| Popis: |
The Department of Energy, International Energy Agency, Intergovernmental Panel on Climate Change and other agencies have declared a need for commercial buildings to become 70-80% more energy efficient. Growing demand for energy efficient buildings requires robust models to ensure efficient performance over the evolving life cycle of the building. Energy management and control systems can reduce energy consumption in buildings without sacrificing occupant's comfort. However, their full capabilities have not been completely realized, partly due to their inability to quickly detect faults in HVAC systems. An accurate model and an appropriate threshold are the key factors in fault detection. Machine learning-based energy models have proved to be more efficient and accurate where historical time series data is available. This dissertation presents and compares various machine learning algorithms that will aid in the generation of more efficient energy models. This dissertation proposes the use of tree-based ensemble learning algorithms to develop accurate energy models.Traditionally, fixed threshold values have been used to differentiate between system faults and energy model error. These threshold values are often set based on the user-experience, rule-of-thumb, or statistical methods. These traditional methods often lead to missed opportunities to detect faults, delayed detection of faults or false alarms. To improve the effectiveness of fault detection frameworks, this research proposes a dynamic threshold method to determine occurrences of faults in real time. This method adjusts the threshold value dynamically according to the real-time moving average and moving standard deviation of the predictions. The performance of the proposed fault detection framework is compared against existing research papers. The results demonstrate the usefulness of the proposed framework to detect faults early in the course and reduce the number of missed opportunities to detect faults.The fault detection frameworks have been comparatively tested with both synthetic and real datasets. It is shown that the proposed framework, based on tree-based ensemble learning algorithms and the dynamic threshold method, is adequately robust to work well under highly skewed datasets. The results suggest that the proposed framework has a fast reaction speed, i.e. there is no time delay between the occurrence of a fault and its detection. The results also suggest that the proposed framework is better than the existing methods, proposed by researchers in the past, especially at the lower fault severity levels.The proposed fault detection framework is expected to reduce energy wastage, prevent further deterioration of equipment, reduce equipment downtime due to major maintenance, and prolong the equipment life. This framework can be used by the building operators to ensure that no energy is wasted in its operation that could excessively increase the energy cost of the building. This framework can also be used by the maintenance team to locate and repair faults in its early stage before it can cause any major damage to the system. |
