Research On Mixed Gas Glassification And Concentration Prediction Algorithms Based On Ensemble Learning

In the current development of machine olfactory technology,one of the main applications is the detection of toxic gases such as flammable and explosive gases.Machine learning algorithms play a major role,and their accuracy and time efficiency affect the timeliness of the entire system.Ther efore,it is extremely important to study the classification and concentration prediction methods of mixed gases.Based on the ensemble learning model,this paper studies the classification and concentration prediction algorithms of mixed gases.Aiming at the problem of data error in the process of data acquisition,the data set is discussed and analyzed.The baseline processing and dynamic time warping algorithm are used to uniformly process the data,compensate the drift of the sensor array data,and perform preliminary screening of the data.It is considered that the feature processing has a great influence on the algorithm.The feature extraction method of feature cleaning,feature construction and principal component analysis is used to perform feature processing to provide a more effective experimental data set for subsequent algorithms.At the same time,the features are visualized,and the classification and regression difficulty of the data are initially analyzed.The difficult reason for the mixed gas classification algorithm is analyzed.Inspired by the random forest algorithm,this paper proposes an extreme random tree model based on simulated annealing parameter tuning to classify the mixed gas.The use of extremely random tree node splitting is more random,which shortens the running time of the algorithm and matches the timeliness requirements in the mixed gas classification.The extreme random tree algorithm is parameter tuned by a fusion simulated annealing algorithm.The simulation results show that the proposed mixed gas classification method can effectively classify carbon monoxide,ethylene and methane with an accuracy of 99.28%,and the time efficiency is 66.85% higher than that of the random forest algorithm.The reason for the poor prediction of gas concentration regression is discussed.It is believed that multiple algorithms can be integrated by integrated learning algorithm to deal with more complex gas concentration prediction problems.This paper presents a gas concentration predicti on algorithm based on the basic Stacking model.Using random forest,extreme random tree and GBDT algorithm as the base learner,after Stacking fusion,it achieves strong generalization ability and high regression precision.The grid search algorithm is discussed.The grid search parameters are tuned to the base learner,and the Stacking regression model with grid search is realized.The experimental results show that the goodness of fit for carbon monoxide concentration prediction is 0.9901,and the goodness of prediction for ethylene concentration is 0.9913.In this paper,the proposed algorithm has obvious effect on regression fitting precision,and is suitable for concentration prediction,which provides reference and reference value for machine olfactor y algorithm.