基于增强型BP网络的气象传感器标校方法  

Calibration method of meteorological sensor based on enhanced BP network

作　　者：王彦明 贾克斌 刘鹏宇 杨加春 WANG Yanming;JIA Kebin;LIU Pengyu;YANG Jiachun(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Beijing Laboratory of Advanced Information Networks,Beijing 100124,China;Tianjin Huayuntianyi Special Meteorological Detection Technology Co.,Ltd.,Tianjin 300392,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]先进信息网络北京实验室,北京100124 [3]天津华云天仪特种气象探测技术有限公司,天津300392

出　　处：《中国测试》2020年第12期105-111,共7页China Measurement & Test

基　　金：国家重点研发计划资助项目(2018YFF01010100);国家自然科学基金资助项目(61672064)。

摘　　要：针对气象传感器标校过程中测量精度低和生产成本高的问题,将人工智能技术与单片机技术相结合,提出一种气象传感器智能标校方法。该方法将BP神经网络、高斯函数和Levenberg-Marquardt算法相融合,设计一种用于传感器标校的增强型BP网络模型。并将训练好的标校模型移植到单片机中,通过分段多项式来拟合高斯函数,有效减少单片机的计算资源、缩短计算时间。实验结果表明:传统BP网络使气压传感器均方根误差由最初的5.93降低到2.83,减少52.28%的测量误差;而增强型BP网络则使均方根误差降低到0.77,进一步减少34.74%的测量误差。通过分段多项式来拟合高斯函数,显著降低标校模型的计算量,可满足气象探测过程中的时间要求。Aiming at the problems of low measurement accuracy and high production cost in the process of meteorological sensor calibration,this paper proposes an intelligent calibration method of meteorological sensor by combining artificial intelligence technology and microcontrollers technology.By combining back propagation(BP)neural network,Gaussian function and Levenberg-Marquardt(LM)algorithm,an enhanced BP network is realized for sensor calibration.The calibration model is transplanted to microcontroller unit(MCU).The Gaussian function is fitted by piecewise polynomials,which effectively reduces the computing resources and time of MCU.The experimental result show that:The traditional BP network reduces the mean squared error(MSE)of the atmospheric pressure sensor from 5.93 to 2.83,which reduces the measurement error by 52.28%.The enhanced BP network reduces the MSE to 0.77,which further reduces the measurement error by 34.74%.By fitting the Gaussian function with piecewise polynomial,the complexity of calibration model is significantly reduced,and the time requirement in the process of meteorological observation is met.

关 键 词：传感器标校 BP神经网络 高斯函数 LEVENBERG-MARQUARDT 单片机 

分 类 号：TP212.1[自动化与计算机技术—检测技术与自动化装置]