Analysis And Research Of Odorous Gas Chromatography Data Based On Machine Learning

Odorous pollution has seriously affected the quality of life of people.Gas chromatography analyzer is a common instrument for odorous gas detection,but highend products have long been monopolized by Europe,the United States,Japan and other countries in terms of technology and market,and domestic products still have some shortcomings in detection accuracy,intelligence,etc.,and there is still a lot of room for development.In this paper,the classification algorithm and overlapping peak resolution algorithm of gas chromatography data are mainly studied.Aiming at the problems of unblanced distribution of computing power and poor resolution of overlapping peaks of gas chromatography analyzers,a solution is proposed to classify peak shapes first,and then design a special analysis algorithm for each peak shape.In this paper,the classification algorithm of malodorous gas chromatography signal is first studied,and the feature points are located by spectral peak identification,and the feature parameters are calculated according to the feature points as the classification basis,and three machine learning algorithms such as K Nearest Neighbor Algorithm(KNN),Support Vector Machine(SVM)and Random Forest are used to carry out experiments and optimization,and the experimental results show that the Random Forest algorithm has the best performance,with an average accuracy of89.33%,high learning efficiency and great application potential.Then,the research on gas chromatography overlapping peak resolution algorithm based on genetic algorithm is carried out,and the overlapping peak resolution problem is transformed into a single-objective minimization problem of multiple decision variables by taking the genetic algorithm toolbox Geatpy as the implementation tool,the key parameters of the overlapping peak mathematical model as the solution object,and the error minimization between the fitting function and the original function as the goal.And the two application modes of fast detection and full detection were studied.The advantage of the fast detection mode is that it can maintain a high accuracy rate and have a very fast running speed,and the disadvantage is that the application range is small and the anti-interference ability is poor.In the fast detection mode,four algorithms can be used to solve the bimodal overlap problem,among which the optimal can control the parameter error within 4%;There are three algorithms that can be used to solve the trimodal overlap problem,where the optimal one can control the parameter error to less than 6%.The advantage of the full detection mode is that almost all algorithms have a high accuracy,whether it is a bimodal overlap or a trimodal overlap problem,the parameter error is mostly within 0.5%,and the disadvantage is that the operation takes a long time.After solving the problem of the analysis of general overlapping peaks,the research on shoulder-peaks analysis strategy was carried out,the definition and category of shoulder-peaks were first clarified,the characteristics of shoulder-peaks were systematically analyzed,and they were divided into 5 categories according to their characteristics,and the analysis experiments were carried out under different resolutions by using the full detection mode,and the experimental results showed that except for the equal width shoulder-peaks analysis effect,other types can achieve good results when the resolution is greater than 0.3.Then,the analytical strategy of the equalwidth shoulder with poor analytical effect was explored,and the strategy of batch fitting of main and subordinate peaks was applied,and a good analytical effect was obtained.