Performance Evaluation of Prophet and STL-ETS methods for Load Forecasting

This work contributes to Short Term load forecasting methods by investigating the performance of Prophet method and comparing it with that of Seasonality and Trend Analysis with Loess (STL-ETS) method (popular in forecasting time-series for near future) to establish its capability of capturing multiple seasonality, sudden pattern shift, computational ease efficiency, simplicity and accuracy. The experimental outcomes validate prophet-based forecasting model for practical deployment. Both models are built using five-year electricity demand data of Texas region, USA. Metric parameters selected to evaluate the performance are Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). Both forecast models are then utilized to predict and compare the day-Ahead half-hourly load of last 15 days of actual data, year 2015. Research work is carried out in R Package (R-core Team, 2020) version 1.3.1093.