A Preliminary Study on Time Series Simulation using NeuralNetworks
| Autor: | Jhih-Min Liou, 柳志民 |
|---|---|
| Rok vydání: | 2005 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 93 In the past, engineers have used many different methods for modeling time series. These methods have many kinds of complicated nonlinear mathematics model. A lot of parameters of simulation model must be established via some procedures. When neural networks generate time series, which do not need define complicated mathematical model or equation. This article applies a suit of technique and mechanism of neural networks, which provides another time series modeling method, to generate specific time series data. In this article chooses nearly normal time series simulation as the example. Show a two-stage approach. In the first stage, a trained replicator neural network is used as a data compression tool. The replicator neural network compresses the vector of the discrete time series data to vectors of much smaller dimension. In the second stage, trained stochastic neural networks or trained back propagation neural networks learns to relate the compressed time series data to another compressed time series data. Then, the replicator neural network was combined with back propagation neural network or stochastic neural network, which trained to learn to associate the time series data with another time series data, to form TGNN1 and TGNN2. That two time series generator neural networks are becoming nearly normal time series generator. Finally, this article uses TGNN1 and TGNN2 to simulate time series data and analyses results. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|