Realization Of Gas Detection Algorithms For Micro-system Applications

At present,the technology of gas detection and recognition based on machine olfaction has developed rapidly,and has been widely used in many fields such as biomedical diagnosis and treatment.At the same time,due to the requirements of volume,power consumption and integration of detection equipment,the combination of intelligent micro system and biomedical diagnosis and treatment based on olfactory application has gradually become an important research direction.With the support of medical data resources,chip technology,MEMS technology and advanced processing technology,intelligent micro system can realize the intellectualization and miniaturization of medical instruments and equipment,thus greatly improving the efficiency of biomedical diagnosis.However,due to the requirements of low power consumption,miniaturization and real-time performance,the application of intelligent algorithms in micro system is limited by power consumption,time and efficiency,which makes the algorithm design and deployment optimization for resource constrained micro system platform extremely important.In this paper,the design and deployment optimization of mixed gas concentration detection algorithm are studied based on machine learning technology for micro system applications1.in view of the lack of effective data set in the early screening research of type II diabetes based on machine olfactory method,based on the analysis of the exhaled gas content of human body,combined with the main gases that affect the early diagnosis and treatment of diabetic patients,a data set containing four gas mixtures is constructed,which provides an effective reference for subsequent algorithm design and evaluation.2.Aiming at the problem of detection and concentration estimation of specific gas components in mixed gas,based on the analysis of the multi-dimensional cross response of the sensor and the nonlinear characteristics of the output data,an algorithm based on neural network is proposed to estimate the concentration of gas components.A modeling and analysis strategy is proposed to optimize the structural parameters of the algorithm.By modeling the structural parameters and the amount of model reasoning calculation,high performance can be achieved on the premise of meeting the power consumption requirements of the micro system.3.Aiming at the problem of lack of sample data caused by the limitation of experimental environment and the difficulty of sampling,a PCA-SVR algorithm is proposed to estimate the mixed gas concentration under the condition of small sample data training.By constructing the design strategy based on structural risk minimization,the algorithm achieves better regression performance under small sample conditions,which provides a reference for early screening of diabetes screening.4.In order to further improve the detection accuracy of gas mixture,an improved ensemble learning strategy is proposed based on ensemble learning.Through the heterogeneous integration of several basic machine learning algorithms,the dynamic updating of data sets and the strategy optimization design of base learner sequence generation,the algorithm significantly improves the accuracy of gas concentration detection on the basis of maintaining the original operating cost,and makes the model have strong anti over fitting ability.In addition,as an end-toend integrated learning framework,the model can adapt to any end-to-end regression algorithm,which provides good scalability for further optimization of design space exploration.