Elman Neural Networks

Forecasting of Short-Term Daily Maximum and Minimum Temperature Using Elman-Recurrent Neural Network
Full-size image (145 K)

by: Alfonsus J. Endharta

Neural Network (NN) is one of the methods mostly used for forecasting non-linear data in many countries. NN which is mostly used id many researches is Feed-Forward Neural Network (FFNN) or Autoregressive Neural Network (AR-NN). AR-NN is not able to catch and represent moving average (MA) order effects in time series data. This research is done to study another NN type application, i.e. Elman-Recurrent Neural Network (Elman-RNN) which is able to represent MA order, and to com-pare the forecast accuracy with Seasonal Autoregressive Integrated Moving Average (SARIMA) mo-del. Maximum and minimum temperature data from BMG Meteorology Station Kalianget is used as the case study. The analysis results that the best ARIMA model for forecasting short-term maximum temperature is ARIMA([1, 19], 1, 1)(0, 0, 1)365 and ARIMA([1, 6, 14], 1, [1, 5, 11])(1, 0, 0)365 for the minimum temperature data. Both model give white noise residuals, but they don’t follow the normal distribu-tion. Elman-RNN inputs which is studied and applied for forecasting the case study data are ARIMA lag inputs. Therefore, there are three nets built for each maximum and minimum temperature data. This net use one hidden layer with tangent sigmoid function and one output with linear function. The forecast accuracy comparison based on out-sample MAPE value shows that the best net for the maximum temperature data is Elman-RNN(4, 5, 1) and Elman-RNN(11, 9, 1) for the minimum tempe-rature data. Those nets are the best model for forecasting maximum and minimum temperature in BMG Meteorology Station Kalianget compared with SARIMA model.

Source: Alfonsus J. Endharta

A Bradford Book / The MIT Press Rethinking Innateness: A Connectionist Perspective on Development (Neural Network Modeling and Connectionism)
Book (A Bradford Book / The MIT Press)

You might also like:

Recurrent neural network state trace for 5 dimensions - symmetric
Recurrent neural network state trace for 5 dimensions - symmetric
Aggregation of 10 Simulated e-puck Robots
Aggregation of 10 Simulated e-puck Robots
Guppies - Evolving neural networks (wip) Part II
Guppies - Evolving neural networks (wip) Part II

Soros has slammed US Treasury Secretary Hank Pau

2008-09-17 12:45:33 by MasterOFDisaster

Billionaire investor George Soros has slammed US Treasury Secretary Hank Paulson for behaving in the same manner as bankers in the 1930’s and mishandling a financial crisis that threatens a repeat of the Great Depression.
Soros told BBC Newsnight that the world was merely at the beginning of a financial storm and warned, “We mustn’t allow the financial system to collapse as it did in the 1930s.”
Referring to Hank Paulson, the US Treasury Secretary, Soros stated, “The way Paulson is handling the situation is reminiscent of the way the bankers handled it in the 1930s.”
He added: “The financial system has gone overboard and the financial engineering has grown to big, it takes up too big a share in the world’s resources.”
“Now it is shrinking. When it becomes...

How to Play the Euro Now  — Wall Street Journal
Oanda, a Forex.com rival, recently introduced a currency-trading system that allows investors to sell the euro against a trade-weighted basket of the US dollar, UK pound, Swiss franc, yen, Norwegian krone and Swedish krona.

A Bradford Book Exercises in Rethinking Innateness: A Handbook for Connectionist Simulations (Neural Network Modeling and Connectionism)
Book (A Bradford Book)

Related posts:

  1. AI Neural Networks
  2. Spline Neural Networks
  3. Code: Neural Network
  4. Language Neural networks