An artificial neural network algorithm and time series for improved forecasting of oil estimation: A case study of south korea and united kingdom (2001-2008)

This paper presents an Artificial Neural Network (ANN) algorithm to improve oil production forecasting. ANN algorithm is developed by different data preprocessing methods and considering different training algorithms and transfer functions in ANN models. Bayesian regularization backpropagation (BR), Levenberg-Marquardt back propagation (LM) and Gradient descent with momentum and adaptive learning rate backpropagation (GDX) are used as training algorithms. Also, log-sigmoid and Hyperbolic tangent sigmoid are used as transfer functions. 240 ANN in 6 groups are examined with one to forthy neuron in hidden layer. The efficiency of constructed ANN models is examined in South Korea via mean absolute percentage error (MAPE). One of feature of the proposed algorithm is utilization of Autocorrelation Function (ACF) to define input variables whereas conventional methods use trial and error method. Monthly oil production in South Korea January 2001 to July 2008 is considered as the case of this study.