基于神经网络的超声波流速测量模型

来源期刊：中南大学学报(自然科学版)2003年第4期

论文作者：黄挚雄 罗安 杨勇

文章页码：394 - 397

关键词：神经网络;模型仿真;超声波;多普勒频率;流速

Key words：neural network; model simulation; ultrasonic wave; Doppler frequency; flow velocity

摘    要：传统的超声多普勒测流采用线性定斜率K的方法,不能准确反映水流随环境变化的多变性.应用人工神经网络的建模方法,采用多层感知器的模型结构和自适应学习速率的BP学习算法,建立了基于超声多普勒频率测量流速的功能模型,并将辨识模型的仿真结果与系统实验测量数据比较,以检验神经网络模型的可靠性.仿真实验采用不同性能的实测数据作为训练样本,通过训练得出了流速测量神经网络模型的结构和参数,以及网络输出均方误差曲线,并作出了垂线流速分布图.实验结果表明:该流速测量建模方法具有较高的精度和较强的适应性,能更好地反映系统输入输出之间的非线性关系,模型的应用符合水力学规律.

Abstract: Traditional method to measure flow velocity by ultrasonic Doppler frequency used linear slope K. It does not exactly reflect the changeability of flow in different environment. The functional model measuring flow velocity by ultrasonic Doppler frequency is built in this paper. The model is a multi-layered artificial neural network that employs adaptive BP learning algorithm. The comparison between the new method and traditional way of system identification is discussed, mainly by comparing the simulated result of model with the data measured in experiment, so as to check the reliability of neural network. Structure and parameters of neural network model for flow measurement are obtained by simulated experiment, and vertical distribution of flow velocity is drawn. The result shows this new way to build the model of flow velocity measurement has high accuracy and strong adaptability. It perfectly reflects the non-linear relationship between input and output of system. The application of the model conforms to hydraulic laws.