摘要
提出了一种基于 RBF神经网络的传感器非线性校正方法。传感器的输出及待测物理量的实际数值用于训练 RBF神经网络 ,以得到非线性校正用的逆模型。只需较少的神经元就可构成上述逆模型 ,便于单片机软件实现或“固化”在硬件中。通过一个二维位移传感器的例子表明 ,采用 RBF神经网络的传感器非线性校正精度和网络训练速度均大大优于 BP神经网络 ,能满足实用要求。
The RBF neural network method for calibration of a sensor that suffers from nonlinearities is described. The RBF neural network is used as an inverse model that is trained to perform the mapping among the sensor's readings and the actual sensed properties. This application requires a relatively small number of neurons,which can be implemented in singlechip software of 'fixed' hardware. As an example we study a 2-D displacement sensor. The example demonstrates that the calibrated data precision by a RBF neural network is much better than a BP neural network,and its training speed is also much faster. This approach is valuable for practical applications.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2003年第1期96-98,共3页
Chinese Journal of Scientific Instrument
基金
浙江省自然科学基金资助项目 (No 60 2 1 4 5)
关键词
径向基函数
神经网络
传感器
校正
Radial basis function(RBF) Neural network Sensor calibration Intelligent sensor
