eduzhai > Physical Sciences > Earth Sciences >

Effects of Model Structural Complexity and Data Pre-Processing on Artificial Neural Network (ANN) Forecast Performance for Hydrological Process Modelling

  • carsar
  • (0) Download
  • 20210408
  • Save

... pages left unread,continue reading

Document pages: 18 pages

Abstract: The choice of a particularArtificial Neural Network (ANN) structure is a seemingly difficult task; worthyof relevance is that there is no systematic way for establishing a suitablearchitecture. In view of this, the study looked at the effects of ANNstructural complexity and data pre-processing regime on its forecast performance.To address this aim, two ANN structural configurations: 1) Single-hiddenlayer, and 2) Double-hidden layer feed-forward back propagation network were employed. Results obtained revealed generallythat: a) ANN comprised of double hidden layers tends to be less robust andconverges with less accuracy than its single-hidden layer counterpart underidentical situations; b) for a univariate time series, phase-spacereconstruction using embedding dimension which is based on dynamical systemstheory is an effective way for determining the appropriate number of ANN inputneurons, and c) data pre-processing via the scaling approach excessively limitsthe output range of the transfer function. In specific terms consideringextreme flow prediction capability on the basis of effective correlation:Percent maximum and minimum correlation coefficient (Rmax and Rmin ), on the average for one-day ahead forecast duringthe training and validation phases respectively for the adopted networkstructures: 8 7 5 (i.e., 8 inputnodes, 7 nodes in the hidden layer, and 5 output nodes in the output layer), 8 5 2 5 (8 nodes in the input layer, 5 nodes in the first hidden layer, 2nodes in the second hidden layer, and 5 nodes in the output layer), and 84 3 5 (8 nodes in the input layer, 4 nodes in the first hidden layer, 3 nodesin the second hidden layer, and 5 nodes in the output layer) gave: 101.2, 99.4; 100.2, 218.3; 93.7, 95.0 in allinstances irrespective of the training algorithm (i.e., pooled). On the other hand, in terms of percent of correctevent prediction, the respective performances of the models for both low andhigh flows during the training and validation phases, respectively were: 0.78,0.96: 0.65, 0.87; 0.76, 0.93: 0.61, 0.83; and 0.79, 0.96: 0.65, 0.87.Thus, it suffices to note that on the basis of coherence or regularity ofprediction consistency, the ANN model: 8 4 3 5 performed better. Thisimplies that though the adoption of large hidden layers vis-à-vis correspondinglarge neuronal signatures could be counter-productive because of networkover-fitting, however, it may provide additional representational power. Basedon the findings, it is imperative to note that ANN model is by no means asubstitute for conceptual watershed modelling,therefore, exogenous variables should be incorporated in streamflow modelling and forecasting exercise because of theirhydrologic evolutions.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×