基于遗传算法优化BP神经网络的生石膏超细磨预测效果研究

Research on the prediction effect of ultra-fine grinding of raw gypsumbased on genetic algorithm optimized BP neural network

为提高BP神经网络对生石膏超细磨效果的预测准确性,采用Pearson相关系数对超细石膏粉体正交试验产品细度与影响因素的显著性进行分析,并利用遗传算法优化BP神经网络对超细石膏粉体试验产品的d_(50)和d_(90)进行预测,结果表明:超细石膏粉体制备过程中影响细度因素的显著性由大到小依次为排矿口宽度、矿浆质量分数和超细磨时间。利用排矿口宽度和矿浆质量分数两个主要影响因素,利用遗传算法对BP神经网络进行优化,与未优化的BP神经网络相比,经遗传算法优化的BP神经网络具有更高的精度,预测误差也更小,其d_(50)平均绝对误差为0.7575,均方误差为0.7977,均方误差根为0.8931,平均绝对百分比误差为4.4838%;d_(90)平均绝对误差为0.7870,均方误差为0.8294,均方误差根为0.9107,平均绝对百分比误差为1.6658%。研究成果可为超细粉体的制备提供参考。

In order to improve the prediction accuracy of BP neural network for superfine grinding effect of raw gypsum,Pearson correlation coefficient was used to analyze the fineness and the significance of influencing factors for orthogonal experimental products of superfine gypsum powder,and BP neural network optimized by genetic algorithm was used to predict d_(50)and d_(90)for test products of superfine gypsum powder.The results showed that the significance of influencing factors on fineness in the preparation process of superfine gypsum powder,from highest to lowest,was the width of the discharge port,the quality fraction of the slurry,and the superfine grinding time.By utilizing the two main influencing factors of ore discharge width and quality fraction of the slurry,the genetic algorithm was used to optimize the BP neural network.Compared with the unoptimized BP neural network,the genetic algorithm optimized BP neural network has higher accuracy and smaller prediction error.Its d_(50)average absolute error is 0.7575,mean square error is 0.7977,root mean square error is 0.8931,and average absolute percentage error is 4.4838%;The average absolute error of d_(90)is 0.7870,the mean square error is 0.8294,the root mean square error is 0.9107,and the average absolute percentage error is 1.6658%.The research results can provide reference for the preparation of superfine powders.