• predict temperature measured by temperature sensor in the next 15 minutes
• predict valve level of radiator in the next 15 minutes

The project was created in cooperation with LERTA and the data was shared by them. The data was registered in an office building located in Poznań. The registered data consisted of:

• room temperature from 3 different sensors in a room [°C]
• valve level of radiator [%]
• set temperature [°C]
• timestamp for each measurement
• number of people in a room
• window direction

• Python 3.8
• Listed in requirements.txt
• Input data for 1 prediction consisted of last 7 days before the prediction.

thermal_comfort/
    common/ - common functions
    data/ - data for training and evaluation
    models/ - trained models
    processing/ - functions to predict values
    train/ - training code
    main.py - main code

To predict values run main.py script with 2 arguments.

python3 main.py input_file.json results.csv

• input_file.json - define time interval for prediction
• results.csv - file to save predicted values and ground truth values

Both values are predicted using Random Forest Regressor. Both models can be found in models directory

The results of a project was evaluated using Mean Average Error (MAE).

• room temperature: 0.057°C
• valve level: 1.74 %

• room temperature: 0.068°C
• valve level: 5.05 %