基于改进神经网络的粉尘浓度软测量研究

发布时间：2018-06-07 03:48

本文选题：粉尘静电信号 + 新LMS算法　；参考：《山东科技大学》2017年硕士论文

【摘要】：本论文针对传统粉尘检测中存在的问题,提出一种基于神经网络软测量的粉尘浓度检测新方法,对粉尘的研究具有重要的现实意义。首先,设计了一种新的变步长LMS算法来修改滤波器系数,更好地提高了低信噪比下算法的收敛性和稳态性能,实验和仿真结果证明该算法能很好地滤除静电信号中的随机噪声。其次,对滤波后的静电信号进行时频分析,提出一种改进相似极值的EEMD信号特征提取方法,并和EMD算法进行了仿真对比,通过静电信号的能量和能量熵值分析出静电信号的变化和粉尘浓度的变化趋势具有正相关性。然后,分别建立了 BP和RBF两种神经网络粉尘浓度软测量模型,根据实验和经验相结合的方法确定了两种模型的参数,构建出综合性能最优的模型。并通过分析比较和实验验证了 BP网络模型在训练精度与泛化能力方面略优于RBF网络模型。最后,采用改进遗传算子的遗传算法对BP神经网络粉尘浓度软测量模型进行优化和改进。实验和仿真结果证明模型优化后其收敛性能、运行时间和均方误差均得到了改善,提高了模型的精确率和效率,证明了基于神经网络的软测量模型对粉尘浓度进行监测的可行性。
[Abstract]:Aiming at the problems in traditional dust detection, a new method of dust concentration detection based on neural network soft sensing is proposed in this paper, which is of great practical significance to the study of dust. Firstly, a new variable step size LMS algorithm is designed to modify the filter coefficients, which improves the convergence and steady-state performance of the algorithm under low SNR. The experimental and simulation results show that the algorithm can effectively filter the random noise in electrostatic signals. Secondly, the time-frequency analysis of the filtered electrostatic signal is carried out, and an improved EEMD signal feature extraction method with similar extremum is proposed and compared with the EMD algorithm. By analyzing the energy and energy entropy of electrostatic signal, it is found that there is a positive correlation between the change of electrostatic signal and the change trend of dust concentration. Then, two kinds of BP and RBF neural network soft sensor models of dust concentration are established, and the parameters of the two models are determined according to the method of combining experiment and experience, and the optimal comprehensive performance model is constructed. The BP neural network model is better than the RBF network model in training accuracy and generalization ability. Finally, the improved genetic operator genetic algorithm is used to optimize and improve the BP neural network soft sensor model of dust concentration. Experimental and simulation results show that the convergence performance, running time and mean square error of the model are improved, and the accuracy and efficiency of the model are improved. The feasibility of monitoring dust concentration by soft sensor model based on neural network is proved.
【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TP274

【参考文献】