Study On Gas Concentration Prediction Algorithm In Electronic Nose System

The electronic nose system consists of a gas sensor array,a signal pre-processing module,and a pattern recognition module.When the electronic nose system is applied to project the gas to be predicted,the information of the gas is collected by the sensor array,then the response signal is processed by the signal pre-processing module.Finally,the qualitative analysis or quantitative regression of the gas to be predicted is realized by the pattern recognition module.At present,the research of electronic nose mainly focuses on the qualitative analysis,whose research is far from enough.The pattern recognition algorithm of electronic nose is crucial to the accuracy of gas concentration detection.Research on the current pattern recognition algorithm of electronic nose is mainly based on the traditional neural network algorithm,which can project the gas concentration better.However,there is still much room to improve the accuracy of gas concentration detection.For the limitation of the current pattern recognition algorithm in electronic nose,the gas to be predicted is analyzed in two different scenarios in this thesis,and optimizes the parameters of the neural network and enhances the robustness and accuracy of the quantitative regression of the electronic nose by using the optimization algorithm.The algorithm proposed is as follows:1.Gas concentration prediction algorithm based on PSOABC-ELM.The algorithm is embedded and fused by particle swarm optimization and artificial bee colony algorithm.Firstly,the particle swarm optimization is used as the main algorithm,and the following and reconnaissance stages in artificial bee colony algorithm are improved and combined with the particle swarm optimization.By adding control parameters and judging the execution stage of the optimization algorithm,the fused algorithm is used to optimize the weights and thresholds of the extreme learning machine.Finally,the optimized neural network model is used to project the concentration of a single gas,in order to realize the quantitative regression of electronic nose to single gas.2.PSO-LSTM based gas mixture concentration prediction algorithm.Based on the data pre-processing of convolution neural network,the algorithm uses particle swarm optimization to optimize the hyperparameters of long and short time memory neural network,which reduces the combination time of hyperparameters according to experience and many experiments,and also eliminates the hidden trouble that can not be determined by the hyperparameter combination.When the mixed gas is input,the concentration of each type of gas is output by preprocessing and pattern recognition,so as to realize the quantitative regression of electronic nose to the mixed gas.The experiments use the data set of California Irvine University,the single gas concentration data set is used to verify gas concentration prediction algorithm based on PSOABC-ELM,and mixed gas concentration data set is used to verify PSO-LSTM based gas mixture concentration prediction algorithm.The experimental results show that the algorithms are effective.