基于GRA-GSBP选矿成本预测的研究

Mineral Processing Cost Prediction based on GRA-GS BP

针对选矿成本影响因素较多,各因素间存在耦合和非线性关系以及BP神经网络隐含层节点数难以选择的问题,提出一种基于灰色关联分析与黄金分割法改进BP神经网络的成本预测法。首先运用灰色关联分析法计算各因素与选矿成本的关联度,选取关联度最大的四个变量作为BP网络的输入;其次采用黄金分割法搜索历史数据区间中的理想数值,在高精度的要求下,对隐含层节点数频繁出现的区间进行拓展,求得非线性映射能力更强的隐含层网络节点数;最后利用仪表柜中储存的现场数据对成本预测模型进行验证,验证结果证明该方法能够实时准确地预测选矿成本的变化趋势。

To solve the problems in mineral processing, including multiple affecting elements, coupling or non-linear relationships existed among the factors, difficulty while determining the hidden layer nodes of BP neural network, this paper put forward an improved BP neural network cost forecasting method based on gray relational analysis and golden segmentation. Firstly, GRA is used to calculate the correlation between the factors and the cost of ore dressing, and then four variables which have the largest correlation are chosen as the inputs of the BP network. Secondly, the principle of golden segmentation is applied to search the desired value in the region of historical data. To obtain the number of hidden layer nodes which can approach more strongly, the region where the desired number of hidden layer appears continually is expanded for higher precision. Finally, the data which have been stored in meter cabinet were used to verify the cost forecasting model. The results showed that the method can predict the trend of mineral processing cost in real time.