摘要
为了提高动态温度测量的精度,提出了基于卷积网络的双热电偶动态温度测量方法。采用不同强度的高斯噪声,仿真获得两支时间常数互异的热电偶在高低温双温度源激励下的理论测量值。将理论测量值视为时间序列,采用卷积网络中的卷积神经网络和时间卷积网络对其进行建模,重建真实温度。计算结果表明,基于时间卷积网络的测量方法的均方根误差(RMSE)更小、拟合优度(R^(2))更大。以两支K型镍铬镍硅热电偶为对象,给出了一个算例,实验结果也表明,基于时间卷积网络的测量方法优于基于卷积神经网络的,该测量方法的RMSE为0.0284,R^(2)为0.9940。
In order to improve the accuracy of dynamic temperature measurement, a dual-thermocouple dynamic temperature measurement method based on convolutional network is proposed.Using different intensity of Gaussian noise, the theoretical measurement data sets of two thermocouples with different time constants excited by high and low temperature dual temperature sources are obtained.The theoretical measured values are regard as time series, which are modeled by convolutional neural network and temporal convolutional network in convolutional network to reconstruct the true temperature. The results show that the measurement method based on temporal convolutional network has a lower root mean square error(RMSE)and higher goodness of fit(R^(2)). Two K-type nickel chromium nickel silicon thermocouple is used as a case study. The experimental results show that the measurement method based on temporal convolutional network is better than that based on convolutional neural network, with RMSE of 0.028 4 and R^(2)of 0.994 0.
作者
虞思思
李文军
崔志文
金敏俊
YU Sisi;LI Wenjun;CUI Zhiwen;JIN Minjun(College of Metrological Technology and Engineering,China Jiliang University,Hangzhou Zhejiang 310018,China)
出处
《传感技术学报》
CAS
CSCD
北大核心
2022年第1期22-29,共8页
Chinese Journal of Sensors and Actuators
关键词
动态温度测量
时间序列
时间卷积网络
双热电偶
卷积神经网络
dynamic temperature measurement
time series
temporal convolutional network
dual-thermocouple
convolutional neural network