基于分布式多子网神经网络的传感器静态误差修正

A Method Based on Distributed Multi-Subnet Neural Network for Static Error Correction of Transducer

为消除温度等非目标参数对传感器输出特性的影响 ,本文提出了一种采用多传感器数据融合技术的传感器静态误差修正新方法———传感器冗余法。该方法的显著优点是系统不需测量非目标参数 ,通过对多种传感器输出数据融合达到消除非目标参数对传感器输出的影响 ,实现正确检测传感器的输入。同时本文首次提出了一种新的神经网络结构 :分布式多子网神经网络 ,并将其用于实现该修正方法。模拟实验表明 ,该网络在泛化能力、学习速度等方面均优于BP网络和RBF网络。其多子网、自动分解任务的特点尤其适用于复杂样本的学习 ,具有很好的应用前景。

A new method based on multi sensor information fusion for static error correction of transducer is proposed in this paper. In comparison with the traditional method, the most significant advantage of the proposed method is the fact that the untargeted parameters measurement are not required for correction. The application of a distributed multi subnet neural network makes it easy for realizatin of the proposed method. The simulation result shows that it achieves better performance than BP or RBF network on generalization ability and training speed etc. The particular of dividing a complex task automatically by subnet, into a series of subtask which are processed by their corresponding subnet allows it owning a better performance for complex task learning.