ETHOS Student Projects

This project is offered in collaboration with the Urban Energy Systems group at EMPA.

The proposed project will develop a common database of metering and sub-metering data from Swiss households. This data will be used to develop non-intrusive load monitoring (NILM) algorithms and customer segmentation models. The project will also evaluate the demand flexibility potential of Swiss households and develop targeted demand flexibility programs. The project will develop new tools and knowledge that can be used to design and implement effective demand flexibility programs in Switzerland. The project will also contribute to the development of a more reliable and sustainable power system.

The key steps envisioned for the completion of the project are as follows:

• Literature review
• Data collection and preparation
• NILM algorithm development and validation
• Customer segmentation modeling
• Development of household energy model
• Demand flexibility analysis
• Documentation

Supervisor: Andrew Sonta ([email protected])

Household energy consumption patterns, accounting for ~25% of European electricity demand, play a pivotal role in demand flexibility to support the grids under increasing intermittent renewable generations. The specific patterns of household appliance usage and time preferences can be the complex consequence of asset and facility conditions, household economic status, resident occupational and recreational lifestyles, and local social-organizational context. We have also been working towards integrating household energy models with socio-economic survey data to emulate these complex and heterogeneous patterns. Agent-based modeling (ABM) via large language models (LLMs) is a promising approach to reflect individual household properties and simulate complex human-like reasoning, behavioral adaption, and interactions in this process via LLM.  In this project, we will leverage existing LLM-agent frameworks to simulate Swiss households’ energy behavior using our collected demographic and time-use survey data, and gain understanding of  populational behavioural shifts & individual reactions to different demand response policy and extreme weather scenarios.

Supervisors: Vasantha Ramani ([email protected]) and Yufei Zhang ([email protected])

The purpose of this research is to answer the following questions: How can causal loop diagramming be used to model and understand equity-related factors in the built environment? What key feedback loops and relationships drive inequities? What are the potential leverage points for reducing inequities within the built environment system?

As an initial exploration, this project focuses on mining the text of existing research studies to infer causal loop diagrams of factors related to (in)equity in the built environment. Students should be comfortable with Python and willing to develop and learn methods for mining information from large text databases.

Supervisor: Andrew Sonta ([email protected]) 

The importance of the design of the built environment on human behavior and experiences has been studied, but largely from a theoretical perspective and lacking empirical evidence. This project aims to understand where data-driven analysis is needed and to conduct data analysis to better characterize the human and built environment relationships. Some components of this topic include investigating how physical features of the existing built environment influence human outcomes and experiences (e.g., social cohesion and well-being) and how human behaviors are interacting with the built environment (e.g. active mobility such as walking). The detailed focus of this project will be shaped based on students’ interests and background. 

Supervisor: Kanaha Shoji ([email protected])

Comfort and satisfaction of occupants are key objectives of building management and lead to complex trade-offs with energy and emission targets. Recent advances in sensing and machine-learning have catalyzed a new paradigm of personalized comfort models that can predict and analyze individuals’ dynamic comfort statuses based on streamed localized/portable/wearable sensors alongside comfort perception feedback from mobile apps. This study aims to progress from the previously-collected personalized comfort datasets and recently developed models in our lab, to address the remaining issues in modeling and analysis strategies. We may together address one or several of the following subtasks, depending on the duration of the project: 1). data-mining bodily and ambient measurements: critical change in ambient and occupant’s bodily condition may be associated with potential comfort status change. We want to apply widely-used change point detection & data-drift detection algorithms to spot those changes. 2). Heterogeneity analysis of individual comfort perception: occupants’ perception of comfort status can be rather heterogeneous by nature, and it is necessary to propose metrics that effectively cluster representative types of comfort preference and sensitivity.  3). Feature analysis for interpretable personalized comfort prediction: we will further explore recently-developed machine-learning comfort status prediction and feature analysis pipeline to draw critical guidelines for successful real-world deployment of personalized comfort models, such as minimally-required features set that are most relevant, and representative scenarios of comfort status transition under interactions of multiple features. 

