Comparing Building Thermal Dynamics Models and Estimation Methods for Grid-Edge Applications

TL;DR

This paper evaluates and compares the accuracy and efficiency of RC-network and structured regression models, along with various parameter estimation methods, for modeling building thermal dynamics in grid-edge applications.

Contribution

It introduces a comparative analysis of different grey-box modeling approaches and estimation techniques for building thermal dynamics.

Findings

RC-network models with MLE perform best on simulated data.

Structured regression models with Almon Lag structure show competitive accuracy.

Parameter estimation methods significantly impact model performance.

Abstract

We need computationally efficient and accurate building thermal dynamics models for use in grid-edge applications. This work evaluates two grey-box approaches for modeling building thermal dynamics: RC-network models and structured regression models. For RC-network models, we compare parameter estimation methods including Nonlinear Least Squares, Batch Estimation, and Maximum Likelihood Estimation. We use the Almon Lag Structure with Linear Least Squares for estimating the structured regression models. The performance of these models and methods is evaluated on simulated house and commercial building data across three different simulation types.