Optimization of indoor temperature sensor layout using proper orthogonal decomposition and Greedy algorithm

Accurately estimating the distributed indoor thermal environmental parameters with limited sensors is crucial for indoor environmental quality and building energy efficiency. This study combines proper orthogonal decomposition and greedy algorithm for sensors layout’s optimization in a large-space thermal environment. Firstly, choose the initial quantity and positions of sensors, and collect steady temperature field information based on the feasible range of environmental variables; Secondly, extract features from the collected dataset using proper orthogonal decomposition, and determine the optimal number of sensors based on the energy proportion of the eigenvalues; Thirdly, select the sensor positions iteratively based on the correlations of eigenvectors and sensors using greedy algorithm. A field experiment for performance validation is conducted using a matrix of 72 temperature sensors in a large cafeteria. By using the collected 27 snapshot datasets, the temperature field can be reconstructed by Linear Stochastic Estimation using only six optimal sensors (steady-state error: 0.2433, RMSE). The proposed POD-Greedy optimization strategy is also compared with another heuristic inference method. The better performance shows great potential for engineering practice and applications.