Electricity Load Forecasting: A Weekday-Based Approach
We present a new approach for building weekday-based prediction models for electricity load forecasting. The key idea is to conduct a local feature selection using autocorrelation analysis for each day of the week and build a separate prediction model using linear regression and backpropagation neural networks. We used two years of 5-minute electricity load data for the state of New South Wales in Australia to evaluate performance. Our results showed that the weekday-based local prediction model, when used with linear regression, obtained a small and statistically significant increase in accuracy in comparison with the global (one for all days) prediction model. Both models, local and global, when used with linear regression were accurate and fast to train and are suitable for practical applications.
Keywordselectricity load forecasting autocorrelation analysis linear regression backpropagation neural networks weekday-based prediction model
Unable to display preview. Download preview PDF.
- 5.Shamsollahi, P., Cheung, K.W., Chen, Q., Germain, E.H.: A neural network based VSTLF for the interim ISO New England electricity market system. In: 22nd IEEE PICA Conf., pp. 217–222 (2001)Google Scholar
- 6.Sood, R., Koprinska, I., Agelidis, V.: Electricity load forecasting based on autocorrelation analysis. In: International Joint Conference on Neural Networks (IJCNN), Barcelona, pp. 1772–1779 (2010)Google Scholar
- 7.Koprinska, I., Rana, M., Agelidis, V.: Yearly and Seasonal Models for Electricity Load Forecasting. In: International Joint Conference on Neural Networks (IJCNN 2011), USA (2011)Google Scholar
- 11.Australian Energy Market Opeartor (AEMO) http://www.aemo.com.au