Chromatographic Spectral Analysis Method Based On Generation Network And Optimization Algorithm

In the experiment of mixture analysis,it is very important to separate the substances from the mixture.The method of High Performance Liquid Chromatography with Diode Array Detector(HPLC-DAD)is used for the qualitative,quantitative and purity analysis of compounds.The data set generated by this method is called HPLC-DAD data set,which is a two-dimensional matrix formed by the fusion of chromatographic peak and spectral curve.However,most of the algorithms used to separate HPLC-DAD data set have their own disadvantages,for example,it is necessary to know in advance the number of components in the HPLC-DAD data,the initial spectral estimation matrix or the standard curve of the chromatographic peak,which is difficult to obtain in most cases.Especially when the components contained in HPLC-DAD data are more complex and the chromatographic peaks of each substance overlap with each other seriously,it is more difficult to separate each substance successfully.Therefore,this paper proposes to separate HPLC-DAD data by using the improved Conditional Deep Convolutional Generative Adversarial Network and the mixed signal separation algorithm based on optimization algorithm.The main research contents of this paper include:1.Aiming at the limitations of the existing HPLC-DAD data separation methods,a method combining the improved chromatographic peak generation network and the mixed signal separation algorithm based on GRCM model is proposed.Firstly,the principle of GRCM model is introduced,and the characteristic of the GRCM model is analyzed by using simulation data set and real data set.Then,the parameter optimization algorithm MIPSO and LBFGS algorithm involved in the above methods are analyzed.Finally,the validity and applicability of the mixed signal analysis algorithm based on GRCM model are proved by experiments.2.Four methods are used to improve the Conditional Deep Convolutional Generative Adversarial Network(C-DCGAN),and a Conditional Wasserstein Generative Adversarial Network(C-WGAN)is proposed.Based on this model,a specific signal generator is constructed.Through many experiments,it is proved that the improved C-WGAN model can generate signals with specific distribution attributes according to different training sets.On the basis of DCGAN,C-WGAN uses Wasserstein distance to replace the loss model in the original model,and adds spectral normalization to the generation network and discrimination network to obtain a stable network structure;Secondly,adding condition information y to the generation network and discrimination network can solve the problem of uncontrollable samples generated by the generator;Finally,the optimal noise input dimension of samples is obtained through maximum likelihood estimation algorithm and experimental estimation.3.Combining C-WGAN model and GRCM model,the framework of HPLC-DAD data separation method was established.Firstly,the C-WGAN model was constructed based on DCGAN.Then,the C-WGAN model was trained by using a large number of real chromatographic peak curves and the chromatographic peak curves constructed based on Gaussian curves,so that the C-WGAN model could generate a reference curve that was more similar to the real chromatographic peak in morphology.Finally,based on the optimization strategy of random input vector of GRCM framework and C-WGAN model,Quasi Newton Hybrid Algorithm and Multi-Target Intermittent Particle Swarm Optimization are combined,and the simulation HPLC-DAD dataset and real HPLC-DAD dataset are respectively used for experimental verification.It is proved that the method proposed in this paper can not only improve the accuracy of data separation in HPLC-DAD,but also improve the training speed.The thesis includes 25 figures,5 tables and 69 references.