基于符号回归的静电电压测试补偿算法  

Compensation algorithm for electrostatic voltage test based on symbolic regression

作　　者：冯亚辉 满梦华 魏明 FENG Yahui;MAN Menghua;WEI Ming(Key Laboratory on Electromagnetic Environmental Effects,Shijiazhuang Campus of Army Engineering University,Shijiazhuang 050003,China)

机构地区：[1]陆军工程大学石家庄校区电磁环境效应重点实验室,河北石家庄050003

出　　处：《现代电子技术》2023年第8期15-20,共6页Modern Electronics Technique

基　　金：国防科技重点实验室基金(6142205190101)。

摘　　要：为解决非接触式静电电位传感器测量数据的补偿问题,文中提出一种基于符号回归的波形补偿算法。设计一种非接触式静电电位传感器校准实验平台,通过遍历测试距离、校准电压和校准频率的参数组合,获得测试结果的样本训练数据;再利用符号回归方法建立从测试结果到校准电压的补偿函数模型,以降低测量距离对测量结果的影响。实验结果表明,基于符号回归的波形补偿算法可行、有效,能够为提高非接触式电位测量的准确性提供一种新手段。另外,基于模型参数与补偿结果,提出一种符号回归的改善方法,即引入损失函数用来减小函数模型的复杂度,利用神经网络卷积学习的方法发现种群与原始数据中的隐藏特征,从而明确演化方向,并针对特定的问题设定特定的评价指标,进一步优化符号回归模型的准确性。A waveform compensation algorithm based on symbolic regression is proposed to slove the problem of compensating the measured data of contactless electrostatic potential sensors.A calibration experiment platform for contactless electrostatic potential sensor is designed.By traversing parameter combinations of testing distance,calibration voltage,and calibration frequency,sample training data of testing results are obtained.A compensation function model from testing results to calibration voltage is obtained by means of the symbolic regression method,which greatly reduces the impact of measurement distance on measurement results.The experimental results verify the effectiveness of the waveform compensation based on the symbolic regression algorithm and provide a new means to improve the accuracy of the non⁃contact potential measurement.Based on the model parameters and the compensation results,the improvement method of symbolic regression is proposed,namely introducing the loss function to decrease the complexity of the function model,using the neural network convolution learning method to find the hidden features in the population and raw data,thus clarifying the evolution direction and set specific evaluation indicators for specific problems can further optimize the accuracy of the symbolic regression model.

关 键 词：符号回归 静电电位传感器 电位测量 波形补偿 距离校准 非接触式 样本训练 

分 类 号：TN911.23-34[电子电信—通信与信息系统]