Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Development of a supervised ml model for occupant detection in smart buildings
A robust, grey-box approach for occupant detection based on indoor air quality data is developed, using previous development and Empa NEST's infrastructure.
Hello curious reader, looking forward to hear from you :). In the sections below you find information to the project, your work, what we expect and what you can expect.
If you have further questions contact me directly via email: michael.locher@empa.ch
**The Project:**
In this semester project, we offer the opportunity to work directly on an industrial project. The aim is to draw conclusions about the number of people in a room based on room climate values and taking into account potential disturbance factors. The occupant-centered information / number of people in a room is one of the most important key features in the smart building sector and the starting point for the development of energy optimization applications. All solutions for occupant's detection, currently available on the market, are based on personalized data, obtained by surveillance systems, which could lead to potential violations of privacy. Using indoor air quality data, which is inherently anonymous data, would allow for non-intrusive occupants detection
For this project we use therefore the indoor climate data from our living lab NEST (Energy Hub: nestcollabo-ration.ch) e.g. indoor air temperature, humidity, air pollutants and factors influencing the indoor climate: ventilation, open windows, doors and number of occupants.
The project is divided into 3 phases: Development of a deterministic model, extension using ML methods, and field tests.
In the first phase, we equipped several units in NEST with people counting sensors and developed and tested a deterministic model based on physical laws, such as the change of CO2 content in a room by the breathing of individuals as well as room specifications and ventilation operations. The model already shows appealing results in closed rooms, but currently neglects more complex relationships and variables of the room climate, e.g. interaction of open windows, doors and running air conditioning, which influence each other. These factors are extremely difficult to describe using physical laws, and are modelled using ML approaches. This is the focus of the second phase and of this semester project.
In the second phase, we want to design and test different supervised ML models to account for all interrelationships. For this purpose, the data will be labelled manually and validation strategies will be worked out, which will be used for the subsequent automation. The complete and correct data will be analyzed to get a space-specific understanding of the interrelationships and their effect sizes, which will feed into the architecture of the supervised ML solution (see attached picture).
In the last phase, the results of the previous phases will be combined in a so-called grey-box model (physical models in combination with ML approaches) and tested in practice.
**Your task:**
First you will be introduced to the NEST environment, our data management and data. The data includes in-door climate and its control variables e.g. ventilation, heating etc. Afterwards, we will prepare the data and work out and implement validation strategies, which will also be used later for a fast and automated access. Finally, the data will be used to analyze the dynamics of the interrelationships and their effect sizes. The results will then be used for modelling, especially for feature engineering and ML conceptualization.
**
You have:**
Experience in coding (Python preferred) and a good command of standard data processing and statistical analysis (descriptive statistics and statistical inference).
**Facts:**
Introduction NEST and its data management
Active integration into the project team
Collaboration between Empa UES and industry partner OAW
Hello curious reader, looking forward to hear from you :). In the sections below you find information to the project, your work, what we expect and what you can expect. If you have further questions contact me directly via email: michael.locher@empa.ch
**The Project:** In this semester project, we offer the opportunity to work directly on an industrial project. The aim is to draw conclusions about the number of people in a room based on room climate values and taking into account potential disturbance factors. The occupant-centered information / number of people in a room is one of the most important key features in the smart building sector and the starting point for the development of energy optimization applications. All solutions for occupant's detection, currently available on the market, are based on personalized data, obtained by surveillance systems, which could lead to potential violations of privacy. Using indoor air quality data, which is inherently anonymous data, would allow for non-intrusive occupants detection For this project we use therefore the indoor climate data from our living lab NEST (Energy Hub: nestcollabo-ration.ch) e.g. indoor air temperature, humidity, air pollutants and factors influencing the indoor climate: ventilation, open windows, doors and number of occupants.
The project is divided into 3 phases: Development of a deterministic model, extension using ML methods, and field tests.
In the first phase, we equipped several units in NEST with people counting sensors and developed and tested a deterministic model based on physical laws, such as the change of CO2 content in a room by the breathing of individuals as well as room specifications and ventilation operations. The model already shows appealing results in closed rooms, but currently neglects more complex relationships and variables of the room climate, e.g. interaction of open windows, doors and running air conditioning, which influence each other. These factors are extremely difficult to describe using physical laws, and are modelled using ML approaches. This is the focus of the second phase and of this semester project.
In the second phase, we want to design and test different supervised ML models to account for all interrelationships. For this purpose, the data will be labelled manually and validation strategies will be worked out, which will be used for the subsequent automation. The complete and correct data will be analyzed to get a space-specific understanding of the interrelationships and their effect sizes, which will feed into the architecture of the supervised ML solution (see attached picture).
In the last phase, the results of the previous phases will be combined in a so-called grey-box model (physical models in combination with ML approaches) and tested in practice.
**Your task:** First you will be introduced to the NEST environment, our data management and data. The data includes in-door climate and its control variables e.g. ventilation, heating etc. Afterwards, we will prepare the data and work out and implement validation strategies, which will also be used later for a fast and automated access. Finally, the data will be used to analyze the dynamics of the interrelationships and their effect sizes. The results will then be used for modelling, especially for feature engineering and ML conceptualization. ** You have:** Experience in coding (Python preferred) and a good command of standard data processing and statistical analysis (descriptive statistics and statistical inference).
**Facts:** Introduction NEST and its data management Active integration into the project team Collaboration between Empa UES and industry partner OAW
collecting, processing and validating ground truth data for model development; automate validation process; analyse and understand interrelationships for non intrusive occupant detection for different environments;
collecting, processing and validating ground truth data for model development; automate validation process; analyse and understand interrelationships for non intrusive occupant detection for different environments;