Artifi cial neural networks applied to forecasting time series
نویسندگان
چکیده
have aroused great interest in fi elds as diverse as biology, psychology, medicine, economics, mathematics, statistics and computer science. The main reason underlying this interest lies in the fact that ANN are general, fl exible, nonlinear tools capable of approximating any sort of arbitrary function (Hornik, Stinchcombe, & White, 1989). Due to their fl exibility as function approximators, ANN are robust methods in tasks related with pattern classifi cation, the estimate of continuous variables and time series forecasting (Kaastra & Boyd, 1996). In this latter case, ANN offer several potential advantages with respect to alternative methods —mainly ARIMA time series models— when it comes to dealing with problems concerning nonlinear data which do not follow a normal distribution (Hansen, McDonald, & Nelson, 1999). The fi rst advantage lies in the fact that ANN are extremely versatile and do not require formal specifi cation of the model or the fulfi lment of a certain probability distribution for the data. Regarding the second advantage, Masters (1995) shows that ANN are capable of tolerating the presence of chaotic components in better conditions than most alternative methods. This capacity is particularly important due to the fact that many of the relevant time series possess systematic chaotic components. In this way, ANN have been successfully applied in time series forecasting in different knowledge fi elds such as biology, fi nance and economics, energy consumption, medicine, meteorology and tourism (Palmer, Montaño, & Franconetti, 2008). The most widely used neural network in time series forecasting has been the MLP (Multilayer Perceptron) (Bishop, 1995). However, recent studies have evidenced the excellent performance of other neural network models with respect to the MLP model in this type of task (Liu & Quek, 2007). As far as ANN are concerned, the literature has not established a general procedure of application of this technique in time series forecasting, but rather aspects in which there is no agreement between several authors have been presented (Nelson, Hill, Remus, & O’Connor, 1999). In this sense, this study offers a description and a comparison of the main ANN models that have been shown to be useful in time series forecasting, as well as a standard procedure for the practical application of ANN in this type of task. The models analyzed are: the Multilayer Perceptron (MLP), Radial Basis Function (RBF), Generalized Regression Neural Network (GRNN) and Recurrent Neural Networks (RNN). Psicothema 2011. Vol. 23, no 2, pp. 322-329 ISSN 0214 9915 CODEN PSOTEG www.psicothema.com Copyright © 2011 Psicothema
منابع مشابه
Electricity demand forecasting using a SARIMA-multiplicative single neuron hybrid model
The combination of SARIMA and neural network models are a common approach for forecasting nonlinear time series. While the SARIMA methodology is used to capture the linear components in the time series, artifi cial neural networks are applied to forecast the remaining nonlinearities in the shocks of the SARIMA model. In this paper, we propose a simple nonlinear time series forecasting model by ...
متن کاملAN EXTENDED FUZZY ARTIFICIAL NEURAL NETWORKS MODEL FOR TIME SERIES FORECASTING
Improving time series forecastingaccuracy is an important yet often difficult task.Both theoretical and empirical findings haveindicated that integration of several models is an effectiveway to improve predictive performance, especiallywhen the models in combination are quite different. In this paper,a model of the hybrid artificial neural networks andfuzzy model is proposed for time series for...
متن کاملA hybrid computational intelligence model for foreign exchange rate forecasting
Computational intelligence approaches have gradually established themselves as a popular tool for forecasting the complicated financial markets. Forecasting accuracy is one of the most important features of forecasting models; hence, never has research directed at improving upon the effectiveness of time series models stopped. Nowadays, despite the numerous time series forecasting models propos...
متن کاملA NEW APPROACH BASED ON OPTIMIZATION OF RATIO FOR SEASONAL FUZZY TIME SERIES
In recent years, many studies have been done on forecasting fuzzy time series. First-order fuzzy time series forecasting methods with first-order lagged variables and high-order fuzzy time series forecasting methods with consecutive lagged variables constitute the considerable part of these studies. However, these methods are not effective in forecasting fuzzy time series which contain seasonal...
متن کاملCombination of Transformed-means Clustering and Neural Networks for Short-Term Solar Radiation Forecasting
In order to provide an efficient conversion and utilization of solar power, solar radiation datashould be measured continuously and accurately over the long-term period. However, the measurement ofsolar radiation is not available to all countries in the world due to some technical and fiscal limitations. Hence,several studies were proposed in the literature to find mathematical and physical mod...
متن کامل