Improving Indoor Occupancy Prediction using Graph Neural Networks and Positional Encodings

Yu Sheng

doi:10.1016/j.enbuild.2026.118077

Improving Indoor Occupancy Prediction using Graph Neural Networks and Positional Encodings

2026-08-11

0

OA

AI

Y

Yu Sheng *

A

Ali Değer Özbakır

D

Deniz İren

C

Clara Maathuis

S

Stefano Bromuri

DOI：10.1016/j.enbuild.2026.118077

Original

Original request for help

Save

Abstract

En 中文

• Compare PE and GNN-based spatial representations for encoding room connectivity. • Use real-home multi-sensor data to predict whether five rooms will be occupied within the next six hours. • Show that sensor placement and label reliability strongly affect room-level performance.

Keywords:

occupancy prediction

spatial embedding

graph neural networks

energy management

Improving Indoor Occupancy Prediction using Graph Neural Networks and Positional Encodings

Abstract

Journal

Organization

Cited Papers

Citing Papers