Hybrid Modeling Of Component Content In The Rare Earth Extraction Process Based On Data-driven Compensation

As one of the few rare earth resources in the world,China has assumed the role of rare earth supplier in the past decades.But at the moment due to technical limitations,most of the rare earth production enterprises on the change of component content in the extraction of tub can’t rapid monitoring,makes the actual working condition of the distribution of components in appeared deviation fail to adopt effective means to adjust working condition,the export product quality is hard to realtime control problems,restricted the production to improve the economic benefits of the enterprise.In this paper,aiming at the shortcomings of traditional soft measurement methods of component content in rare earth extraction process,such as low precision of single mechanism model and opaque modeling process of single data model,a hybrid modeling method of component content in rare earth extraction process based on data-driven compensation was proposed.The main research contents of this paper are as follows:Firstly,the mechanism model of rare earth extraction process was established.By analyzing the Ce Pr/Nd cascade extraction separation process in actual production,the piecewise assembly simplification method was used to simplify the process to reduce the dimension of the mechanism model,simplify the modeling process and reduce the calculation of the model.The mechanism model of the simplified process was established step by step based on the principle of material balance,and the mechanism model with parameter description was identified by the recursive least square method to achieve the detection of component content.In view of the above mechanism model in the process of modeling set a lot of assumptions to avoid noise interference,so there is inevitable deviation in the actual apply,this paper uses the data-driven model to compensate the error generated by the mechanism model.Due to different rare earth ions with special structure,component content of rare earth ions in the solution under the irradiation of visible light will show different color characteristics,using the relationship between color features and components in error compensation based on the random forest algorithm(RF)model,combined with detection mechanism model and error compensation model components in hybrid model.To solve the problem of small sample size in data compensation modeling,a data set enhancement method based on generative adversation network(GAN)was proposed.Uses KMeans-Smote algorithm for data sets of data compensation model extension,adopts the random forest discriminant for extension increases the data is in accordance with the real data distribution,according to the topology based on the data obtained from data set billiton after toughening detection method of the rare earth extraction process components in hybrid model,further improve the detection precision of the hybrid model.In conclusion,the hybrid model based on data extension method has good tracking performance for component content change in Ce Pr/Nd extraction process,and meets the requirements of component content change detection in production site.