摘要
采用硬件的冗余配置是提高可靠性的传统方法,虚拟传感器可实现对汽轮发电机组监测诊断系统中参数测点的软件冗余。该文对利用基神经网络建立虚拟传感器数学模型的几个问题进行了讨论,其中包括基神经网络的选择,采用遗传算法对输入参数的组合进行优选。通过归一化、加入白噪声等方法优选出训练样本,以及对虚拟传感器扩展性和可移植性等进行研究,为虚拟传感器的工程应用提供可借鉴的数值经验。
Traditionally, redundant hardware is used to enhance the reliability of steam turbine unit monitoring and diagnosis system. As redundant software, a virtual sensor is possible to be used in an effort to insure reliability In this paper some issues concerning a virtual sensor based on a neural network are discussed. The issues include selection of the type of neural network, the training parameter optimization, selecting training samples by normalization and flat noise, the expansibility and applicability of virtual sensor, as well as others. The discussion results provide some valuable parameters and experiences for engineering applications of virtual sensors.
出处
《中国电机工程学报》
EI
CSCD
北大核心
2004年第7期253-256,共4页
Proceedings of the CSEE
关键词
汽轮发电机组
监测诊断系统
虚拟传感器
数学模型
神经网络
Thermal power engineering
Steam turbine unit
Virtual sensor
Mathematical model
Neural network
Genetic algorithm