Personalized Energy Services: A Data-Driven Methodology towards Sustainable, Smart Energy Systems

The rapid pace of urbanization has an impact on climate change and other environmental issues. Currently, 54% of the global population lives in cities accounting for two-thirds of global energy demand. Sustainable energy generation and consumption is the top humanity’s problem for the next 50 years. Faced with rising urban population and the need to achieve energy efficiency, urban planners are focusing on sustainable, smart energy systems. This has led to the development of Smart Grids (SG) that employs intelligent monitoring, control and communication technologies to enhance efficiency, reliability and sustainability of power generation and distribution networks.

While energy utilities are optimizing energy generation and distribution, consumers play a key role in sustainable energy usage. Several energy services are provided to the consumers to know households' hourly energy consumption, estimate monthly electricity cost and recommendations to reduce energy consumption. Furthermore, advanced services such as demand response, can now control and influence energy demand at the consumer-end to reduce the overall peak demand and re-shape demand profiles. The effectiveness and adoption of these services highly depend on the consumers’ awareness, their participation and engagement. Current energy services seldomly consider consumer preferences such as their daily behavior, comfort level and energy-consumption pattern. In this thesis, we investigate development of personalized energy services that strive to achieve a balance between efficient-energy consumption and user comfort.

Personalization refers to tailoring energy services based on individual consumers’ characteristics, preferences and behavior. To develop effective personalized energy services a set of challenges need to be tackled. First, fine-grained data collection at user and appliance level is required (data collection challenge). Mechanisms should be devised to collect fine-grained data at various levels in a non-intrusive way with minimal sensors. Second, personalized energy services require detailed user preferences such as their thermal comfort level, appliance usage behavior and daily habits (user preference challenge). Accurate learning models to derive user preferences with minimal training and intrusion are required. Third, energy services developed needs to be easily scalable, from one household to tens and thousands of households (scalability challenge). Mechanisms should be developed to tackle the deluge of data and support distributed storage and processing. Fourth, energy services should deliver real-time feedback or recommendations so that users can promptly act upon it (real time challenge). This calls for development of distributed and low complexity algorithms.

This thesis moves away from traditional SG services -- which hardly consider consumer preferences and comfort -- and proposes a novel approach to develop effective personalized energy services. The proposed energy services provide actionable feedback, raise awareness and promote energy-saving behavior among consumers.

In this thesis, we follow a bottom-up data-driven methodology to develop personalized energy services at various scales -- (i) nano: individual households, (ii) micro: buildings and spaces, and (iii) macro: neighborhoods and cities. To this end, we present our approach -- physical analytics for sustainable, smart energy systems -- that combines IoT data, physical modeling and data analytics to develop intelligent, personalized energy services. Physical analytics fuses data from various Internet of Things (IoT) devices such as smart meters, smart phones and smart watches, along with physical information such as household type, demographics and occupancy to infer energy-usage patterns, user behavior and discover hidden patterns. This approach is used to learn and model user preferences and energy usage, subsequently, employed to develop personalized energy services.

This thesis is organized into three parts. Part I describes how to derive fine-grained information with minimal sensors and intrusion. We present two novel algorithms viz., LocED and PEAT that derive fine-grained information from appliance and user level, respectively. This real-time information is used to raise awareness on energy-usage behavior among occupants. Part II presents personalized energy services targeted at households and buildings. We develop services that shift and/or reduce energy consumption and cost by considering individual consumers’ preferences and comfort. These energy services are aimed at providing actionable feedback to occupants towards sustainable energy usage. Part III presents energy services targeted at neighborhood and city level. These energy services aim to identify target consumers in a neighborhood based on their energy-usage pattern and preferences for various DR programs. Finally, we present data-processing architectures that investigate how to cope with the overwhelming data generated from smart meters towards design and development of sustainable, smart energy systems.

This thesis advocates that the design and development of energy services should follow personalized approach with consumer preferences and comfort given paramount importance. Results show that the personalized energy services developed has significant potential to raise awareness, reduce energy consumption and improve user comfort in smart -- homes, buildings and neighborhoods.