基于Transformer的生物质流化床快速热解瞬时流率预测  

Predicting Instantaneous Mass Flow Rates of Biomass Fast Pyrolysis in a Fluidized Bed by Transformer

作　　者：王若晨 钟汉斌[1,2] Wang Ruochen;Zhong Hanbin(Xi'an Key Laboratory of Low-carbon Utilization for High-carbon Resources,Xi'an Shiyou University,Shaanxi,710065;Shaanxi Engineering Research Center of Green Low-carbon Energy Materials and Processes,Xi'an Shiyou University,Shaanxi,710065)

机构地区：[1]西安石油大学西安市高碳资源低碳化利用重点实验室,陕西710065 [2]西安石油大学陕西省绿色低碳能源材料与过程工程技术研究中心,陕西710065

出　　处：《当代化工研究》2024年第22期49-51,共3页Modern Chemical Research

基　　金：陕西省自然科学基础研究计划资助项目“低能耗分区控制变径流化床大颗粒生物质快速热解过程基础研究”(项目编号2023-JC-YB-119)。

摘　　要：传统的生物质流化床快速热解实验需要投入大量的时间和经济成本,因此,CFD模拟已经成为研究快速热解的主要工具。但大型反应器或宽操作条件下的CFD模拟仍需要大量的计算资源,所以为了进一步节省时间和经济成本,使用PaddleTS的Transformer深度学习模型对CFD模拟中的生物质流化床快速热解瞬时流率数据进行了训练和预测。首先,将收集到的CFD数据输入Transformer模型中进行训练。为了优化模型性能,对多个参数进行了调整,包括学习率、输入输出长度、向量维度、多头注意力头数、编码器和解码器层数等。为了进一步提升模型的泛化能力,还使用了L2正则化以降低过拟合风险。最终所获得的模型预测结果较为理想,表明了Transformer模型在处理生物质流化床快速热解瞬时流率预测方面的有效性和潜力。Traditional biomass fluidized bed fast pyrolysis experiments require significant time and economic costs,and CFD simulation has become the main tool for studying fast pyrolysis.However,CFD simulation still demands substantial computational resources for simulating large reactor or under wider operating conditions.To further save time and economic costs,the Transformer model of deep learning in PaddleTS was employed to train and predict the instantaneous mass flow rates collected from CFD simulations of biomass fast pyrolysis in a fluidized bed.First,the collected CFD data was input into the Transformer model for training.Multiple parameters,including learning rate,input and output lengths,vector dimensions,the number of attention heads,and the number of encoder and decoder layers,were adjusted to optimize model performance.To further enhance the model's generalization ability,L2 regularization was used to reduce the risk of overfitting.The final model's prediction results were satisfactory,indicating the effectiveness and potential of the Transformer model in predicting instantaneous flow rates in biomass fluidized bed fast pyrolysis.

关 键 词：CFD 深度学习 TRANSFORMER 快速热解 瞬时流率 

分 类 号：TQ9[化学工程]