Occupancy Prediction for Building Energy Systems with Latent Force Models

TL;DR

This work tackles occupancy-driven energy optimization in building energy systems by modeling occupancy as a latent disturbance with Gaussian Processes and embedding it in a Latent Force Model. The GP is converted to a state-space form, enabling Kalman-filter-based estimation and forecasting that feed an MPC controlling building energy use. The approach is validated on EnergyPlus and a Bosch Renningen campus dataset, showing energy reductions (up to ~14% with Pre-Comfort modes) and improved or comparable thermal comfort relative to occupancy-aware baselines, with zone-dependent variations. The results demonstrate the practical potential of LFM-MPC to leverage occupancy patterns for more efficient, comfortable operation in real and simulated BES, while highlighting challenges in irregular occupancy zones and the need for kernel/adaptation strategies. Overall, the work advances occupancy-aware MPC by combining data-driven learning with mechanistic dynamics in a scalable state-space framework.

Abstract

This paper presents a new approach to predict the occupancy for building energy systems (BES). A Gaussian Process (GP) is used to model the occupancy and is represented as a state space model that is equivalent to the full GP if Kalman filtering and smoothing is used. The combination of GPs and mechanistic models is called Latent Force Model (LFM). An LFM-based model predictive control (MPC) concept for BES is presented that benefits from the extrapolation capability of mechanistic models and the learning ability of GPs to predict the occupancy within the building. Simulations with EnergyPlus and a comparison with real-world data from the Bosch Research Campus in Renningen show that a reduced energy demand and thermal discomfort can be obtained with the LFM-based MPC scheme by accounting for the predicted stochastic occupancy.

Figures (5)

• Figure 1: Exemplary control loop for the LFM-MPC. The observer provides state estimates for the mechanistic model and the latent states, which are used to compute the estimated disturbance trajectory $\hat{d}(\tau)$.
• Figure 2: 3D view of the building used for the EnergyPlus test case.
• Figure 3: Zone layout of the considered part of the R&D building at the Bosch Research Campus. Only zone 9 and zone 21 to 24 are considered.
• Figure 4: Energy reduction and thermal discomfort for the EnergyPlus test case. The thermal discomfort is broken down to the two offices in the north and south.
• Figure 5: Energy reduction and thermal discomfort for the Renningen test case.