Dynamic temperature measurement is important in industry. The traditional method with a dual-thermocouple is easily disturbed by noise, so it is only suitable for a high signal-to-noise ratio (SNR). To obtain more accurate dynamic temperature, a dual-thermocouple method based on a one-dimensional convolutional neural network is proposed. This method extracts and fuses the temperature signals of two thermocouples with the same material and but different diameters, thereby weakening noise. The method was simulated with a numerical simulation dataset. Evaluations under different SNRs reveal that the method has a higher fit percentage (FIT), lower root mean square error (RMSE), and lower dynamic error. The algorithm was also applied to an experimental dataset. The experimental results also confirm that this measurement method is better than the traditional methods, with a FIT of 96.49% and RMSE of 0.0149. The average dynamic error is 0.0078 and the standard deviation is 0.0127.

动态温度测量在工业中很重要。双热电偶的传统方法容易受到噪声的干扰，因此只适用于高信噪比（SNR）。为了获得更准确的动态温度，提出了一种基于一维卷积神经网络的双热电偶方法。该方法提取并融合两个材质相同但直径不同的热电偶的温度信号，从而减弱噪声。该方法使用数值模拟数据集进行模拟。在不同 SNR 下的评估表明，该方法具有更高的拟合百分比 (FIT)、更低的均方根误差 (RMSE) 和更低的动态误差。该算法还应用于实验数据集。实验结果也证实该测量方法优于传统方法，FIT 为 96.49%，RMSE 为 0.0149。平均动态误差为 0.0078，标准差为 0.0127。