Power demand forecasting through social network activity and artificial neural networks

Ana Luna, Miguel Nunez-Del-Prado, Alvaro Talavera, Erick Somocurcio Holguin

Research output: Contribution to conferencePaper

4 Scopus citations


Long-term and short-term term national power demand forecasting is a well known and open issue for many countries. In this paper, we focus and study the short-term Peruvian national power demand forecasting. Thus, we tackle this problem using indirect and direct method for prediction. The former method relies on Social Network Activity to estimate national needs using regression models. The latter method is based on Artificial Neural Networks (ANNs). The network was used subsequently for predictions of the power for the last day of April, May and June 2016. The result was highly satisfactory with a mean absolute percentage error (MAPE) of 0.36 % for April and 0.34% in May and June. The ANN cumulative model proved to be a fast, reliable and accurate method for predicting power demand in Peril. In the case of the social activity generated by tweets, there is an increase in the MAPE values of an order of magnitude, reaching a maximum value of 7.3% for June. Nevertheless, the power demand forecasting using Twitter posts is a good indicator as a first approximation.
Original languageEnglish
StatePublished - 27 Jan 2017
EventProceedings of the 2016 IEEE ANDESCON, ANDESCON 2016 -
Duration: 27 Jan 2017 → …


ConferenceProceedings of the 2016 IEEE ANDESCON, ANDESCON 2016
Period27/01/17 → …


  • Artificial neural networks
  • Power demand forecast
  • Power systems
  • social network activity


Dive into the research topics of 'Power demand forecasting through social network activity and artificial neural networks'. Together they form a unique fingerprint.

Cite this