Backpropagation in Hydrological Time Series Forecasting

Abstract

One of the major constraints on the use of backpropagation neural networks as a practical forecasting tool is the number of training patterns needed. We propose a methodology that reduces the data requirements. The general idea is to use the Box-Jenkins model in an exploratory phase to identify the 'lag components' of the series, to determine a compact network structure with one input unit for each lag, and then apply the validation procedure. This process minimizes the size of the network and consequently the data required to train the network. The results obtained in eight studies show the potential of the new methodology as an alternative to the traditional time-series models.

Forecasting nonlinear time series with feed-forward neural networks: a case study of Canadian lynx data

“…These architectures include: multilayer perceptron (MLP) (Lisi and Schiavo, 1999;Faraway and Chatfield, 1998;Stern, 1996;Hill et al, 1996;Lachtermacher and Fuller, 1995;Jayawardena and Fernando, 1995); recurrent networks (Freeman, 1994, section 6.2); radial basis functions (RBF) Hutchinson, 1994); comparison of MLP and RBF Jayawardena et al, 1996).…”

“…According to Lisi and Schiavo (1999), based on a statistical test, there was no significant difference between FFNN and the chaos models, but both of these models performed significantly better than the traditional random walk model, which is usually the best model for such data. Lachtermacher and Fuller (1995) mention that Farber (1987, 1988) generated two deterministic nonlinear time series, which look chaotic, and found neural networks performed excellently in generating forecasts. They think that neural networks have a key role to play in time series forecasting.…”

“…On the contrary, Lachtermacher and Fuller (1995) suggest using the Box-Cox transformation recommended by Hipel and McLeod (1994) in their modelling framework. They employ the Box-Jenkins method approach to build a suitable neural network structure by identifying the lag components of the time series.…”

“…Therefore, FFNN models should be more widely applied to this type of data not only for forecasting purposes but also for other reasons such as checking the performance of developed statistical models or producing combinations of forecasts as is done by Lachtermacher and Fuller (1995). Especially when a time series is nonlinear or messy and statistical modelling is difficult, FFNN models can be advantageous in providing quick and accurate forecasts of the series.…”

Χωροχρονική Ανάλυση Προσομοίωση Και Πρόγνωση Ξηρασίας Στην Υδρολογική Λεκάνη Πηνειού Ποταμού Θεσσαλίας

“…Άλλες μελέτες έδειξαν ότι η επαναληπτική μέθοδος είναι ελαφρώς καλύτερη για βραχυπρόθεσμες προγνώσεις, αλλά για μεσοπρόθεσμες προγνώσεις η ευθεία μέθοδος πρέπει να προτιμάται για μετεωρολογικούς δείκτες ξηρασίας [Mishra et al, 2007;Vasiliades and Loukas, 2007]. Οι Lachtermacher and Fuller [1995] επισημαίνουν ότι η ευθεία μέθοδος χρειάζεται περισσότερα δεδομένα από την επαναληπτική μέθοδο για την αποφυγή της υπερπροσαρμογής των ΑΝΝs. O Kline [2004], σε μία συγκριτική μελέτη τεχνικών πρόβλεψης πολλαπλών χρονικών βημάτων με νευρωνικά δίκτυα έδειξε ότι η ανεξάρτητη ευθεία πρόγνωση πολλαπλών χρονικών βημάτων είναι καλύτερη από την ευθεία πρόγνωση πολλαπλών εξόδων των νευρωνικών δικτύων αλλά ο αριθμός των δειγμάτων εκπαίδευσης και ο χρονικός ορίζοντας μπορεί να επηρεάσει την υπεροχή της ανεξάρτητης ευθείας πρόγνωσης.…”