基于机器学习的复杂环境下APD最优偏置电压补偿方法  

APD Optimal Bias Voltage Compensation Method Based on Machine Learning in Complex Environment

作　　者：陈梦强 杨家志[1,2] 于广旺 沈洁 CHEN Mengqiang;YANG Jiazhi;YU Guangwang;SHEN Jie(College of Information Science and Engineering,Guilin University of Technology,Guilin 541006,Guangxi,China;Guangxi Key Laboratory of Embedded Technology and Intelligent System,Guilin 541006,Guangxi,China)

机构地区：[1]桂林理工大学信息科学与工程学院,广西桂林541006 [2]广西嵌入式技术与智能系统重点实验室,广西桂林541006

出　　处：《实验室研究与探索》2023年第1期147-152,共6页Research and Exploration In Laboratory

基　　金：国家自然科学基金项目(41961065);广西创新驱动发展专项项目(桂科AA18118038);广西科技基地和人才专项(桂科AD19254002)。

摘　　要：雪崩光电二极管(APD)在激光雷达探测系统中的信噪比受工作距离和背景辐射的影响较大,传统探测方法通过离线式或事先根据外界环境影响因素进行预补偿,不能在线动态调节,难以适应复杂环境。一种基于机器学习的APD偏置电压最佳补偿方法,可以准确判断APD当前的工作状态,对偏置电压进行二分补偿,使APD工作在最优状态。通过比较多种机器学习模型,选用准确率维持在98%以上的随机森林算法来判断APD工作状态。在不同距离下测试,该方法可保证偏置电压始终处于最优工作电压下,提高激光雷达探测系统的性能。The signal-to-noise ratio of avalanche photodiode(APD)in lidar detection system is greatly affected by working distance and background radiation.The traditional method cannot adapt to complex environment by off-line compensation or pre-compensation according to the influence factors of external environment.In this paper,an APD bias voltage optimal compensation method based on machine learning is designed,which can accurately judge the current working state of APD,so that the bias voltage can be compensated in two ways to make APD work in the optimal state.By comparing various machine learning models,the random forest algorithm with 98%accuracy is selected to judge the working state of APD.Finally,under different distances,the method can ensure that the bias voltage is always under the optimal operating voltage,so as to improve the performance of the lidar detection system.

关 键 词：雪崩光电二极管 机器学习 偏置电压补偿 二分法补偿 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]