基于预测补偿网络的锌扫选尾矿品位预测

Grade prediction of zinc scavenging tailings based on predictioncompensation network

针对泡沫浮选关键性能指标预测准确率低的问题,提出一种基于预测补偿(PC)网络的品位预测方法。该预测补偿网络分为两部分:第一部分,构建基于GRU的锌浮选尾矿品位预测模型,充分利用泡沫图像的时序信息,得到初始品位预测值;第二部分,为解决未知样本的输入输出难以精确匹配的问题,建立由残差诱因导出模块和改进Choquet模糊积分(ICFI)聚合模块组成的动态残差补偿(DRC)模型,对初始品位进行补偿以获取更精确的结果。研究结果表明:相较于传统的神经网络,所提出的预测补偿网络有更强的拟合能力和稳定性,提高了预测精确性和可靠性。

Aiming at the problem of low prediction accuracy of key performance indicators of froth flotation,a grade prediction method based on prediction-compensation(PC)network was proposed.The prediction-compensation network was divided into two parts.The first part constructed a GRU-based zinc flotation tailings grade prediction model,which made full use of the time series information of the froth image to obtain the initial grade prediction value.In the second part,in order to solve the problem that the input and output of unseen samples were difficult to map accurately,a dynamic residual compensation(DRC)model composed of the residual trigger derivation module and the improved Choquet fuzzy integration(ICFI)aggregation module was established to compensate for the initial grade prediction value to obtain more accurate results.The results show that compared with the traditional neural network,the proposed prediction-compensation network has better fitting ability and stability,and improves the prediction accuracy and reliability.