Application Of Convolution Neural Network In Dynamic Temperature Measurement

Temperature is an important parameter to characterize the working state of the system.There is a wide demand for accurate measurement of dynamic temperature in the industrial field.Multi-thermocouple method is one of the methods of dynamic temperature measurement.The thermocouple has thermal inertia,the dynamic response has hysteresis,and the dynamic temperature measurement has dynamic error.In order to reduce the dynamic error,this paper applies convolutional neural network to dynamic temperature measurement to improve the accuracy of dynamic temperature measurement.It is not easy for the measured medium to produce a dynamic temperature that changes regularly.Therefore,the step signal,pseudo random binary signal and amplitude-modulated pseudo random signal are used as the measured medium to simulate the temperature respectively,so as to obtain two ideal response of the thermocouples in the simulation environment,and the simulated measured values of two thermocouples under different signal-to-noise ratios are obtained by adding Gaussian white noise of different intensities.The simulated measured values are regarded as time series,which are modeled by long short-term memory network(LSTM),convolutional neural network(CNN)and temporal convolutional network(TCN)to reconstruct the measured medium temperature.The root mean square error(RMSE)and goodness of fit(R~2)were used to evaluate the dynamic temperature reconstruction of the three models.The calculation results show that under the set simulation conditions,compared with LSTM and CNN models,the RMSE between the dynamic temperature reconstruction of TCN model and the real temperature is smaller,and R~2 is larger,indicating that the dynamic temperature reconstruction accuracy of TCN model is higher.An experimental system for measuring the dynamic temperature of gas flow with dual-thermocouple is built,and the experimental values of two thermocouples excited by gas medium are obtained.The experimental measured values are regarded as time series,which are modeled by LSTM,CNN and TCN to reconstruct the real temperature.The dynamic temperature reconstruction of the three models was evaluated by RMSE and R~2.The experimental results also show that the dynamic temperature reconstruction accuracy of TCN model is higher than that of LSTM and CNN models.