基于灰色神经网络的传感器分段标定

针对大量程高精度传感器不能一次完成标定实验的情况，提出一种将优化灰色GM(1,1)模型与BP神经网络相结合来预测分段标定过程中特征值缺失的方法，从而实现传感器的分段标定。首先，根据实验数据建立传统灰色GM(1,1)模型，对待标定传感器和标准传感器的测量值进行缺失数据的预测；然后，为弱化传统灰色GM(1,1)模型序列变化的幅度，提高模型的预测精度，利用中心逼近的思想对传统的GM（1,1）模型进行优化；最后，利用BP神经网络对优化的灰色GM(1,1)残差序列进行修正，以较高的精度实现对分段标定过程中缺失特征值的预测。结果表明，待标定传感器和标准传感器组合预测模型的平均残差分别为0.023%和0.401%，证明了组合预测模型的有效性。所提出方法为解决大量程高精度传感器分段标定时静态特性曲线的拟合提供了一种新思路。

Segmented Calibration of Transducer Based on Grey Neural Network

As large range and high precision transducer could not complete the calibration with just one experiment, an integrated modeling method was proposed, which incorporated optimized grey GM(1,1) and BP neural network to predict the missing values in calibration, and the segmented calibration of transducer was realized. Firstly, according to experimental data, traditional grey GM(1,1) model was established to predict the missing values, which were measured by both calibrated transducer and standard transducer. In addition, in order to weaken the scope of the sequence and improve mode prediction accuracy, the idea of center approach was used to optimize traditional grey GM(1,1) model. Finally, BP neural network was applied for modifying the residuals of optimized grey GM (1,1), realizing the prediction of the missing values in calibration with a high accuracy. The results show that the residual mean of the combined model of calibrated and standard transducer are 0.023% and 0.401% respectively, the effectiveness of the combined predicting model is proved, so it can be used to predict the missing values for the segmented calibration of transducer, and a new method is proposed to solve characteristic curve fitting problem, which is related to segmented calibration of large range and high precision transducer.