【作者】 黄超；

【导师】 戴一阳；

【作者基本信息】 四川大学 ， 化学工程， 2023， 硕士

【摘要】 在能源危机与环境危机的双重影响下,新能源发电开始在国际国内大规模发展,新能源所产生的电能具有间歇性、随机性、波动性的特点,随新能源发电装机规模逐步扩大,电力全部进行电网消纳会对电网产生极大的冲击。因此新能源发电系统配备就地消纳电解制氢,可以解决偏远地区、电网架构薄弱地区新能源电力外送的问题。电解制氢为新能源系统与化工领域深度拓展和应用提供了技术支撑,碱性电解水作为目前成熟度最高,大规模工业应用前景最大的电解制氢技术,其技术瓶颈必须突破。因此本论文以碱性电解水系统为研究对象,基于过程模拟技术建模分析,并结合机器学习技术对碱性电解水系统关键指标预测,旨在从系统的角度对碱性电解水工艺进行模拟分析,评价系统的工作效率、操作弹性、安全边界。总体工作包含以下两部分。1)针对碱性电解水系统进行机理建模,建模过程分为两个步骤,首先以Aspen Custom Moduler(ACM)工具搭建电解槽单元模型,该电解槽模型考虑了各种因素对杂质传质过程、极化曲线的影响,是基于机理计算的电解槽模型。再将ACM电解槽模型集成到HYSYS系统模型中。整个电解系统模型搭建完毕后,采用清华四川能源互联网研究院的2 N/m~3电解系统实验数据对电解模型修正,修正后的HYSYS模型运行结果与真实实验装置运行数据相符。基于搭建的HYSYS模型对系统的效率、气体杂质进行分析,可以对实际操作运行过程进行指导。2)基于HYSYS模型建立了数字孪生工厂,以HYSYS模型运行数据为数据集,选取了12个参数输入,采用PCA-ANN网络对系统的氧中氢浓度(Hydrogen to oxygen,HTO)进行预测。HTO是系统安全运行的重要指标,当HTO高于2%时,系统强制停机,在新能源场景下,制氢装置弹性操作过程中,基于数据驱动的预测可以保证过程安全。最终PCA-ANN模型实验表明预测结果具有准确性和可行性,模型预测的判定系数(R Squared,R~2)达到了0.9508,均方差(Mean squared error,MSE)、平均绝对误差(Mean Absolute Error,MAE)值均在可接受范围内。同时进一步验证了工业数据清洗步骤对数据驱动模型的影响,数据集的质量决定了模型预测的上限。更多还原

【Abstract】 Under the double influence of energy crisis and environmental crisis,new energy power generation has begun to develop on a large scale both at home and abroad.The electric energy generated by new energy is characterized by intermittence,randomness and volatility.With the gradual expansion of the installed scale of new energy power generation,all the electricity consumed by the grid will have a great impact on the grid.Therefore,the new energy power generation system is equipped with local absorption electrolytic hydrogen production,which can solve the problem of new energy power transmission in remote areas and areas with weak grid architecture.Electrolytic hydrogen production provides technical support for the deep expansion and application of new energy systems and chemical industry.Alkaline electrolytic water is the most mature electrolytic hydrogen production technology with the largest prospect of large-scale industrial application,and its technical bottleneck must be broken through.Therefore,this paper takes alkaline electrolytic water system as the research object,modeling and analysis based on process simulation technology,combined with machine learning technology to predict the key indicators of alkaline electrolytic water system,aiming to simulate and analyze the alkaline electrolytic water process from the perspective of the system,and evaluate the working efficiency,operating elasticity and safety boundary of the system.The overall work consists of the following two parts.1)The mechanism modeling of alkaline electrolytic water system is divided into two steps.First,Aspen Custom Moduler(ACM)tool is used to build the cell model.The cell model takes into account the influence of various factors on the mass transfer process of impurities and polarization curve,which is an electrolytic cell model based on mechanism calculation.Then the ACM cell model is integrated into the HYSYS system model.After the entire electrolytic system model was built,experimental data of 2 N/m~3 electrolytic system from Sichuan Energy Internet Research Institute of Tsinghua University was used to modify the electrolytic model.The operation results of the modified HYSYS model were consistent with the actual experimental device.Based on the established HYSYS model,the efficiency and gas impurities of the system are analyzed,which can guide the actual operation process.2)A digital twin factory was established based on the HYSYS model.The operating data of the HYSYS model was used as the data set,12 parameters were input,and the Hydrogen to oxygen(HTO)concentration of the system was predicted by PCA-ANN network.HTO is an important indicator for safe operation of the system.When HTO is higher than 2%,the system will be forcibly shut down.In the new energy scenario,data-driven prediction can ensure the process safety during the elastic operation of hydrogen production device.The final PCA-ANN model experiment showed that the prediction results were accurate and feasible,and the decision coefficient(R Squared,R~2)of the model prediction reached 0.9508,Mean squared error(MSE),Mean Absolute Error(Mean Absolute Error,MAE values were within the acceptable range.At the same time,the influence of industrial data cleaning steps on the data-driven model is further verified.The quality of data sets determines the upper limit of model prediction..更多还原