Search

The ASHRAE Great Energy Predictor III (GEPIII) competition was held in late 2019 as one of the largest machine learning competitions ever held focused on building performance. It was hosted on the Kaggle platform and resulted in 39,402 prediction submissions, with the top five teams splitting $25,000 in prize money. This paper outlines lessons learned from participants, mainly from teams who scored in the top 5% of the competition. Various insights were gained from their experience through an online survey, analysis of publicly shared submissions and notebooks, and the documentation of the winning teams. The top-performing solutions mostly used ensembles of Gradient Boosting Machine (GBM) tree-based models, with the LightGBM package being the most popular. The survey participants indicated that the preprocessing and feature extraction phases were the most important aspects of creating the best modeling approach. All the survey respondents used Python as their primary modeling tool, and it was common to use Jupyter-style Notebooks as development environments. These conclusions are essential to help steer the research and practical implementation of building energy meter prediction in the future. [ABSTRACT FROM AUTHOR]

Grid-interactive efficient buildings (GEBs) require quick and accurate manipulation of flexible electrical loads. Heating, ventilating, and air conditioning (HVAC) loads are particularly well suited for demand flexibility control by utilizing thermal mass in building structures. The HVAC load and its responding zone temperature (ZT) can be optimized by model predictive control (MPC) to minimize the operation cost with the indoor comfort constraint. However, due to the lack of reliable and affordable energy meters, the current HVAC demand control is implemented using ZT feedback control with the responding ZT optimized by MPC as the setpoint and the HVAC load as the response. Unfortunately, the building thermal mass makes the actual HVAC load significantly different than the optimized HVAC load, which is the on-demand setpoint, with potential demand penalty. Low-cost virtual flow meter technology makes the cooling load measurement at air handling units (AHUs) available, thus energy feedback control can be applied to actively control the cooling load of AHUs at its on-demand setpoint. The objective of the paper is to preliminarily investigate the demand flexibility control performance of the energy feedback control at AHUs. First, the energy feedback control at AHUs is introduced. Then, the performance of a single zone AHU is simulated and compared between the ZT feedback control and energy feedback control. Finally, a field experiment is conducted on a 7-ton single duct variable air volume AHU with these two different controls. Experimental results show that the cooling load with the current ZT feedback control drops to the on-demand setpoint after 25 minutes. On the other hand, the energy feedback control can reduce the cooling load to the on-demand setpoint in less than 3 minutes. Therefore, the active energy feedback control can significantly advance the current ZT feedback control by quickly and accurately controlling the cooling load at AHUs to avoid the demand penalty. [ABSTRACT FROM AUTHOR]

Operational deficiencies such as inappropriate scheduling and malfunctioning components in HVAC systems attribute to substantial energy losses in commercial buildings and are often left unaddressed due to a lack of accessible analytical tools which can identify energy-saving measures. A novel multisource, data-driven energy management toolkit was proposed, which synthesized established inverse energy modeling, anomaly detection and diagnostics, load disaggregation, and occupancy and occupant complaint analytics methods in the literature. The toolkit contains seven functions that input HVAC control, energy meter, Wi-Fi device count, and work order log data to detect hard and soft faults, improve sequences of operation, and monitor energy flows, occupancy patterns, and occupant satisfaction. These capabilities were demonstrated on a case study building and the generated insights identified a fault in operational logic which exacerbated heating energy use during afterhours. [ABSTRACT FROM AUTHOR]