摘要
对流行的神经网络算法和无学习率的神经网络算法做了比较.流行的人工神经网络算法在误差反演过程中需要加入学习率,依次减少误差,逐渐逼近正确的拟合多项式,计算精度很高.无学习率的神经网络算法在进行权值调整时不需要加入学习率,减少了计算量,增加运算速度,计算精度也很高.它们可以应用于传感器信号处理中,流行的神经网络算法适用于压力传感器的温度补偿,无学习率的神经网络算法可用于对范德堡函数多项式拟和.
A comparison between the popular neural-network algorithm and a new neural-network algorithm without learning rate is introduced. The former has to use learning rate to reduce the error scale gradually in its iterative processing until getting a right fitting polynomial. Therefore, it has a more perfect precision. The latter, however, dose not use learning rate in the adjustment of weight value. As a result, the amount of calculation decreases, and the operating speed increases. It has also a more perfect precision. Both are all used for the Signal Processing in Sensors. For example： the former is used for the temperature compensation for pressure sensors, the latter is used for the polynomial fitting for Van der Pauw s＇ function.
出处
《河北工业大学学报》
CAS
北大核心
2010年第1期56-61,共6页
Journal of Hebei University of Technology
关键词
BP神经网络算法
流行的神经网络算法
温度补偿
学习率
范德堡多项式
back propagation neural-net algorithm
popular neural-net algorithm
temperature compensate
learning rate
the polynomial of Van der Pauw＇s function