Supervisor: Yufei Zhang and Matteo Favero ([email protected])

With the advent of low-cost sensing devices, it is becoming easier to measure the qualities of the indoor environment, such as air quality, occupant presence, noise levels, and more. Doing so can improve our understanding of the performance of the building and also gather non-invasive information about the occupant’s experience. This project will develop an indoor environment sensing strategy and gather initial data. The aim of the sensing strategy will be to focus on human-building interactions, such as occupant presence and occupant social interactions in the office. The data will be analyzed to determine what occupant interactions can be inferred through the non-invasive, privacy-preserving data collection.

Supervisor: Andrew Sonta and Vasantha Ramani ([email protected]) 

Household energy consumption comes from the operation of various electrical appliances, space and water heating, space cooling, lighting and other end uses.  The distribution of energy for various end uses may vary depending on the type of dwelling, household size, income level, types of appliances, weather conditions and several other factors. This study is aimed at understanding the differences and similarities in occupancy and occupant behavior globally based on smart meter data and end use energy distributions in residential units.

Supervisor: Vasantha Ramani ([email protected])

Demand side management involves changes in energy consumption by the customers in order to reduce, remove or shift demand. This study is aimed at developing household-specific demand management strategies based on their domestic appliance usage. The main energy meter data is cumulative of the energy load from all the appliances, which can be disaggregated to identify various appliance usage using non-intrusive load monitoring (NILM) techniques. The extracted appliance signature is indicative of time and duration of appliance usage, which can then be used to identify deferrable and non-deferrable appliances in the household. This study is aimed at identifying different user types based on their interaction with domestic appliances and developing strategies for demand response.

Supervisor: Vasantha Ramani ([email protected])

With the amount of data about urban environments steadily rising, we have new possibilities to understand the structure of cities. Data about urban form—such as characteristics of buildings, pedestrian and vehicle street networks, public space, and more—can tell us a great deal about how cities are planned and designed. As different cities, or different parts of the same city, would be expected to have different characteristics of design, we would also expect those cities to create different experiences for their inhabitants. In this project, we will use machine learning clustering tools to identify different “natural” city forms, and we will attempt to interpret the results of the clustering algorithm. The results of this investigation will enable new research that allows comparison of different forms vis-a-vis human experiences and behaviors in the urban built environment.

Supervisor: Andrew Sonta ([email protected]) 

Summary: This project aims to develop a composite index of walkability and cyclability, integrating physical (infrastructure), social (mobility practices), and environmental (air pollution, noise) dimensions, across multiple spatial scales (street segment, neighborhood, municipality). The framework will be piloted in the Canton of Vaud.

Data: TLM networks (sidewalks, cycling paths), Open data (OSM / Ouverture).

Prerequisites: Advanced Python + advanced statistics, Spatial Data Science, GIS; interest in active mobility, sustainable urbanism, and accessibility.

Contact: Andrew Sonta ([email protected]) 

Partner: Bureau Action Située

Summary: This project explores the relationship between the shape and extent of individuals’ activity spaces (covered areas, distances traveled, spatial dispersion) and their daily carbon footprint. The aim is to identify low-carbon mobility profiles while maintaining accessibility to essential activities, location innovation rate, spatial consumption, etc.

Data: Panel Lémanique 2023 (GPS data and questionnaires), TLM networks, emission factors (OFEV / MOBITOOL), GTFS datasets, and Open Data on transport networks.

Prerequisites: Advanced Python + advanced statistics, Spatial Data Science, GIS; strong interest in low-carbon lifestyles and sustainable mobility.

Contact: Andrew Sonta ([email protected]) 

Partner: Bureau Action Située

Summary: This project investigates how transport nodes—such as train stations, bus stops, and multimodal hubs—can be redesigned as inclusive centralities that better accommodate care-related mobility (e.g. childcare, elderly care, dependent relatives). The goal is to characterize care-related mobility chains and develop multidimensional accessibility indicators to evaluate and diagnose caregiving-friendly hubs.

Data: Panel Lémanique 2023 (GPS data and questionnaires), TLM networks, GTFS datasets, Open Data on local services and facilities.

Prerequisites: Advanced Python + advanced statistics, Spatial Data Science, GIS; interest in daily mobility patterns and social inequalities.

Contact: Andrew Sonta ([email protected]) 

Partner: Bureau Action Située