Research On The Missing Data Imputation And Its Application In The Leaching Process Modeling

As the original information of production process, industrial production data is the important basis and foundation of process modeling and optimization. But in the actual industrial process, the data acquisition operation, the restriction of instrument operation environment and malfunctions often lead to values missing, which makes it hard to use the original values. Missing Data Imputation is an effective way to solve this problem. As a step of metal extracting, hydrometallurgy leaching process directly affects the metal recovery rate. However, because of the complexity of leaching process, data missing is a vital issue. Therefore, under the condition of date missing, the research on missing data imputation and modeling in the leaching process has important theoretical and practical significance.At the beginning, this thesis takes the acid leaching of cobalt compound ore as research background. The technological process of leaching process, process mechanism model and the development status of leaching process modeling are presented firstly in this paper. Missing data characteristics and classification, the basic principle of simple imputation methods including BP neural network, LS-SVM and GMM-EM imputation, and fundamental of MI are introduced in detail next. Based on the analysis of data deficient and its features during the acid intermittent leaching of cobalt compound ore, this article will launch the research from the deficiency of influential crucial values, such as sulfur dioxide flow, PH value of leaching agent, leaching rate, and apply various data packing methods into leaching process and modeling. On the basis of the mechanism model, studies are developed from the modeling and simulation of filling single or multiple missing variable data by the above different packing and modeling methods. According to the simulation results, this thesis evaluates the application performance of different methods in such aspects as packing results and ultimate modeling accuracy, concludes the method with which leaching process data is well packed and modeled under the condition of different kinds of data missing, and verifies the effectiveness of the leaching process modeling method based on data imputation. In this context, by analyzing the features of MI and GMM-EM, this paper presents theGMM-MI which combines GMM and MI. The simulation result shows that this method has a better comprehensive performance in the different situations of data loss. Thereby, the matter of missing data online imputation in the process of leaching rate prediction model online applications is well settled. Finally, through analyzing the system and modularizing the overall function, leaching rate modeling and prediction system is designed and implemented by using C# and MATLAB. And the simulation result proves that this system is effective and feasible.