AN IMPROVED MULTILAYER PERCEPTRON BASED ON WAVELET APPROACH FOR PHYSICAL TIME SERIES PREDICTION

摘要: The real world datasets engage many challenges such as noisy data, periodic variations on several scales and long-term trends that do not vary periodically. Meanwhile, Neural Networks (NN) has been successfully applied in problems in the domain of time series prediction. standard NN adopts computationally intensive training algorithms can easily get trapped into local minima. To overcome drawbacks ordinary NN, this study focuses using a wavelet technique filter at pre-processing part NN. However, this study exposed towards an idea to develop model called An Improved Multilayer Perceptron based Wavelet Approach for Physical Time Series Prediction (WMLP) to overcome W-MLP, network with a wavelet technique added network, is trained standard backpropagation gradient descent algorithm tested with historical temperature, evaporation, humidity wind direction data Batu Pahat 5-years-period (2005-2009) earthquake North California 4-years-period (1995-1998). Based obtained results, proposed method W-MLP yields better performance compared existing filtering techniques. Therefore, it be concluded be alternative mechanism ordinary NN one-step-ahead prediction those five events.