e-space
Manchester Metropolitan University's Research Repository

    An Investigation into Net Zero Energy Public Buildings with a Focus on Machine Learning Methods

    Scott, Connor Lee (2024) An Investigation into Net Zero Energy Public Buildings with a Focus on Machine Learning Methods. Doctoral thesis (PhD), Manchester Metropolitan University.

    [img]
    Preview

    Available under License Creative Commons Attribution Non-commercial No Derivatives.

    Download (5MB) | Preview

    Abstract

    Global warming’s negative effect on the environment necessitates the timely need to reduce CO2 emissions. Commercial buildings are major contributors into such emissions in the UK that consume 277TWh from the Government’s latest data collection and 89% of non-domestic energy consumption was from the grid. Therefore, government legislations worldwide are becoming increasingly stricter on carbon dioxide emissions from buildings. However, the buildings must continue to run, fully functioning, while still reducing emissions. While some building management systems (BMS) can manage the demands of the building and reduce energy waste, they rely on the current demand instead of planning for future demand. Moreover, some space heating methods are not fully understood such as with recently developed infrared technologies. Electric vehicle charging schedules are not previously optimised, and neither is energy management from on-site renewable generation. In this research work, already existing machine learning algorithms (MLA) methods are applied to optimise energy efficiency of commercial buildings. Neural networks, random forests, support vector machines, and linear regression are developed to accurately forecast the buildings’ energy characteristics for various resolutions and horizons to determine the best method and application to the BMS. From the developed MLA’s, the random forest has an accuracy as high as 96% and can forecast the energy demand in 15-minute resolutions on 24-hour horizons. One outcome of this research is a developed strategy that saves 64.7% on costs through using the energy capacity of electric vehicles. This allows energy to be purchased from the EV instead of from the grid at peak-times. Moreover, the heating consumption of a lecture hall is reduced by 75.97% through using infrared heating combined with MLA occupation density forecasting. Furthermore, neural networks, random forests, support vector machines, and linear regression to forecast the active solar panels are developed and critically analysed to determine data requirements and surrounding affecting factors. The MLA’s used to forecast the energy consumption of the case study building are trained on a 16GB intel core i7, with a dataset of 97,185 samples of 11 features, with average training times of 172s (neural network), 61.9s (random forest), 270.3s(support vector machine), and 41.7s (linear regression) respectively.

    Impact and Reach

    Statistics

    Activity Overview
    6 month trend
    80Downloads
    6 month trend
    55Hits

    Additional statistics for this dataset are available via IRStats2.

    Repository staff only

    Edit record Edit record