Time Series Analysis Of Dynamic Temperature

Different temperature sensors show different indicated temperatures when excited by the same dynamic temperature source.In order to quantify the difference and further use the difference to verify the temperature sensor,considering that the dynamic temperature is sequential under time uniform sampling,this paper proposes treating dynamic temperature over a period of time as a dynamic temperature time series and conduct research and analysis.The main research contents are as follows:(1)Research on dynamic temperature measurement technology.The method of dynamic temperature measurement was discussed.The dynamic characteristics of thermocouple were studied and the dynamic excitation signal was analyzed.The similarity measures of dynamic temperature time series was studied,and the principles of Euclidean distance and discrete Fréchet distance were expounded.Finally,the classification model of dynamic temperature time series was proposed,and the principles of k nearest neighbor model(classification model based on similarity measures)and long short term memory-fully convolution network model(deep learning model)were summarized and analyzed.(2)Design dynamic temperature measurement simulation experiment.The dynamic temperature measurement simulation datasets were constructed.Based on the above datasets,the similarity and classification model of dynamic temperature time series were simulated and analyzed.The simulation results indicate that compared with Euclidean distance,discrete Fréchet distance has strong robustness and high sensitivity,and is more suitable for indicating the similarity of dynamic temperature time series.Compared with the k nearest neighbor model,the long short term memory-fully convolution network model has strong anti-noise ability and is more suitable for the measurement environment with low signal-to-noise ratio.(3)Design the dynamic temperature measurement experiment of airflow.The dynamic temperature measurement experimental datasets of airflow were constructed.Based on the above datasets,the similarity and classification model of airflow dynamic temperature time series were experimentally analyzed.According to the simulation results,the k nearest neighbor model based on discrete Fréchet distance and the long short term memory-fully convolution network model were established.The experimental results indicate that the discrete Fréchet distance is more suitable to indicate the similarity of airflow dynamic temperature time series.The long short term memory-fully convolution network model has better classification accuracy and effect.