Research On Prediction Of Flotation Concentrate Grade Based On Improved ASO Optimized Neural Network

Concentrate grade is an important economic indicator to measure flotation production,which is of great significance for improving the economic efficiency of steel enterprises and implementing the boutique strategy.However,the traditional flotation concentrate grade control still relies on experienced workshop staff to adjust the condition parameters.The non-linearity of the flotation process and many uncontrollable factors on site cannot guarantee the optimal control of the concentrate grade.Therefore,this paper uses BP neural network and intelligent optimization algorithm as the main technical means,and builds a flotation concentrate grade prediction model based on improved atomic search optimization to optimize BP neural network to realize the optimal control of flotation concentrate grade.Aiming at the problem of flotation concentrate grade prediction,the classic BP neural network prediction model is introduced and the basic structure,parameter settings and network structure of BP neural network are described in detail.Four limitations affecting the prediction of BP neural network are discussed in detail:parameter settings,function selection,network structure and data samples.To deal with these four deficiencies,corresponding solutions,have been proposed.Aiming at the defect that the BP neural network is sensitive to the initial weight and threshold,which leads to the problem of low prediction accuracy,an intelligent optimization algorithm is introduced.To further improve the performance of intelligent optimization algorithms,two improved atomic search optimizations based on hybrid strategies are proposed,namely CALFASO and ASOINU algorithms.In the CALFASO algorithm,the introduction of the cellular automata structure provides good information exchange for the population.The long-short jump mechanism of Levy’s flight effectively helps the algorithm avoid local optimization and premature convergence.The linear inertia weight balances the global search and the local search;In the ASOINU algorithm,the neighborhood topology learning mechanism enriches the population diversity,the global optimum and worst individual update mechanism further improves the local search ability,and the introduction of nonlinear weights better balances exploration and mining.The CALFASO and ASOINU algorithms and other classic algorithms have been tested on the CEC2017 test set and two sets of engineering design cases.The results show that the improved ASO algorithm proposed in this paper has better calculation accuracy and convergence performance.Taking the ASOINU algorithm as an example,a hybrid prediction model ASOINU-BP is built.To preliminarily verify the prediction performance of ASOINU-BP,an evaluation system for standard data sets was established,which was verified on three sets of UCI machine learning regression data sets with other classic prediction models.The results show that ASOINU-BP has better fitting ability and smaller prediction error on UCI machine learning regression data set.The ASOINU-BP model is used to study the flotation concentrate grade prediction.First,SPSS and Matlab software are used to preprocess the flotation data collected on the spot,and the model of the flotation concentrate grade prediction example is determined to determine the input and output variables.Secondly,the parameters are determined for the intelligent optimization algorithm and neural network parameters in ASOINU-BP.Finally,an industrial data prediction evaluation system was established,and the flotation concentrate grade prediction evaluation was carried out.The results show that the ASOINU-BP algorithm has good predictive ability and reliability on the flotation concentrate grade problem.