Conference Agenda

Space heating accounts for a large share of energy consumption in Germany due to fossil fuel combustion. To reduce combustion-related emissions, space heating electrification with heat pumps is promising. However, heat pumps increase the power demand related to the electrical grid. Therefore, minimizing the additional electrical power consumed by heat pumps is crucial, which can be achieved by either lowering the heating demand or enhancing the overall efficiency. While lowering the heating demand due to improvements to the building envelope is very cost-intensive, increasing the heat pump efficiency is substantially more cost-effective.

To achieve high efficiencies with heat pumps, low supply temperatures are required. Today, static heating curves (HCs) connect the heat pump with a specific building. Conventionally, the HC correlates the ambient temperature with the supply temperature. This method captures the fundamental principle that colder ambient conditions necessitate higher supply temperatures. While HCs are favored for their simplicity in implementation, they fail to ensure maximal heat pump efficiency because HCs are not tailored to the conditions of the specific building and its usage. Hence, a method to tailor the supply temperature to the building and its usage is promising to maximize efficiency while maintaining thermal comfort.

This paper introduces a self-optimized HC (soHC), which adapts to the building and its usage. For this purpose, a data-driven technique generates the process models in the form of artificial neural networks and gaussian process regressors. The underlying data set is generated using a conventional HC, so that the thermal discomfort during the system identification is held low. With this data, a new soHC is generated. We assess the performance of the soHC in different buildings and heating systems in annual building energy system performance simulations. The soHC outperforms the conventional HC in terms of efficiency while maintaining thermal comfort. In further research, the soHC should be extended by further specific conditions such as solar irradiation, presence detection, and domestic hot water production

Power over Ethernet (PoE) technology optimizes lighting systems by transmitting power and data transmission over the same Ethernet cable. This technology holds the promise of significant advancements in energy efficiency, cost-effectiveness, and flexibility in lighting design. Through PoE, lighting systems can be centrally controlled and managed, allowing for dynamic adjustments in brightness, color temperature, and fixture behavior. This enhances user comfort and leads to energy savings by optimizing usage. Furthermore, PoE eliminates the need for separate electrical wiring, simplifying installation and reducing upfront costs. This makes it an attractive choice for both new construction projects and retrofitting existing spaces. This paper explores the design and implementation of a PoE based DC (Direct Current) lighting system and the power savings and versatility offered in doing so. The system uses a PoE Switch or Relay (PoES) to direct the data and the power to the DC lights. PoE Controllers (LINCS) are used to supply the power to the lights themselves. Kinetic switches are used to supply a wireless and unpowered method of turning on and off the lights. Finally, the PoE Central Control (COR-TAP) is configured to receive input from the kinetic switches and issue commands to different LINCs. This system was installed in a real-world residential building for testing. Power consumption savings were quantified by comparison to simple AC lighting system. Through this comparative study, it is evident that the new system can improve lighting efficiency, connectivity, and flexibility compared to traditional AC lighting systems.

The advent of smart connected thermostats and their capability of data collection has spurred many studies exploring how occupants adjust thermostat setpoints to achieve comfort, along with the various factors influencing their setpoint preferences. The objective of this study is to investigate thermostat-adjustment behavioral patterns in households with single-stage heat pumps coupled with backup heaters and smart thermostats. A field study was carried out in 30 houses within a newly constructed residential community. The experiments consist of two parts: 1) a baseline mode featuring a heat pump paired with an auxiliary heater controlled by default thermostat heuristic rules, and 2) a comparison mode where the auxiliary heater is activated to provide the majority of heating. The findings from the field study suggest that several occupants exhibit lower setpoint preferences during the winter season for the comparison mode that has higher supply air temperatures. Furthermore, four distinct setpoint-increasing actions are identified, contributing to the setpoint differences between the two modes. Among these, the behavior associated with staging the auxiliary heater and lower heat pump capacity during cold weather conditions is the primary difference between the two operation modes. A noticeable decrease in this behavior is observed in the comparison mode.

