Abstract:

To address the challenges of low gasification efficiency and poor hydrogen selectivity in the production of green hydrogen from biomass gasification, a decoupling process involving pyrolysis followed by chemical looping reforming of volatiles for hydrogen production is proposed. In the theoretical analysis of the above process, it was found that the complex relationship between the yield and composition of pyrolysis volatiles and the properties of biomass and pyrolysis operating conditions is difficult to be accurately associated with the traditional modeling method, which restricts the precise analysis and regulation of the above process. Therefore, this paper establishes a neural network model for the product distribution of the biomass fast pyrolysis process using machine learning methods and determines the optimal pyrolysis conditions using the particle swarm optimization algorithm. The goal is to maximize the hydrogen atom ratio and heating value of the pyrolysis volatiles while minimizing the oxygen atom ratio. Subsequently, the process of hydrogen production from chemical looping reforming of volatiles was analyzed and optimized through process simulation. The results show that the established neural network model can accurately predict the yield of the three-phase pyrolysis products, the detailed composition of the pyrolysis gas, the elemental distribution of the pyrolysis oil, and the higher heating value, etc. The average coefficient of determination of the predictions is 0.821, and the average root mean square error is 2.00, in the test set of the above output parameters. After optimization, the pyrolysis volatiles yield for herbaceous biomass (wheat straw, corn stover) and woody biomass (ficus, pine wood) ranged from 64.49% to 78.62%, with a hydrogen atom ratio between 3.77% and 4.39%. Under optimal conditions at a reforming temperature of 700 ℃ and a steam-to-biomass mass ratio of 0.71 to 0.88, wheat straw showed the highest hydrogen yield and CO₂ negative emission capability, with values of 0.60 m³/kg and -1.74 kg /m³, respectively. Using chemical looping reforming of biomass volatiles for hydrogen production, the hydrogen yield from the four types of biomass increased by 61%, 35%, 16%, and 34% respectively compared to conventional gasification. The research results provide effective foundational support for the production of green hydrogen from biomass.

Key words: machine learning, chemical looping reforming, gasification, biomass, optimization

摘要：

针对生物质气化制备绿氢过程中气化效率低、氢气选择性差的挑战，提出了一种热解串联挥发分化学链重整制氢的解耦工艺。在对上述过程理论分析时发现，热解挥发分的产量和组成与生物质性质和热解操作条件之间的复杂关系难以通过传统模型化方法被准确关联，从而制约了上述工艺的精确分析调控。因此本文基于机器学习方法建立了生物质快速热解过程的产物分布预测的神经网络模型，并结合粒子群优化算法确定最佳热解条件，使热解挥发分的氢原子比和高位热值最大化，氧原子比最小化。随后，基于流程模拟对挥发分化学链重整制氢工艺进行了分析和优化。研究结果显示，所建立的神经网络模型能够准确预测热解三相产物的产率、热解气的详细组成、热解油的元素分布及高位热值等。在上述输出参数的综合测试集中，模型的平均决定系数为0.821，平均均方根误差为2.00。优化后，草本生物质（小麦秸秆、玉米秸秆）和木本生物质（榕树、松木）的热解挥发分产率为64.49%~78.62%，氢原子占比在3.77%~4.39%之间。在重整温度700 ^oC，蒸汽/生物质质量比0.71~0.88的优化工况下，小麦秸秆的氢气产量和CO₂负排放能力最高，分别为0.60 m³/kg与-1.74 kg /m³。采用生物质挥发分化学链重整制氢工艺，四种生物质的氢气产量相较常规气化分别增加了61%、35%、16%和34%。研究结果为生物质制备绿氢提供了有效的基础支撑。

关键词: 机器学习, 化学链重整, 气化, 生物质, 优化