ReadMe

Required files/inputs:

1. The Meters dictionary: Data/Meters_dictionary.csv

2. The Model: Model/Model-CNNLSTM.h5

3. The path to save the predictions: Results/preds.csv

4. The Weather Condition dictionary: Data/WeatherCondition_dictionary.csv

5. The Weather Wind dictionary: Data/WeatherWind_dictionary.csv

Data and Model Overview

Data

The dataset used in this solution comprises historical energy consumption data from various buildings and weather data from the EEKA station. The data sources are:

• Elering Dashboard: Historical energy consumption data was obtained from Elering Dashboard.
• Weather Data: Weather data was collected from the EEKA station on Wunderground.

For training the model, one year (2022) of historical data from the buildings' consumption was utilized. The final merged columns used for training include:

• Temperature
• Dew Point
• Humidity
• Wind Speed
• Wind Gust
• Pressure
• Precip.
• Wind
• Condition
• Soome
• Läti
• Venemaa(Leningrad)
• Venemaa(Pihkva)
• Tarbimine
• Tootmine
• Tuuleparkide toodang
• Päikeseenergia toodang
• month
• hour
• dayofweek
• Meter_code
• Meter_consumption

Model

The predictive model utilized in this solution is a Convolutional Neural Network (CNN) combined with Long Short-Term Memory (LSTM) architecture. The model is trained to forecast energy consumption based on historical data and external factors such as weather conditions.

• Model File: The trained model is saved as Model/Model-CNNLSTM.h5.
• Python Packages: The solution requires the following Python packages: pandas, numpy, keras, selenium, and bs4.

Carbon Intensity Calculation

Elering provides planned energy and I/E data ('Planned production', 'Wind farm production forecast', 'Solar energy production forecast'), allowing for the computation of carbon intensity using the following formula:

G = (f + (I * 0.1)) / (|E| + pred)

Where:

• ( G ) is the carbon intensity.
• ( f ) is the planned energy minus the sum of wind and solar energy predictions.
• ( I ) is the total positive import of energy.
• ( E ) is the total negative export of energy.
• ( pred ) is the predicted energy consumption.

This formula calculates the carbon intensity based on predicted energy consumption, Elering data, and the current datetime.

Getting Predictions

To obtain predictions for energy consumption, follow these steps:

1. Replace Paths: Adjust the paths in the command according to your directory structure.
2. Run the Script: Execute the predictions script Predictions.py using Python 3, providing the necessary arguments, including the meter name (--Meter_name), the new energy consumption value (--new_value), and the paths to data dictionaries and the model.

List of Meters names (18 meters in 8 different buildings):

• ICT_JK1_EM1.Total
• ICT_JK1_EM2.Total
• ITK_JK1_EM1.Total
• LIB_JK1_EM1.Total
• LIB_JK1_EM2.Total
• LIB_JK2_EM1.Total
• NRG_JK2_EM1.Total
• NRG_JK2_EM2.Total
• SOC_JK1_EM1.Total
• SOC_JK1_EM2.Total
• SPORT_JK1_EM1.Total
• SPORT_JK1_EM2.Total
• U03_JK1_EM1.Total
• U03_JK1_EM2.Total
• VK1_JK1_EM1.Total
• VK1_JK1_EM2.Total
• VK2_JK1_EM1.Total
• VK2_JK1_EM2.Total

Example Command:

python3 Predictions.py --Meter_name=ICT_JK1_EM1.Total --new_value=0.08536 --d_path=Meters_dictionary.csv --d_WC_path=Data/WeatherCondition_dictionary.csv --d_WW_path=Data/WeatherWind_dictionary.csv --model_path=Model/Model-CNNLSTM.h5 --pred_path=Results/preds.csv

This command will generate predictions for energy consumption based on the specified meter and input value.