Python script for time series forecasting with Random Forest

Technical assignment (TA) for the development of a program for forecasting electricity consumption time series

Project goal:

Develop a program for forecasting electricity consumption based on time series using the Random Forest method. The program should accept historical electricity consumption data, train the model, save the model and scaler, and use the trained model to forecast future consumption and save the results in a file.

Input data:

CSV file with historical data containing the following columns:

year: Year of observation.

month: Month of observation.

day: Day of observation.

hour: Hour of observation.

weekday: Day of the week.

temp: Air temperature (°C).

function: Electricity consumption (MW/h).

Target variable:

The target variable is electricity consumption (column function).

Main stages of the program:

1. Data loading:
   • Import data from CSV files.
   • Split data into features and target variable (electricity consumption).
2. Data scaling:
   • Use a scaler (e.g., StandardScaler) to bring data to a common scale (normalization or standardization).
3. Creating time windows:
   • Form feature sets and target values for each time window (e.g., 24 hours).
   • Features: year, month, day, hour, weekday, temp.
   • Target variable: function (electricity consumption).
4. Model training:
   • Use the RandomForestRegressor model to train on time series data.
   • Split data into training and testing sets for model evaluation.
   • Evaluate the model using RMSE or other suitable metrics.
5. Saving the model and scaler:
   • Save the trained model and scaler to files for future use.
6. Using time windows for forecasting:
   • Use the model to forecast future values based on new data.
7. Saving forecasting results:
   • Forecasted values should be saved in a CSV file for further analysis.
8. Technical requirements:
   
   
   • 
     
   • Programming
     language: Python 3.x.
   • Libraries used:
     • 
       
     • pandas
     • scikit-learn
     • numpy
9. Success criteria:
   
   
   1. The program correctly trains the model on historical data and saves it to a file.
   2. The program uses the trained model for forecasting on new data and
      saves the results.
   3. The program logs all key processes (training, forecasting,
      errors).
   4. The program structure and code are well documented, with instructions for running and
      using the program.
   5. The program should provide forecasts with a total deviation of no more than 5%. This means that
      the average error between actual data (in the test set) and the forecast
      should not exceed 5% of the actual values.

Additionally, there is ready-made code with a result error above 17%. It may be useful to you.