基于神经网络的传感器输出特性样本集研究

研究了压力传感器输入/输出特性样本集结构和样本集的实时校正。首先,通过加速试验获得了传感器特性的时漂以及受温度影响的变化规律,并据此构建了样本集。基于BP神经网络模型,对样本集的融合精度进行了动态对比试验,进而验证了样本集构建的合理性。此外,提出一种对样本集进行实时校正的方法,校正过程由程序控制。将实时校正后的数据与初始标定样本数据、加速试验600 h后的标定数据对比,最大偏差仅为0.08 kPa,样本集经校正后,数据准确度提高了近1个数量级。

Research sample set of sensor''''s intput/output characteristic based on neural network

The structure and real-time correction of sample set of pressure sensor's intput/output characteristic are studied.First,the time-drift of sensor's characteristic and its change law with temperature are obtained by the accelerated test,then the sample set is established.The dynamic contrast experiment for the fusion accuracy of sample set based on BP neural network model is made and the rationality of sample set structure is demonstrated.Furthermore,a kind of method of correcting sample set real-time is presented.The correction process is controlled by programme.Making comparison between real-time correction data and calibration data,and another calibration data which is obained after the accele-rated test in 600 h.The maximum error is only 0.08 kPa.After sample set is corrected,its accuracy is improved almost an order of magnitude.