Prediction of Short-Term Electricity Consumption by Artificial Neural Networks Levenberg-Marquardt Algorithm in Hormozgan Province, Iran

Abstract

The prediction of short-term electricity consumption is of vital importance for designing and management of energy production and storage systems. Forecasting the next 24 h of electrical load allows operators to prevent power blackout and optimize their resources. In this paper, we present a more accurate prediction methodology for short-term energy consumption utilizing optimized artificial neural networks (ANNs) for Hormozgan. Variant architectures were developed for hourly load prediction. For the determination of best combinations of learning algorithms, layers and neurons, the main training algorithm of Levenberg-Marquardt was used. In the proposed implementation of the network, data groups are modeled by Levenberg-Marquardt backpropagation algorithm containing two layers and 20 neurons. Weather conditions, holidays, weekends, etc. cause increase and decrease in electricity consumption and temperature is known as the meteorological variable with the highest effect. Therefore, a total of five kinds of parameters were considered in this model. The evaluation of performance of the model was based on mean absolute, mean absolute percentage error and daily peak forecast error. Using the proposed framework, the error values in the forecast in the order of 2.83% have been achieved in the third training. The Hormozgan province load data are used to train and validate the forecast prediction.