Study On Prediction Of The Rare Earth Element Composition Based On GRA-JITL

Rare earth is known as an ’ industrial vitamin ’.It is an indispensable raw material for aerospace,high-speed trains,sophisticated weapons.It is an important strategic resource.The state attaches great importance to the production and research of rare earth.Although China is a big country of rare earth,rare earth separation enterprises are still in the stage of offline analysis and manual adjustment in extraction separation control,and has the low automation level of extraction separation control,which leads to weak competitiveness in the rare earth market.The key to improving the automation level of enterprises lies in whether the online and accurate detection of rare earth element component content can be realized.At present,the research on the prediction of rare earth element component content mainly focuses on offline modeling,which can’t cope with the sudden situation of extraction conditions.Aiming at the problems of off-line,large time delay,and weak anti-interference ability of the existing rare earth element component content model,this paper proposes a just-in-time learning algorithm based on grey relational analysis(GRA-JITL)to establish the online detection model of component content in rare earth extraction process.Aiming at the uncertainty of model parameters,a genetic algorithm with a stagnation backtracking strategy(SBS-GA)is proposed.R(red),G(green),B(blue),H(hue),S(saturation),I(brightness),RR(relative red component),RG(relative green component),RB(relative blue component),CVA(color vector angle)ten kinds of rare earth solution image color features are used as auxiliary variables for experimental verification.The main research contents are as follows:1)Analyzing the relationship between the color characteristics of rare earth solution images and the content of element components and determine the model auxiliary variables.In this paper,the extraction and separation process of Pr/Nd elements is taken as the research object.According to the color characteristics of Pr/Nd solution,the relationship between the color characteristic components and the content of Nd element components is analyzed.The results show that the color characteristics of R,G,B,H,S,I,RR,RG,RB,and CVA have a strong correlation with the content of Nd element components,which can be used as input variables of the model.2)Aiming at the lack of online prediction ability of traditional just-in-time learning algorithm model,the GRA-JITL algorithm is proposed.Considering the shortcomings of the traditional just-in-time learning algorithm in the selection of learning sets,the grey relational analysis method is combined with the hash table to select the learning set,and then the least squares support vector machine is used for local modeling,and the database update strategy is used to improve the adaptive ability of the GRA-JITL algorithm.Based on Pr / Nd extraction solution samples,with 5 and 10 color feature components as input variables,the GRA-JITL was compared with offline models such as ELM(extreme learning machine)and LSSVM(least squares support vector machine),as well as just-in-time learning algorithm based kvector nearest neighbor(k-VNN-JITL)and just-in-time learning algorithm based mutual information weighted similarity criterion(MI-SJITL-LSSVM).The results show that the GRA-JITL model proposed in this paper has the best prediction effect when the ten color features of R,G,B,H,S,I,RR,RG,RB,and CVA are used as input variables.3)Aiming at the uncertainty of the above model parameters,a genetic algorithm with a stagnation backtracking strategy(SBS-GA)is proposed for parameter optimization.Considering that the traditional genetic algorithm has the advantages of global optimization and simple model structure,as well as the disadvantages of low efficiency and premature convergence,this paper introduces the stagnation backtracking strategy to improve it,analyzes the convergence,and uses the standard multimodal function to verify the validity.The improved genetic algorithm with stagnation backtracking is applied to the component content model of the rare earth extraction process based on GRA-JITL.The experimental results show that the prediction accuracy of the GRA-JITL component content model optimized by SBS-GA is further improved.In summary,the SBS-GA algorithm proposed in this paper can ensure the global solution of the optimization parameters.The GRA-JITL algorithm proposed in this paper has high prediction accuracy when it is applied to the online detection of component content in the rare earth extraction process and can be used for the online detection of element component content in rare earth extraction production sites.