Physics informed neural networks for control oriented thermal modeling of buildings

Buildings constitute more than 40% of total primary energy consumption worldwide and are bound to play an important role in the energy transition process. To unlock their potential, we need sophisticated controllers that can understand the underlying non-linear thermal dynamics of buildings, consider user comfort constraints and produce optimal control actions. A crucial challenge for developing such controllers is obtaining an accurate control-oriented model of a building. To address this challenge, we present a novel, data-driven modeling approach using physics informed neural networks. With this, we aim to combine the strengths of two prominent modeling frameworks: the interpretability of building physics models and the expressive power of neural networks. Specifically, we use measured data and prior information about building parameters to realize a neural network model that is guided by building physics and can model the temporal evolution of room temperature, power consumption as well as the hidden state, i.e., the temperature of building thermal mass. The main research contributions of this work are: (1) we propose two new variants of physics informed neural network architectures for the task of control-oriented thermal modeling of buildings, (2) we show that training these architectures is data-efficient, requiring less training data compared to conventional, non-physics informed neural networks, and (3) we show that these architectures achieve more accurate predictions than conventional neural networks for longer prediction horizons (as needed for effective control). We test the prediction performance of the proposed architectures using both simulated and real-word data to demonstrate (2) and (3) and argue that the proposed physics informed neural network architectures can be used for control-oriented modeling.

Automated summary

This paper proposes a data-driven modeling approach using physics informed neural networks for control-oriented thermal modeling of buildings. By utilizing measured data and prior information, this approach accurately predicts the temperature, power consumption, and hidden state of building thermal mass, achieving higher prediction accuracy and data efficiency compared to conventional neural networks.