基于RBF神经网络的磨矿粒度软测量模型及实现

Soft Sensor Model Based on RBF Neural Network for Particle Size of Grinding Circuit and Implementation

摘要:

磨矿分级作业是选矿生产中的重要环节,磨矿粒度的好坏直接影响浮选的精矿品位和尾矿回收率；在实际生产中粒度的测量有在线粒度分析仪,但存在成本高、维修率高,离线实验室化验又有时间延迟大的问题；对实际磨矿分级作业过程进行了分析,提出用径向基函数(RBF)神经网络建立磨矿粒度软测量模型,采用正交最小二乘法(OLS)算法对网络进行训练学习,泛化校验；仿真结果表明,在较少训练数据下该网络非线性处理能力和逼近能力依然很强,学习时间短,模型基本符合要求；通过OPC技术将Matlab与PKS控制系统相结合,实现实时软测量磨矿粒度。

Abstract:

Grinding grading operation is a key link in mineral processing production, and particle size of grinding circuit directly affect the grade of concentrate and the recovery rate of tailings in the process of flotation. In the actual production, the measurement of particle size, one side we can use online particle size analyzer, but it’s too high cost and high maintenance rate; in other side we can measure it in off-line laboratory tests, but it’s too large time delay. Through the analysis of the actual grinding grading operation process, put forward a soft sensor model of grind size with RBF neural network and training this network by using orthogonal least squares learning algorithm. The simulation results show that the network under less training data is still very strong nonlinear processing ability and approaching ability, learning time is short, model conforms to the basic requirements. Through the OPC technology can contact the Matlab and PKS control system, which can realize real-time soft sensor particle size of grinding circuit.