Design Of Copper Ore Grade Prediction Algorithm And Software Oriented To On-Stream Analyzer

In the copper ore flotation process,real-time analysis of concentrate grade can guide the on-site operation process in real-time and improve the recovery rate of beneficiation.The current carrying grade analyzer is an important analytical instrument in the copper ore flotation process,and sample calibration is one of the core processes for measuring the target metal grade.However,in the process of sample calibration,the modeling methods used,such as least squares or neural networks,have insufficient measurement accuracy.On the other hand,the flotation process is complex,with poor production conditions and significant concept drift,resulting in insufficient adaptability of offline models.To solve the above problems,this theory proposes a model that combines the penalty likelihood function with the online sequential extreme learning machine(OS-ELM),and develops the calibration software of the analyzer based on this algorithm,which is used to analyze the grade of pulp samples in real time.The main research content of the theory is as follows:(1)A novel online learning algorithm based on OS-ELM neural network and Nonnegative Garrot(NNG)is proposed to address the complex modeling problem in the calibration process of current carrying grade analyzers.Firstly,the online learning capability of OS-ELM is utilized to address the continuous decline in offline model accuracy caused by fragmented data streams in industrial processes;Then,the sparse ability of NNG is utilized to compress the input weights of the OS-ELM network and filter out relevant variables;Finally,the effectiveness of the proposed algorithm was verified through numerical simulation and real industrial numerical simulation.The experimental results show that the proposed algorithm can effectively screen out relevant variables during sample calibration modeling,improve model accuracy and calibration efficiency.(2)A variable selection and structural optimization based OS-ELM network model(OP-NNG-OSELM)is proposed to address the issue of conceptual drift in rapidly growing data streams in industrial processes.Firstly,the online learning ability of OS-ELM is utilized to adapt to the impact of concept drift on the model,thereby reducing the problem of low efficiency caused by retraining the model.Secondly,the sparsity ability of NNG is utilized to impose varying degrees of penalties on input nodes and hidden layer nodes,achieving the simplification and performance optimization of the entire model.Finally,the effectiveness of the proposed algorithm was verified through artificial numerical simulation experiments and actual industrial datasets containing concept drift.The experimental results showed that the OP-NNG-OSELM algorithm can still accurately predict target variables when concept drift occurs in the data stream,and its prediction accuracy is higher than NNG-OSELM.(3)Aiming at the problem that the linear regression modeling method currently used in calibration software can lead to low measurement accuracy in sample calibration,this thesis introduces the online learning algorithm studied into the calibration and regression software,and then designs a calibration and regression software that has the ability to handle nonlinear and real-time analysis through Python,database,and Qt software.Finally,the dataset from a copper mine current carrying analyzer measuring the metal copper grade was used for the performance test of the calibration software.The results show that the software can handle the nonlinear and redundant issues of variables during modeling,and the software has real-time analysis capabilities,greatly improving production efficiency.