Study Of Multi-model Fusion Modeling Method And Its Application In Chemistry

Data-driven modeling is one of the hot topics in current modeling methods.Because it does not need to fully understand the internal operating mechanism of the system,it can establish effective information mapping relationships based on input and output data and have played an important role in complex system measurement,control,optimization,and other aspects.To deal with the shortcomings of single data-driven model,such as poor stability,incomplete extraction of effective information and unsatisfactory prediction accuracy,multi-model fusion modeling methods were developed.The research work mainly includes multi-model fusion modeling methods based on homogeneous models and multi-model fusion modeling methods based on heterogeneous models,which are used to solve the problem of describing the global characteristics of complex systems and improve the overall prediction performance of the model.The proposed modeling methods are applied to the prediction of the solubility of CO₂ in ionic liquids,the validity of the modeling method is verified,and the modeling method is comprehensively evaluated.The main research work is as follows:（1）A systematic review of common data-driven modeling methods is conducted,analysis and induction by consulting a large number of relevant domestic and foreign literature.Aiming at solving the modeling problems in various chemical and chemical processes,this paper systematically analyzes the data-driven modeling methods,and the basic theories and methods of modeling in the application background.It summarizes the current research status at home and abroad,and reviews in detail the advantages and disadvantages of various methods.The design ideas,research status and research prospects of multi-model fusion models in detail.（2）To deal with the limitations of a single data-driven modeling method such as incomplete extraction of effective information,poor stability and low prediction accuracy,a multi-model fusion modeling method based on homogeneous models was proposed.Firstly,by reasonably selecting different model parameters of the same type of model,a set of sub-models with parameter differences are established by training set,and different information of the research object is extracted by using parameter diversity method.Than the optimal sub-models which are selected by evaluation sets uses the linear fusion method based on bias entropy,the nonlinear fusion method based on BP neural network to establish two fusion models.Finally,the test set is used to test the prediction performance of the fusion models.The flow data set in the UCI database was used to verify the feasibility of the proposed modeling methods;（3）In order to solve the problem of extraction and fusion of effective information as well as establish a fusion model with better stability and predictive performance from the model structure level,a multi-model fusion modeling method based on heterogeneous models was proposed.First use the training set to build a set of sub-models containing multiple types of models（neural network,support vector machine,extreme learning machine）.Then use the evaluation set to select the optimal neural network,support vector machine and extreme learning machine from each type of sub-model,and two kinds of fusion models are established by the method of minimizing the error sum of squares and the information entropy method,respectively.Finally,the prediction performance of the fusion model is tested by the test set.The concrete flow data set in the UCI database was used to verify the feasibility of the proposed modeling method;（4）Taking the solubility of CO₂ in ionic liquid as the prediction object,the application research of multi-model fusion modeling method was carried out.First of all,for the prediction of CO₂ solubility of imidazole ionic liquids,the input variables of model are reasonably selected.Then collect a large amount of data through literature,and preprocess and group of sample sets.Finally,multi-model fusion modeling methods based on homogeneous models and heterogeneous models were used to build prediction models of the solubility of CO₂ in ionic liquids.The research results show that,compared with single data-driven model,the prediction models established by two multi-model fusion modeling methods have better prediction performance;compared with multi-model fusion modeling method based on homogeneous models,the fusion method based on heterogeneous models have the better prediction accuracy since sub-models broaden the adaptability of the effective information extracted from the data;a multi-model fusion modeling using an information entropy fusion method based on homogeneous models has better prediction performance and stability than the fusion method by minimum squared error,because it dispersed model differences and error factors.The research and application of two multi-model fusion modeling methods are carried out,and the results show that the multi-model fusion modeling method has important theoretical research significance and practical application value.The applications in chemistry and chemical industry are of great reference significance.