Study On The Key Issues Of Long-term Drift Calibration Of Breathting Analysis System

As the tradition way of medicine in China, Chinese medicine is broad and profound,which has far-reaching significance, it is the gem of Chinese traditional culture. In recent years, with the rapid development of science and technology, the modernization of the Chinese medicine, which is called “looking, smelling, and asking and pulse-taking” become more and more urgent. This dissertation mainly focuses on the research of the modernization of Chinese medicine "smell", that is, the detection of the concentration of the oral gas composition by the respiratory gas analyzer. Respiratory gas analyzer uses a variety of gas sensors to achieve the perception of gas, but over time, due to the sensor's own aging will lead to long-term drift, which is inevitable phenomenon. Therefore, it is necessary to correct the long-term drift of the sensor.Our research group independently developed a breathing gas analyzer, which is composed of eleven gas sensors, different sensors have different sensitive gases. In order to observe the rules of the sensor drift, firstly, we use 10 kinds of standard gas consisted of 4 kinds of calibration samples to carry out long-term tracking experiments. Then, Then the baseline difference method is used to deal with the initial response values of the sensor. Extracting the steady state characteristic value is used to reduce the complexity of the experiment. Through observation and analysis of response data of selected sensor corresponding to the sensitive gas, according to the drift law of sensor's sensitive gas, establishment the long term drift model in order to use in the latter part of the compensation.The design of a composition calibration method based on the principal component analysis method in this research. The algorithm gets the coefficient matrix by applying the principal component analysis to the data of a calibration sample collected by sensor. Then the test sample is projected onto the drift coefficient matrix to obtained the drift of the test sample, and by this way the test sample is calibrated and can be used to screen for the disease. After the calibration, a support vector machine classification algorithm is used to classify the data before and after the compensation, the sample classification after correction accuracy is improved, proves the necessity of compensation algorithm.Long term drift correction algorithm is proposed based on the multiplicative correction method. First, we observed the sensor drift law of the collected long term calibration data of the calibrated sample, then using smooth function to smooth the data samples and established the piecewise linear model for the data, and finally calibrated the data using the multiplicative correction method. In order to verify effectiveness of multiplicative correction method, a classification has been applied to the data before and after compensation using regularized logistic regression classification has been applied to the data before and after compensation using regularized logistic regression classification algorithm and K nearest neighbor classification algorithm.In short, the two correction algorithm proposed by this dissertation have a good performance on the compensation for the long-term drift of the sensor, which improved the screening accuracy of late stage samples collected by respiratory gas analyzer.