基于深度神经网络的SOFC电堆温度场建模  

MODELLING OF SOFC STACK TEMPERATURE FIELD BASED ON DEEP NEURAL NETWORK

作　　者：武鑫[1] 吴万哲 白浩 王祺 熊星宇 Wu Xin;Wu Wanzhe;Bai Hao;Wang Qi;Xiong Xingyu(School of Energy Power and Mechanical Engineering,North China Electric Power University,Beijing 102206,China)

机构地区：[1]华北电力大学能源动力与机械工程学院,北京102206

出　　处：《太阳能学报》2023年第7期55-60,共6页Acta Energiae Solaris Sinica

基　　金：国家重点研发计划(2017YFB0601900);中央高校基本科研业务费项目(2019MS018)。

摘　　要：基于石英光纤温度传感器和直角坐标机械手,设计并搭建一种SOFC电堆温度场测量系统。然后应用上述系统测量模拟电堆的阴极气道温度数据。根据采集的数据,基于深度神经网络方法建立模拟电堆温度场模型,并与基于支持向量机方法的电堆温度场模型进行对比。结果显示:深度神经网络电堆温度场模型的训练时间更短,预测精度更高,其平均绝对误差和均方根误差分别为支持向量机电堆温度场模型的45.2%和47.4%,更有利于该文中电堆温度场建模。Based on the quartz fiber temperature sensor and the Cartesian coordinate manipulator,this paper designs and builds the SOFC stack temperature field measurement system.Then,the temperature data of the cathode port in the emulation SOFC stack are measured through the built measurement system.Based on the collected data,the emulation stack temperature field model is established based on deep neural network method,and compared with the stack temperature field model based on support vector machines(SVM).The results show that the stack temperature field model with deep neural network owns shorter training time and higher prediction accuracy.Its mean absolute prediction error and root mean square error are 45.2%and 47.4%respectively of the stack temperature field model of support vector machines,which is more convenient for the application of SOFC stack temperature field modelling in this paper.

关 键 词：固体氧化物燃料电池 深度神经网络 支持向量机 温度传感器 电堆温度场测量 

分 类 号：TK91[动力工程及工程热物理]