Title :
A method to determine the input variable for the neural network model of the electrical system
Author :
Sovann, N. ; Nallagownden, Perumal ; Baharudin, Z.
Author_Institution :
Electr. & Electron. Eng. Dept., Univ. Teknol. PETRONAS, Tronoh, Malaysia
Abstract :
This paper highlights input variable selection for neural network of the electrical system to predict the load demand in 168 hours ahead. Autocorrelation (ACF), partial autocorrelation (PACF) and cross correlation (CCF) analysis are used to identify the correlated input for the forecast model. The combination of time, time indicator, lagged load and respective weather variables are considered as forecast model inputs in order to enhance the accuracy of short-term term load forecast. Several case studies are designed to analyse the highly correlated forecast model inputs and evaluate its performance using artificial neural network (ANN) based forecast model. The accuracy of MAPE 1.79% is achieved for the model, by applying the input variables of lagged load, time, time indicator and forecasted weather.
Keywords :
load forecasting; neural nets; power engineering computing; ACF analysis; ANN; CCF analysis; MAPE; PACF analysis; artificial neural network based forecast model; autocorrelation analysis; cross correlation analysis; electrical system; forecast model inputs; input variable determination method; lagged load; load demand prediction; partial autocorrelation analysis; respective weather variables; short-term term load forecast; time 168 hour; time indicator; Artificial neural networks; Correlation; Input variables; Load modeling; Predictive models; Weather forecasting; Artificial Neural Network (ANN); Autocorrelation Function (ACF); Cross Correlation Function (CCF); Partial Autocorrelation Function (PACF);
Conference_Titel :
Intelligent and Advanced Systems (ICIAS), 2014 5th International Conference on
Conference_Location :
Kuala Lumpur
Print_ISBN :
978-1-4799-4654-9
DOI :
10.1109/ICIAS.2014.6869491
