Study On Modeling Of Rare Earth Extraction Based On SCN

Recently,the artificial intelligence technology has been gradually applied to various industrial production processes,with continuous advancing and popularizing.As a resource of strategic,rare earth elements play an irreplaceable role in many high-tech industries,hem also have an important influence on national economy and social development.In order to transform the advantages of rare earth resources into industries,the rare earth industry was higher required.In addition,because of the characteristics of rare earth industrial process such as large time delay,nonlinear,time-varying,strong coupling,multivariable and so on,the traditional control theory and information processing technology cannot satisfy the urgent demand for advanced automation technology in rare earth industry.In the process of rare earth extraction and separation,component content is crucial on the design of control system,product quality control and energy consumption.But the traditional off-line analysis method cannot give the component content in time,which is not conducive to the operation of the control system.In this context,based on the analysis of rare earth element component content detection method and soft sensor modeling method,this paper studied a soft sensor modeling of rare earth extraction process by using stochastic configuration network.The specific content is as follows:In order to ensure the feasibility of the method presented in this paper,the regression performance of the stochastic configured network is verified firstly,a real regression task is used to verify.After comparing with several excellent machine learning methods,the results show that the stochastic configuration network is appropriate in this kind of actual regression task.Then,the standard stochastic configuration network was used to model the process of rare earth extraction and separation.After comparing with several similar neural network models,the results of simulation proved the feasibility of applying the network in this field.Subsequently,with the increasing number of hidden layer neurons in generating the stochastic configuration network,the network structure will gradually become more complex,and the problem of over fitting may occur,which will reduce the generalization performance of the model.In addition,due to the randomization of the network parameters and the uncertainty of network structure,the network will be unstable.Therefore,this paper first joined the regularization method in stochastic configuration network to improve its generalization ability.And then in order to ensure the selection of initial parameters is reasonable and further guarantee the measurement performance and stability of the model,this paper used the modified Grey Wolf Optimization Algorithm to optimize the parameters' selection of stochastic configuration network.Then this paper builds a stochastic configuration network model based on the optimizer of the modified Grey Wolf Optimization Algorithm.The model established in this paper is compared with several other models,and the two indexes used to evaluate the performance of the model are low,which confirms the effectiveness of the method.The research results provide an effective method for soft sensing modeling of component content in rare earth extraction production process,and have certain reference significance for promoting the automation of rare earth production process.