A comparison of artificial neural networks algorithms for short term load forecasting in Greek intercontinental power system

  • Authors:

  • G. J. Tsekouras;F. D. Kanellos;V. T. Kontargyri;C. D. Tsirekis;I. S. Karanasiou;Ch. N. Elias;A. D. Salis;N. E. Mastorakis

  • Affiliations:

  • Department of Electrical & Computer Science, Hellenic Naval Academy, Terma Hatzikyriakou, Piraeus, Greece;School of Electrical and Computer Engineering, National Technical University of Athens, Zografou, Athens, Greece;School of Electrical and Computer Engineering, National Technical University of Athens, Zografou, Athens, Greece;School of Electrical and Computer Engineering, National Technical University of Athens, Zografou, Athens, Greece;School of Electrical and Computer Engineering, National Technical University of Athens, Zografou, Athens, Greece;School of Electrical and Computer Engineering, National Technical University of Athens, Zografou, Athens, Greece;School of Electrical and Computer Engineering, National Technical University of Athens, Zografou, Athens, Greece;Department of Electrical & Computer Science, Hellenic Naval Academy, Terma Hatzikyriakou, Piraeus, Greece

  • Venue:
  • COMPUCHEM'08 Proceedings of the 2nd WSEAS international conference on Computational chemistry
  • Year:
  • 2008

Quantified Score

Hi-index 0.01

Visualization

Abstract

The objective of this paper is to compare the performance of different Artificial Neural Network (ANN) training algorithms regarding the prediction of the hourly load demand of the next day in intercontinental Greek power system. These techniques are: (a) stochastic training process and (b) batch process with (i) constant learning rate, (ii) decreasing functions of learning rate and momentum term, (iii) adaptive rules of learning rate and momentum term, (c) conjugate gradient algorithm with (i) Fletcher-Reeves equation, (ii) Fletcher-Reeves equation and Powell-Beale restart, (iii) Polak-Ribiere equation, (iv) Polak-Ribiere equation and Powell-Beale restart, (d) scaled conjugate gradient algorithm, (e) resilient algorithm, (f) quasi-Newton algorithm, (g) Levenberg-Marquardt algorithm. Three types of input variables are used as inputs: (a) historical loads, (b) weather related inputs, (c) hour and day indicators. The training set is consisted of the actual historical data from three past years of the Greek power system. For each ANN training algorithm a calibration process is conducted regarding the crucial parameters values, such as the number of neurons, etc. The performance of each algorithm is evaluated by the Mean Absolute Percentage Error (MAPE) between the experimental and estimated values of the hourly load demand of the next day for the evaluation set in order to specify the ANN with the smallest value. Finally the load demand for the next day of the test set (with the historical data of the current year) is estimated using the best ANN of each training algorithm, so that the verification of behaviour of ANN load prediction techniques should be demonstrated.