Monitoring real-time internal and solar heat gains and indoor environmental conditions is essential for demand-driven HVAC and lighting operation in buildings. This paper presents a new framework for real-time estimation of dynamic internal and solar heat gains by employing a low-cost programmable fisheye camera and a convolutional neural network (CNN)-based multi-head classification scheme. The camera captures High Dynamic Range (HDR) images of the space, automatically converted to luminance maps. The CNN model was trained and deployed on the camera to classify real-time changes in occupancy, equipment, lighting and window status in private and open-plan offices. To balance computational load and feature extraction performance, the pre-trained weights in the original Resnet18 architecture were applied to the model with self-attention layer, which can increase the receptive field area of feature pixels. The classification precision and recall results show that the model detects the status of objects in the predefined areas of the scene with great performance and can be transferred efficiently to other spaces.

To assess the impact of dynamic internal and solar gains detection on energy demand, a model of the open-plan office utilized in the experiment was created in EnergyPlus using the followinginputs: (i) baseline schedules of internal and window gains using ASHRAE 90.1 recommendations and (ii) actual detection of dynamic occupancy levels, equipment status, light diming levels and window (shading) status using the deep learning framework. A comparison under the same external conditions showed noticeable differences in total heat gains and thermal loads when real-time internal and solar gains are monitored and quantified in real-time. The heat gains detection framework can be integrated with the Building Management System for efficient control in demand-response applications.

With the widespread adoption of building automation systems, alongside the progress in data analytics, sensing, and machine learning, the domain of data-driven fault detection and diagnostics (FDD) for building heating, ventilation, and air conditioning (HVAC) systems has attracted increasing interest. Numerous studies have tested various algorithms using diverse data sources, including simulations, laboratory testing, and real building environments. However, there is a notable gap in the literature regarding systematic benchmarking of these algorithms against each other using the same open-source datasets. In this study, we undertake a comprehensive benchmarking of different classification algorithms tailored for multivariate time series classification. We employ a publicly available data set provided by Berkeley Labs, which includes ground-truth data concerning the presence and absence of building faults. This dataset covers a wide spectrum of seasons and operational conditions, encompassing multiple building system types. It also includes detailed information on fault severity and data points indicative of measurements in building control systems, which are typically accessible to FDD algorithms. The data compilation leverages both simulation models and experimental test facilities. Our findings suggested that Canonical Interval Forest CIF and K-Nearest Neighbors KNN with dynamic Time Warping DTW have the highest average performance over the datasets analyzed with 0.78 and 0.73 respectively. This is particularly notable given the lower computational resources required by these methods compared to deep learning-based classifiers.

Advanced HVAC data analytics tools such as fault detection & diagnostics (FDD) are growing in popularity in large commercial buildings, and recent research has demonstrated potential for such tools to automatically correct certain faults. However, market adoption of such technologies is very low in small- and medium-sized buildings (SMBs). Furthermore, a lack of on-site maintenance teams and limited maintenance expenses restrict the efficient maintenance activities after faults occur and are identified in HVAC systems. Those factors could cause significant energy waste and downgrading system performance (e.g., decreased occupant comfort and reduced system lifetime). Considering commercial buildings under 50k square feet comprise 94% of all commercial buildings in the U.S, there is a significant need to develop cost-effective solutions for those buildings.

While much attention has been given to the benefits of smart thermostats, more prevalent mid-tier connected thermostats provide significant untapped opportunities to advance the state of operational practice in SMBs. These devices now offer two-way APIs that can be combined with everyday computing resources to open the door to continuous, automated monitoring and correction of the most common problems in HVAC control settings. Such technology presents the potential for a lightweight solution to address monitoring and control barriers.

In this paper, we present the results of a study to develop self-correction algorithms to correct common setting faults in SMBs, including inefficient HVAC setpoints, and wrong schedule setting when using thermostats. The detection and self-correction actions were achieved by employing two-way APIs embedded in thermostat products. The developed correction algorithms were evaluated in a lab environment. The results show that common thermostat setting faults can be efficiently corrected. Consequently, energy waste that commonly goes undiscovered, can be effectively avoided. We conclude with a discussion of how these solutions can be delivered to market by OEMs or as managed service offerings from thermostat installers and other service providers.