基于树突神经网络的MEMS压力传感器误差补偿方法  

Dendritic Neural Network-Based Error Compensation Method for MEMS Pressure Sensors

作　　者：赵月爱[1] 白渊铭 ZHAO Yueai;BAI Yuanming(College of Computer Science and Technology,Taiyuan Normal University,Jinzhong Shanxi 030619,China)

机构地区：[1]太原师范学院计算机科学与技术学院,山西晋中030619

出　　处：《传感技术学报》2023年第1期93-98,共6页Chinese Journal of Sensors and Actuators

基　　金：国家自然基金项目(61273294);国家社科基金项目(20BJL080);山西省重点研发计划项目(201803D121088)。

摘　　要：压力传感器的性能因自身工艺或环境温度变化而造成影响,在实际应用中,因环境温度的改变而造成的传感器输出误差尤为突出。现今微机电系统(MEMS)压力传感器在线性度与灵敏度方面较之前已有了很大提高,但温度漂移问题依然存在。传统的误差补偿方法如线性回归模型、BP神经网络或BP神经网络的优化算法,均能对传感器的误差有修正效果,但还有提升空间。针对以上问题,在MEMS压力传感器中嵌入温度传感器模块,并构建了一种基于树突神经网络的误差补偿模型。该模型首先对采集数据进行逻辑组合关系预处理,然后用树突神经网络对传感器数据进行误差补偿。实验结果表明,使用数据逻辑关系预处理后的BP神经网络模型评估指标平均绝对误差(MAE)由9.1降至0.191,而用树突神经网络模型后,该指标更是降低至0.043 48,精度提升效果非常明显,证明所提方法能够有效地补偿MEMS压力传感器误差。The performance of pressure sensors is affected by their manufacturing processes or by changes in ambient temperature, especially, the sensor output error caused by the change of ambient temperature is significant in practical application. Nowadays, the linearity and sensitivity of MEMS pressure sensors have been greatly improved, but the problem of temperature drift still exist. Traditional error compensation methods such as linear regression models, BP neural networks or optimization algorithms of BP neural networks can correct the error of the sensor, but there are still room for improvement. To address the above problems, a temperature sensor module is embedded in the MEMS pressure sensor, and an error compensation model based on dendritic neural network is proposed. Firstly, the collected data are pre-processed for logical combination relationship, and then, the sensor data error is compensated by using dendritic neural network. The experimental results show that the model evaluation index of mean absolute error(MAE)is reduced from 9.1 to 0.191 by using the BP neural network model after data logical relationship preprocessing, and the MAE is reduced to 0.043 48 by using the dendritic neural network, dewonstrating that the precision improvement effect is very obvious. The proposed method can effectively compensate the MEMS pressure sensor errors.

关 键 词：压力传感器 误差补偿 数据逻辑组合 树突神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]