Application Of Genetic Algorithm Based Principle Component Selection Method In Concentrate Grade And Production Rate Prediction In The Mineral Process

Mineral Processing means that the original ore from the underground is concentrated base on physical or chemical methods. The useful portion concentrated is called concentrate. The degree of it is denoted as concentrate grade while the production of it is called concentrate production rate. They are both very important indices for mineral process. And setting them properly can lead to effective production control and finally to the improvement of the overall profit of the enterprise.Because mineral process is a kind of industrial process, it consists of a series of procedures. And every procedure has its own index to evaluate its effect. All of these procedures indices can influence the value of concentrate production rate and concentrate grade directly or indirectly. Therefore, if we can effectively consider the relationship between all the procedures indices and the concentrate production rate and concentrate grade, and then establish related model for that relationship, the concentrate production rate and concentrate grade can be predicted by procedure indices. This will enable tuning of the setting of production index, which is an effective method to relieve the problem of time lag phenomenon in mineral process control. Thus, for predicting the concentrate production rate and concentrate grade accurately, this paper is focused on predictive modeling for concentrate production rate and concentrate grade.This paper is supported by the grant of "Closed-loop Optimization Algorithm on Technical Indices of Complex Industrial Process in Dynamic Environment". The method:"Genetic Algorithm based Principle Component Selection Method" is proposed. Then it is applied into concentrate production rate and concentrate grade predictive modeling problem,. And the prediction accuracy has been improved by this method. The main work of this paper is list as follows:(1) Propose "Genetic Algorithm based Principle Component Selection Method"Principle Component Analysis is a kind of feature extraction method, where original variables are transformed linearly into principle components and then a number of principle components will be selected. The selected set of principle components should then represent all the original variables. However, the selected set should consist of principle components which are really important to output prediction. Also, an amount of information contained in the set should also be ensured. The information in the set should be enough to ensure the robustness of the predictive model. These explain why the problem of principle component selection is difficult to be solved. In this paper, the proposed method can evaluate the set of principle component based on the performance on training data set, which can ensure that the selected principle components can help predict the output. Also, this method can ensure the total amount of information in the set of principle component. Based on case of study, the effectiveness of the method proposed in this paper is proved.(2) Predictive modeling of concentrate production rate based on "Genetic Algorithm based Principle Component Selection Method" and LS-SVMBecause of the complicated relationship between procedure indices and the overall production index, such as concentrate production rate, mechanism model is unavailable in predictive modeling of mineral process. Therefore, we apply LS-SVM to establish the predictive modeling which can describe the relationship between all the procedures indices and the concentrate production rate. The parameters of LS-SVM are optimized by PSO. And the feature of the input of LS-SVM is extracted by the method we proposed in Chapter Ⅱ. Thus the noise and redundant information can be excluded while the useful information in the input can be maintained. By compared to prior method in concentrate production rate, method proposed by Weijian Cheng, the improvement by our method can be proved.(3) Predictive modeling of concentrate grade based on "Genetic Algorithm based Principle Component Selection Method" and "linear model and nonlinear error compensation model"Considering the characteristic of concentrate grade and related database, we accordingly applied an effective concentrate grade predictive model. And then "Genetic Algorithm based Principle Component Selection Method" proposed by this paper is applied to extract the feature of the model input. By case study, we test the effectiveness of our method and current methods. Then these results are compared and the results show that the result get from the method proposed by this paper is better. This proves that our method has effectively improved the accuracy of concentrate grade prediction.