Random Forest and Xgboost Technique for Short-Term Load Forecasting
Utkarsh Singh, Shelly Vadhera
Abstract
In power system, load forecasting for short-term is a step forward in the course of making a completely reliable grid. With the advent of smart grid and its technologies, the use of load forecasting becomes an extremely important and inevitable task. It helps the electric utilities to prepare and organize based on the load's changing requirements. This eliminates any uncertainty, ensuring optimum and reliable performance of the grid. In this paper, a model which uses Random Forest and Xgboost algorithms to predict one hour ahead load is presented. The load data set obtained is from US electric utility company Commonwealth Edison (ComEd) from the year 2011 to 2018. The weather data of Chicago, Illinois for the above period is used in this model. Data from the years 2011 to 2017 has been taken to train the model and data from the year 2018 has been taken for evaluation. The model has been implemented in the Google Colab environment. Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) is used as evaluation metrics for the predicted output. The Random Forest algorithm gives MAE of 100.79 MW and MAPE of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0.875\pm 0.76}$</tex> . With the Xgboost algorithm, MAE of 85.92 MW and MAPE of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0.738\pm 0.646}$</tex> were obtained.