基于多波段融合网络的相关k分布预测模型  

A Correlated k-distribution Model Based on Multi-Band Fusion Network

作　　者：汪迁文 吴嘉雯 张彪[1] 许传龙[1] 李健[1] WANG Qianwen;WU Jiawen;ZHANG Biao;XU Chuanlong;LI Jian(National Engineering Research Center of Power Generation Control and Safety,School of Energy and Environment,Southeast University,Nanjing 210096,China)

机构地区：[1]东南大学大型发电装备安全运行与智能测控国家工程研究中心,南京210096

出　　处：《工程热物理学报》2025年第4期1294-1300,共7页Journal of Engineering Thermophysics

基　　金：国家自然科学基金(No.52176167);江苏省自然科学基金(No.BK20201279)。

摘　　要：提出了一种多波段融合神经网络模型用于分波段式地预测CO、CO_(2)和水蒸气混合气的气体辐射特性。该模型可以将目标波段拆分成若干个连续的子波段,将在每个子波段上训练得到的单张网络融合成一个紧凑的大模型,从而实现所有子波段上混合气的吸收系数相关k分布的灵活快速预测。结果表明,此模型在测试集上的最大对称平均绝对百分误差(Symmetric Mean Absolute Percentage Error,SMAPE)小于1.83%,在40个连续且步为0.1μm的子波长段上的预测用时约为1.3 s。A multi-band fusion neural network model was proposed for predicting CO in a seg-mented manner,focusing on the gas radiation characteristics of CO_(2) and water vapor mixtures.This model divides the target band into several continuous sub-bands and integrates the individual networks trained on each sub-band into a single,compact model.This approach allows for flexible and rapid prediction of the absorption coefficient correlation(k)distribution across all sub-bands.The results demonstrate that the maximum Symmetric Mean Absolute Percentage Error(SMAPE)of this model on the test set is less than 1.83%,and the prediction time for 40 consecutive sub-wavelength bands,with a step of 0.1μm,is approximately 1.3 seconds.

关 键 词：神经网络 辐射换热 气体辐射特性 辐射传递模型 

分 类 号：TK123[动力工程及工程热物理—工程热物理] TN21[动力工程及工程热物理—热能工程]