MULTISENSORY AND MACHINE LEARNING APPROACH TO IDENTIFY DRIVERS OF INDOOR AIR QUALITY

据估计，全球每年有800万人死于空气污染。因此，解决空气污染及其对人类健康的影响仍然是一项关键的全球挑战。目前，关于室内空气质量及其影响因素仍有许多未解决的问题。特别是，越来越多的证据表明，居住者的行为和活动对室内空气质量有很大的影响。然而，迄今为止，收集有力的证据一直具有挑战性。小型化、离散化和低成本传感器的进步，以及机器学习和分析方法的进步，为揭示这些行为提供了新的希望。使用多传感器方法（例如多种空气污染物、温度、湿度、光、声、能量）和机器学习技术，假设可以从长期、非侵入式监测中成功识别许多关键活动。例如，颗粒物质、二氧化氮和湿度同时达到峰值可能表明进行了特定的烹饪活动，噪音水平和能源数据进一步表明使用了抽油烟机。这个博士学位的总体目标是解开这些特征，以更好地理解居住者的活动和行为对室内空气质量的影响。这将为健康和低能耗建筑设计战略以及改进的运营战略和通风做法提供信息。更具体地说，博士学位的目标是：描述关键活动（烹饪、清洁等）的基线特征。开发数据收集的多感官监测方法。识别和测试适当的机器学习技术。验证此方法并将其应用于更广泛的实地研究。

Air pollution can be attributed to an estimated 8 million global deaths per year. Tacking air pollution and its impact upon human health therefore remains a key global challenge. At present, there are many unresolved questions regarding indoor air quality and the factors that influence it. In particular, there is emerging evidence that occupant behaviours and activities have a strong influence upon indoor air quality. However, to date it has been challenging to gather robust evidence.The advancement of miniaturised, discrete and low-cost sensors alongside advancements in machine learning and analytical methods offers new promise in uncovering these behaviours. Using a multi-sensor approach (e.g. multiple air pollutants, temperature, humidity, light, sound, energy) and machine learning techniques, it is hypothesized that many key activities may be successfully identified from long-term, non-intrusive monitoring. For example, simultaneous peaks in particulate matter, nitrogen dioxide and humidity might indicate a particular cooking activity, with noise levels and energy data further indicating the use of an extractor hood.The overall aim of this PhD would be to unlock these signatures to better understand the influence of occupant activities and behaviours upon indoor air quality. This would inform strategies for healthy and low-energy building design as well as improved operational strategies and ventilation practices. More specifically the PhD would aim to:Characterise the baseline signatures of key activities (cooking, cleaning, etc.).Develop multisensory monitoring approach for data collection.Identify and test appropriate machine learning techniques.Validate this approach and deploy in wider field studies.