基于多源信息融合的浮选过程软测量建模方法研究

发布时间：2018-06-10 21:29

本文选题：浮选过程 + 软测量　；参考：《辽宁科技大学》2015年硕士论文

【摘要】：由于浮选生产过程是一个复杂的物理化学综合反应过程,它具有强非线性、强耦合性等特点,因此精矿品位和浮选回收率很难在线实时的获取。本文提出采用人工神经网络和软测量建模相结合的方法对精矿品位和浮选回收率进行预测。本文具体工作主要有以下几个方面:以浮选过程的精矿品位和浮选回收率预测为目标,提出了一种基于PSO-GSA算法优化的浮选过程前馈神经网络软测量模型。万有引力算法虽然具有较好的寻优能力,但是其收敛速度较慢,并且容易陷入局部最优。本章利用粒子群算法优化万有引力算法中的速度和位置,从而提高收敛速度和预测精度。最后运用所提算法优化前馈神经网络软测量模型参数,并对浮选过程的关键工艺技术指标进行预测和仿真。其次提出了一种基于浮选泡沫图像特征提取的混洗布谷鸟搜索算法优化BP神经网络软测量模型。由于浮选泡沫图像中包含大量有关浮选过程的信息,因此针对浮选泡沫图像的颜色、视觉和形状共14个参数进行特征提取,以作为精矿品位软测量模型的输入变量;并采用等距映射方法对高维输入向量进行降维,降低BP神经网络的输入维数和网络规模;最后提出一种自适应步长的混洗布谷鸟搜索算法优化BP神经网络软测量模型,并对该模型进行仿真验证。最后提出一种基于改进萤火虫优化算法的浮选过程回声状态网络软测量模型。将浮选过程数据和从浮选图像数据提取的图像特征信息共同作为软测量模型的辅助变量,并采用核主元分析方法对高维输入向量进行降维,提取非线性主元,以降低ESN的目标维数和网络规模;然后采用基于拥挤度因子的GSO算法对浮选过程ESN软测量模型进行优化,并对精矿品位和浮选回收率进行预测仿真。总之,通过仿真结果表明了三种神经网络软测量模型均能够取得较好的预测效果,能够提高浮选过程中精矿品位和浮选回收率的预测精度,可以满足浮选生产过程的控制要求。
[Abstract]:Because the flotation process is a complex physical and chemical synthesis reaction process, it has the characteristics of strong nonlinearity and strong coupling, so the concentrate grade and the flotation recovery rate are difficult to obtain on line. In this paper, a combination of artificial neural network and soft measurement modeling is proposed to predict the grade of concentrate and the recovery rate of flotation. The main work of this paper is as follows: in order to predict the concentrate grade of flotation process and the recovery rate of flotation, a soft sensing model of feedforward neural network based on the optimization of PSO-GSA algorithm is proposed. Although the universal gravitational algorithm has good optimization ability, its convergence speed is slow and it is easy to fall into the process. In this chapter, the particle swarm optimization is used to optimize the speed and position of the gravitational force algorithm, thus improving the convergence speed and prediction accuracy. Finally, the proposed algorithm optimizes the parameters of the soft sensing model of the feedforward neural network, and pretests and simulate the key technological parameters of the flotation process. Secondly, a flotation bubble is proposed. The BP neural network soft measurement model is optimized by the mixed washing cuckoo search algorithm. The flotation foam image contains a lot of information about the flotation process, so the color, the vision and the shape of the floatation foam image are extracted with 14 parameters to make the input variable for the soft measurement model of the concentrate grade. An isometric mapping method is used to reduce the dimension of high dimensional input vector and reduce the input dimension and network size of BP neural network. Finally, a kind of adaptive step size mixed washing cuckoo search algorithm is proposed to optimize the soft sensing model of BP neural network, and the model is verified by simulation. Finally, a new algorithm based on improved firefly optimization algorithm is proposed. The soft measurement model of the process echo state network is selected. The flotation process data and the image feature information extracted from the floatation image data are used as auxiliary variables of the soft measurement model, and the kernel principal component analysis method is used to reduce the dimension of the high-dimensional input vector, and the nonlinear principal element is extracted to reduce the target dimension and network size of ESN; then, it is used to reduce the dimension of the target and the network size. Based on the GSO algorithm of crowding degree factor, the ESN soft measurement model of flotation process is optimized, and the concentrate grade and flotation recovery rate are predicted and simulated. In conclusion, the simulation results show that three neural network soft sensing models can achieve better prediction effect, and can improve the concentrate grade and flotation recovery rate in the process of high flotation. The prediction accuracy can meet the control requirements of flotation production process.
【学位授予单位】：辽宁科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TD923;TP18;TP391.41

【参考文献】