbance proactively.

Biogas is obtained by collecting gas (e.g., methane, CO2) generated during the decomposition process of organic matter and storing them as energy. Biomass to biogas process is important in terms of reducing environmental pollution and recycling energy resources, and biogas is utilized in various fields including as fuel for vehicles and in power generation. Particularly, biogas which has high methane content (50% ~ 80%) is a promising raw material for hydrogen production. Hydrogen is a valuable product that does not emit greenhouse gas when used and has a higher energy value than fossil fuel and biogas can be used immediately in the biogas to hydrogen process. There are several processes for producing hydrogen using biogas such as steam methane reforming (SMR), dry reforming and auto thermal reforming. Among these processes, SMR process is the most well-established, offering a proven pathway for converting methane-rich biogas into hydrogen. However, the biogas utilization has several limitations. First, the mechanism of biological reactions is complex, making it difficult to fully understand the enzyme action and the role of the microbial community. Second, production of biogas depends greatly on the type and composition of raw materials and external environmental conditions (e.g., region and temperature). This causes uncertainty in the subsequent process and makes it difficult to optimization. Accordingly, identification of the composition is an important factor in the operation of the process, and then it is necessary to evaluate the effect of the biogas composition on the result of the process. Thus, we developed a solution model that can predict the biogas composition generated by biomass and can contribute to the optimization of the hydrogen production process. This research can provide insight into operation strategies for biomass to hydrogen process.

The purpose of this study is to provide an insight to minimize the uncertainty in the hydrogen production process using biogas and to operate plant optimally according to the composition. We obtained three years of actual data (2020’~2022’) from the biogas plant in Yeongcheon-si, Republic of Korea. Biomass (food wastewater, food waste, livestock manure) irregularly generated in nearby cities was used as a raw material. Organic acids are produced from input materials through oxidation, after hydrolysis in an oxidation tank. Finally, in an anaerobic digestion tank biogas is generated from the organic acid transferred from the oxidizing tank by digestion process. All processes except the input of raw materials were operated with continuous batch type. The initial data frame includes the input amount of each raw material, the chemical characteristics of the process step, and the composition of biogas. We added week’s number and temperature data of Yeongcheon-Si to additionally consider the time series and external environmental conditions. Pearson correlation analysis was performed to quantify the relationship between each data, and the reaction time delay caused by the biochemical reaction was found through time lags shift. By using long short-term memory (LSTM) which is a machine learning model with strong performance for time series analysis, two models for each digestion step were developed. The models trained time series and variables of each process. Then, we integrated the models to predict the composition of biogas based on the raw materials and time points. The prediction accuracy was evaluated in R2-score and mean absolute percentage error (MAPE).

Subsequent hydrogen production processes from biogas involve reaction and separation, and aspen plus V12.1 was used to obtain chemical reaction data including mass and energy balance. In the first reactor, the syngas is synthesized through SMR reaction in the fed biogas. CO in the syngas is then converted to hydrogen in the water gas shift (WGS) process. A simple flash tank is used to separate H₂O from the gas stream from the WGS, before the amine process where most of CO₂ is captured and separated from the mainstream. Finally, high purity hydrogen is obtained through the pressure swing adsorption (PSA) process. Biogas composition, reformer operating temperature, and feedstock ratio were set as variables in the data collection process. In the process simulation step, a comprehensive database of input and output information was generated, including energy consumption and costs for each unit process. We made a huge number of operational data by connecting various concentrations of biogas to each sequential process. A knowledge-based screening including realistic biogas compositional ranges and reasonable operating conditions of the reforming process was applied to ensure valid data selection. Additionally, we set four different criteria, which are calculated by size and cost information as well as mass and energy balance. The criteria include maximum hydrogen production (MHP), unit production cost (UPC), net CO₂ equivalent emissions (NCE), and energy efficiency (EE). As a result, we utilized them to find the optimal economic and environmental pathway for each situation.

This study presents the overall process and results from biomass to hydrogen production via biogas. The machine learning model that learned about real data and time series in biogas plant was developed. We could predict the composition of biogas by using model generated. We set criteria and performed sensitivity analysis about SMR processes. The optimal economic and environmental hydrogen production pathway was found for each situation through the data-driven model. As a result, adjustment of operation conditions according to biogas composition shows better economic and environmental outcomes compare to operating the process under fixed conditions. As a result, our models are expected to help handle the uncertainty of biogas utilization and provide operation strategies for the hydrogen production process.