Research On Mixed Gas Detection Method Based On Unbalanced Sample

A variety of gases exist in industrial and living environments,and these gases are usually in the form of mixtures.Mixed gas detection based on MOS gas sensor array has become a hot research field.However,in practical conditions,the performance of traditional machine learning model is degraded due to unbalanced training samples.This paper studies the method of composition identification and concentration prediction of mixed gases under the condition of sample imbalance.The main research contents are as follows:In order to solve the problem of low recognition accuracy of quasi-unbalanced gas mixture,ADASYN-GA-SVM method is proposed in this paper.First,hierarchical 50-fold cross-validation was used to ensure that the proportion of sample sizes in each subset was the same as that in the original data set.Adaptive synthetic oversampling(ADASYN)method is used to adaptively transfer the decision boundary to the hard-to-learn samples,augment the minority classes of samples in the dataset,and form a new training set with the original training set.Combined with the support vector machine(SVM)good processing ability of small sample data,the mixture gas component identification is realized.In order to further improve the gas component identification accuracy of the model,genetic algorithm(GA)was used to optimize the penalty factor and kernel parameters of the SVM model.The experimental results show that the recognition accuracy of unbalanced samples is improved by using different sample augmentation methods,and the average recognition accuracy of ADASYN-GA-SVM method is 6.8% higher than that before the augmentation.Aiming at the problem of low precision of sample unbalanced mixed gas concentration prediction.This paper put forward SMOTE-PSO-MLSSVR method to predict the concentration of the mixture of gas.Extend the minority class in the training set by using the characteristic of synthetic minority oversampling(SMOTE)method to reduce overfitting and increase sample diversity.Combined with the characteristics of high accuracy and strong robustness of multi-output least squares suppor vector machine(MLSSVR)method,the mixed gas concentration can be predicted.In order to solve the problem of two regularization parameters and kernel parameter selection in MLSSVR,the particle swarm optimization algorithm(PSO)is used to optimize the superparameters in MLSSVR model,which has the advantages of fast convergence and easy implementation.The experimental results show that the root mean square error(RMSE)and mean absolute percentage error(MAPE)of the model after sample augmentation are increased by 0.3462 and 0.6828,respectively.