Research On XGBoost Soft Sensing Model Of Flotation Concentrate Grade Based On Improved WOA Algorithm

In the process of flotation production of iron ore,concentrate grade is one of the key economic and technical indexes in flotation operation.Because the online concentrate grade analysis instrument is expensive and challenging to run stably for a long time,and the off-line test results based on manual have a large lag in time,they cannot meet the requirements of online flotation process control.Therefore,there are problems of large fluctuation of concentrate grade and poor economic benefit in flotation production.At present,soft sensing technology has become an effective alternative means to solve the problem of real-time monitoring of process variables which are difficult to detect online by industry instruments.In view of the characteristics of large hysteresis,nonlinearity,strong coupling and difficult to establish complex mechanism model in flotation production process,a soft sensor prediction model of concentrate grade was established using Monte Carlo method,improved whale optimization algorithm,XGBoost algorithm and soft sensor technology as main technical means.The prediction accuracy of the soft sensor model for concentrate grade is good,the prediction error distribution is between[-1.204%,1.483%],and the model fitting coefficient R~2is 0.8730,the prediction results can better reflect the change trend of concentrate grade.The model prediction results are of great significance to guide flotation production,which can effectively save production cost and improve product quality.The main work of this paper is as follows:(1)The first chapter is the introduction of the article.This thesis mainly introduces the current research status of flotation production and soft sensor technology,and determines the overall research scheme of this thesis.(2)For the modeling data samples collected in the flotation workshop,there are problems such as high dimensionality and abnormal refutation.Firstly,the modeling data samples were uniformly standardized,and the abnormal data hidden in the modeling data samples were fully eliminated by Monte Carlo abnormal data diagnosis method.Then,the K-means clustering method is used to cluster the modeling data samples,and the modeling training set and testing set are reasonably divided.Finally,the principal component analysis(PCA)is used to reduce the dimensionality of the modeling data samples and verify the clustering effect,simplify the model structure and improve the data quality.It lays the foundation for the establishment of soft sensor prediction model,and improves the calculation speed and prediction accuracy.(3)The theoretical basis of XGBoost algorithm is studied,and the hyperparameters that affect the performance of XGBoost algorithm are analyzed.In order to improve the performance of soft sensor prediction model,a whale optimization algorithm(WOA)is introduced to optimize the hyperparameters of XGBoost algorithm.In order to overcome the shortcomings of the standard WOA algorithm and improve the optimization performance,this thesis uses various strategies to improve the WOA algorithm.Firstly,to obtain high-quality initial population with better diversity,this thesis uses chaos tent mapping combined with reverse learning to expand the whole search space.Secondly,the dynamic adaptive weight factor is combined with the position updating method of the algorithm,and the nonlinear change of the weight factor is used to balance better the global optimization and local exploration of the algorithm.Finally,to accelerate the convergence of the algorithm and avoid the algorithm falling into premature,a greedy selection strategy is introduced.The rationality and convergence of the improved algorithm(TOWWOA)are proved by the comparative simulation results of ten standard test functions in CEC2005 and two classical engineering design problems.(4)Aiming at the problem of online monitoring of flotation concentrate grade,a soft sensor prediction model of flotation concentrate grade optimized by XGBoost based on TOWWOA algorithm was built.TOWWOA algorithm is used to optimize the hyperparameters(max depth,min child weight,subsample,etc.)of XGBoost model.The inverse of the coefficient of determination of ten-fold cross validation is taken as the fitness function of the algorithm.The performance of the soft sensor prediction model was evaluated according to the model’s root mean square error,fitting determination coefficient,and prediction accuracy.Through the test and comparison of different models,the advantages of the TOWWOA-XGBoost model are confirmed,and the prediction accuracy of the model can meet the needs of online monitoring of concentrate grade in flotation production.