Author(s):L. Tian and A. Noore

Reference: International Journal of Electrical Power and Energy Systems, submitted and under review

Abstract: This paper presents a dynamically optimizing artificial neural network modeling approach for electric load forecasting in real-time implementation. A short-term load forecasting system using evolutionary connectionist model based on only multiple delayed power load data is proposed. During an online forecasting process, genetic algorithm is used to globally optimize the number of delayed input neurons and the number of neurons in the hidden layer of the neural network architecture. Modification of Levenberg-Marquardt algorithm with Bayesian regularization is used to improve the ability to forecast future power load profile. The optimized neural network architecture is iteratively and dynamically updated as new power load data arrives. The proposed power load forecasting approach is easily implemented to predict daily load data in real-time. Experimental results show that the ability to forecast electric load using only the load data is comparable to the performance of existing approaches that use multiple datasets such as load, temperature, humidity or price.

Keywords: Short-term load forecasting, neural networks, genetic algorithm