Study On Prediction Of Flotation Concentrate Grade Based On Improved SSA Optimized Neural Network

In the flotation production process,the concentrate grade is the most important investigation factor,and its online prediction plays an important role in improving the flotation process quality and optimizing the operation process,as well as in the economic benefits of iron and steel enterprises.However,the flotation workshop still relies heavily on experienced technicians for operation at present,which makes the adjustment of key indicators in all aspects highly subjective.In addition,the interference of various uncertain factors in the process flow makes the control and management of concentrate grade impossible to be guaranteed.The soft sensor technology can establish a model between the measurable variables of flotation process and concentrate grade,and monitor the flotation concentrate grade online.This thesis proposes a flotation concentrate grade prediction model with computer image processing technology,neural network and intelligent optimization algorithm as the main technical means.A variety of hybrid strategies are introduced to improve SSA(Sparrow Search)algorithm,and then optimizing Elman NN to improve the precision and constancy of the model.(1)In view of the difficulty in measuring the key parameters of flotation,a feasible scheme of online measurement is proposed.The parameters that affect the concentrate grade are obtained through image processing technology,including feed particle size and pulp texture characteristics(energy,entropy,contrast,correlation).(2)According to the wide popularization and application of soft sensor technology,Elman NN prediction model is introduced in this thesis.The fundamentals,arguments setting and network architecture of Elman NN are introduced in detail,but the network also has shortcomings.Aiming at the defects of the above three settings,the corresponding solutions are proposed,and sparrow search algorithm is introduced to optimize the weight and threshold of the model.(3)In an effort to improve the performance of SSA,four methods are used to improve it.At the beginning of the cycle,using Sine map to initialize the population,so as to make the generated population uniform,increase the multiplicity of the population,and increase the quality of the population;Using BOA(butterfly algorithm)to improve the location update of sparrow discoverer,the search space is expanded,and the defect of lack of information exchange between individuals in traditional SSA algorithm is improved;Lévy flight disturbance is introduced to further optimize the algorithm.Lévy flight has the characteristics of long and short distance random jump,which can overcome the question that the algorithm falls into local optimization to advantage.The improved SLSSA algorithm is compared with other five algorithms using standard test functions.The results show that SLSSA algorithm performs well in all aspects of performance.(4)Aiming at the problem of on-line prediction of flotation concentrate grade,a concentrate grade prediction model based on SLSSA algorithm optimized Elman neural network was established.Firstly,the flotation data collected on site and online are preprocessed to determine the input and output variables;Secondly,the parameters of the neural network are set,SLSSA algorithm is introduced to optimize the parameters of Elman neural network,and the optimal parameter combination(weight and threshold)is selected by evaluating the mean value of ten fold cross validation;Finally,the grade of flotation concentrate is predicted,and the performance of the soft sensor model is evaluated through mean square error,root mean square error,and fitting determination coefficient.The precision of the model is increased to advantage.