Optimal Sensor Placement based on Gaussian Process Regression for Shared Office Spaces under Various Ventilation Conditions

Energy efficiency and thermal comfort are the most important objectives for building operation. While energy efficiency is at least partly implemented in operation concepts, individual thermal comfort is commonly not taken into account for commercial buildings due to missing sensors and actuators or suitable control approaches. A first step towards individualized comfort control needs a detailed room model to capture the various conditions within a large occupied space. In this paper a Gaussian process regression model is developed for the prediction of the stationary temperature profile in an exemplary open-plan office. Since sensor count and positions have a significant impact on the achievable prediction accuracy, optimal placement strategies are evaluated based on computational fluid dynamics simulations. Two objective functions - maximizing Mutual Information and minimizing the prediction error - are considered and compared. The differences between the approaches and their effect on the model quality are analyzed and discussed regarding their feasibility for real-world applications. Fewer requirements on the initial system knowledge suggest the usage of Mutual Information as an objective for optimizing the sensor setup, although the simpler application and higher robustness are accompanied by a slightly increased prediction error.