报告详情
  1. 活动首页
  2. 报告清单
  3. 报告详情
Delayed Feedback Adaptation of Virtual Sensor Reliability Under Recurrent Target Outages
编号：166 访问权限：仅限参会人 更新：2026-10-19 18:04:32 浏览：17次 口头报告

报告开始：暂无开始时间（Asia/Shanghai）

报告时间：暂无持续时间

所在会场：[暂无会议] [暂无会议段]

暂无文件

摘要
When a physical target becomes unavailable, a virtual sensor can continue predicting, but current truth and realized error are also unavailable. Runtime reliability must therefore be estimated from target-free evidence rather than verified during the outage. We study a frozen random-forest (RF) virtual sensor whose separate reliability mapping combines RF tree dispersion with Mahalanobis out-of-distribution (OOD) distance. After target recovery, delayed truth can update this evidence-to-reliability mapping for future outages only, without retraining the predictor or revising completed outputs. On the Tetouan City Power Consumption dataset, we evaluated recurrent simulated target outages in a held-out future region using predeclared schedules and experimental tolerances. Compared with static reliability, direct evidence adaptation reduced the equal-tolerance Brier score by 0.0228 and empirical area under the risk-coverage curve by 0.1957; both schedule-bootstrap 95% confidence intervals excluded zero. Improvements were concentrated at medium and relaxed tolerances, whereas strict-tolerance Brier score and AURC worsened. Delayed recovery feedback can therefore improve aggregate probabilistic reliability estimation and selective use without retraining the virtual sensor, but its benefit is tolerance dependent.
 
关键词
Virtual sensing,Reliability estimation,Delayed feedback,Target outage,Selective prediction
报告人
GEZHI ZHAO
Master's Student Multimedia University

稿件作者
GEZHI ZHAO Multimedia University
Yi-Fei Tan Multimedia University
Hezerul Bin Abdul Karim Multimedia University
CHING KING CHIA TM/TMR&D
发表评论
验证码 看不清楚，更换一张
全部评论
重要日期

  • 会议日期

    11月06日

    2026

    至

    11月08日

    2026

  • 10月15日 2026

    初稿截稿日期

  • 11月10日 2026

    注册截止日期

主办单位
IEEE Instrumentation and Measurement Society
承办单位
Sichuan University
联系方式
登录查看完整联系方式
历届会议
移动端
在手机上打开
小程序
打开微信小程序
客服
扫码或点此咨询