Data-Driven Approach to Forecast and Visualize User Behavior

This project proposes a data-driven method to discover user behavior patterns, predict demand under given conditions and identify its relevant influencing factors within the residential energy usage domain. In order to make our models transparent, we apply visual analytics and explainable artificial intelligence to show how the models make decisions.

The aim is to answer the following main research questions for a case study on (smart) buildings and energy demand behavior in them:

• What user behavior patterns change over time?
• What differences are there in user behavior between different countries/regions?
• What kinds of user behavior groups exist?
• What data-driven methods can be used to predict energy demand based on user behavior?
• What are the (indirect) influencing factors on the energy demand of users?
• What role do technical and social innovations (e.g. demand response systems, photovoltaic, electric vehicles, innovative tariff strategies) play in influencing behavior patterns towards long-term societal goals, e.g. for energy system transformation?

These research questions will be answered by collecting unstructured data from several sources (e.g. open questions in socio-economic surveys, social media, technology reports, webpages, etc.) together with structured datasets (e.g. smart meter data, structured answers in socio-economic survey, climate data and other public databases).
 

• Database: Energy Consumption Behavior Database in Residential Buildings
• Workflow/Model: Load Forecast and Visualization of User Behavior
• Working Paper: Explainable Deep Attention Model for Residential Energy Demand Forecast with User Profile Learning
• Master Thesis: Analysis of User Behavior of Residential Energy Demand based on Smart Meter Data