loading page

A Comparative Study of Forecasting Problems on Electrical Load Timeseries Data using Deep Learning Techniques
  • Harish Battula ,
  • Debasmita Panda ,
  • Krishna Reddy Konda
Harish Battula
National Institute of Technology Warangal

Corresponding Author:[email protected]

Author Profile
Debasmita Panda
Author Profile
Krishna Reddy Konda
Author Profile


This work presents a comparative study of forecasting problems with electrical time-series load demand data using deep learning models ANN-MLP, RNN-LSTM, and 1D-CNN for short-term and medium-term electrical load forecast using different timestamps data generated from the source data. Through analysis found that multiple timestamps ahead prediction (MTAP) using 1D-CNN show better results with reduced MAPE. 1D-CNN is selected as the best model for both short-term 1-day ahead load forecast using 1-hour timestamp data with MTAP methodology resulting in mean MAPE 4.62% with standard deviation 0.9 and for mid-term 1-year ahead load forecast using 1-month timestamp data with STAF methodology resulting in MAPE 1.45% with standard deviation 0.34. A simulation study performed on the monthly electricity demand time series for 35 European countries confirmed the high performance of the selected 1D-CNN model with MAPE 4.6% and its comparison with the other hybrid deep learning models using the three forecasting methodologies validates the failure of STAP methodology by comparing with STAF methodology results in the real-time forecast and superiority of MTAP methodology in the real-time forecast.
In this paper, the dataset we adopted contains the hourly electrical consumption profile of 35 European countries, from which Austria country load consumption profile from 1 January 2006 to 31 December 2015 is considered for the preliminary analysis. In this preliminary analysis, we will compare the 9 possible forecasting problems discussed earlier on the basic 3 deep learning models ANN-MLP, RNN-LSTM, and 1D-CNN and conclude which data set and model will be best suitable for short and mid-term load forecasting problems. The complete data set is discussed in chapter V. The data were obtained from the ENTSO-E repository (www.entsoe.eu).