A tungsten ore grade prediction model based on XRT induction characteristics and BP neural network
In order to achieve efficient XRT preselection for a tungsten ore,a model between XRT image and ore grade was established. Firstly,a tungsten ore was taken as a test sample for X-ray perspective imaging. Using MATLAB software,XRT grayscale image of the ore was generated,gray histogram was established and grayscale distribution of the image was counted. Secondly,based on BP and GA-BP neural networks,WO3 grade prediction models were established respectively,and the tungsten ore samples were used for training and testing. Coefficient of determination,root mean square error,mean absolute error,and mean deviation error of two prediction models were analyzed. Finally,based on the accuracy and generalization ability of the two prediction models,the appropriate tungsten ore grade prediction model and optimization method were determined. The results show that the grayscale distribution of XRT grayscale images of ores with different grades is obviously different. The grayscale distribution obtained from the images is highly correlated with the ore grade. The higher the grade of tungsten ore,the greater the proportion of pixel grayscale level at the low grayscale interval. The prediction model of ore grade can be established by the grayscale distribution of XRT images. The prediction model based on GA-BP neural network can obtain larger coefficient of determination and smaller error. It has higher prediction accuracy and stronger generalization ability,which can better predict WO3 grade of tungsten ore. In the case of small samples,GA-BP neural network prediction model is feasible and effective for WO3 grade prediction.
万方数据
维普
