PREDICTION OF TIME-SERIES USING TIME-DELAY NEURAL NETWORK ON THE EXAMPLE OF TOTAL ENERGY LOAD

Abstract

Predicting the total energy load is extremely important for all elements of the energy system for a number of reasons. This is even more pronounced when energy system includes a significant number of volatile renewable sources. This paper examines the ability of the Time-Delay Neural Network model to predict total energy load, on an hourly basis, using real-world data from Spain. The neural network was created, trained and tested in Neurosolutions 5.0 software. The dataset was obtained from Kaggle database and consists of 35,000 real-world hourly total energy load records from Spain. The results showed a high degree of linear correlation (0.975) between the observed and predicted values with satisfactory Relative Mean-absolute-error (1.239%), and relative Root-mean-square error (1.626%) values. Based on the results, TDNN model emerged as a promising method for both energy load prediction, and time-series prediction.