The Research Of Soft Sensor Modeling Method Based On NNG And Neural Network For Copper Flotation Process

The degree of industrial automation is getting higher and higher with the rapid development of modern society,and the requirements for the measurement and control of key variables in industrial processes are also getting higher and higher.Aiming at some key variables in the industrial process which are difficult to measure,lack of real-time performance and high measurement cost,etc.Soft measurement technology provides a method to replace hardware sensors with mathematical modeling and software programming,which has attracted the research and attention of the majority of scholars.The paper takes the flotation process of a copper mine as the research background.The concentrate grade represents the product quality and production efficiency of the flotation process,which is the key technical indicator of the copper mine flotation process.However,the variable is usually obtained by offline manual testing in actual production,which has the following problems: insufficient real-time performance and high measurement cost etc.The paper systematically analyzes the flotation production process of A copper mine and studies soft measurement modeling algorithm based on neural network,which establishes a data-driven soft sensor model for copper concentrate grade.The main research contents of this paper are as follows:(1)For complex systems,a variable selection algorithm based on the combination of nonnegative garrote(NNG)algorithm and extremal optimization(EO)algorithm is proposed for multi-layer perceptron(MLP).First of all,complex systems are modeled based on MLP by using existing data sets,and a trained MLP neural network is obtained;Secondly,the input weight coefficients of the trained MLP neural network are compressed and the input variable is selected by NNG algorithm.On the basis,further local variable selection is performed to obtain a more refined data set through the EO algorithm,and the final MLP model is given.(2)The performance of the proposed algorithm is tested through two different types of numerical simulation examples from various variable sizes,sample sizes and correlations,etc,and the performance of the algorithm is comprehensively compared with other classic MLP soft sensor algorithms.Simulation results show that the proposed algorithm combines the advantages of global compression of NNG algorithm and local search of EO algorithm,which is superior to other algorithms in both algorithm accuracy and variable selection accuracy.(3)In this paper,the technological process of flotation process of a copper mine is analyzed,including the physical and chemical reaction mechanism of flotation,flotation device,and various variables provided by the industrial data system were studied.In view of the problems existing in the measurement of concentrate grade for the current copper ore flotation process,as well as the characteristics of non-linearity,complexity and multiple variables in the process.The researched algorithm is applied to the modeling of copper mine flotation process.The simulation results show that the algorithm can successfully predict the dynamic change of copper grade value of copper concentrate.At the same time,the analysis of variable importance given by the algorithm is consistent with the actual operating experience,which can provide theoretical and technical support for process optimization and control system improvement.