Data and Sourcecode from: Load Prediction for Mixed Use Districts

Data and Sourcecode of: LOAD PREDICTION FOR MIXED USE DISTRICTS

Prediction of electricity demand for Inffeldgasse campus at the university of technology Graz, Austria, based on weather, time-based features, energy consumption and special load information.

Investigation of short- and long-term energy consumption prediction with and without knowing that special loads (e.g. laboratories with really high energy consumption) are active.

A voting regressor (=stacked model) is implemented to combine the advantages of both short- and long-term prediction models.

It turned out that special load information is crucial for predicting the energy consumption where the random forest performed best.

Dataset Description

data/raw/labs_data/

Absolute humidity, dew point, enthalpy, global irradiation diffuse/total, relative humidity and temperature are historical measurements from campus Inffeldgasse.

The remaining files in this directory hold energy consumption data of laboratories with extraordinary high energy consumption.

Later, this information is used to generate binary vectors to indicate if certain laboratories are in operation.

Total energy consumption is sum of sensors 20_000100.csv and 20_999100.csv.

data/raw/weather/

climate_data_graz.csv holds monthly climatic conditions of Graz, Austria, which is used for long-term energy predictions instead of historic measurements.

meteoblue_data.json holds weather forecast data from Meteoblue, which is used for short-term energy prediction instead of historic measurements.

data/processed/

preprocessed.csv holds all features for model training (short- and long-term models) generated by notebooks/1.0-RKr-preprocessing.py.

Features/columns are selected immediately before model training.

special_loads.csv holds energy consumption data of all laboratories.