Loading...
Please wait, while we are loading the content...
Similar Documents
Universiteit Leiden Opleiding Informatica Investigating an Evolutionary Strategy to Forecast Time Series (2009)
| Content Provider | CiteSeerX |
|---|---|
| Author | Wurdinger, Kerstin |
| Abstract | Recently many investigations have been published about finding good solutions to forecast time series. Different linear and non-linear approaches and hybrid models have been used successfully. Approaches of using Genetic Algorithms and Evolutionary Programming could demonstrate suitability, too. However, a third current mainstream of Evolutionary Algorithms – Evolutionary Strategies – is hardly investigated for this issue. This study is aimed to contribute research to the question whether and how suitable an Evolutionary Strategy can be for forecasting time series. Test sequences of different origins were subject to research. We used a combined model of Autoregressive Moving Average Model and a neural network for our experiments; the overall architecture and the parameters of both components have been optimized by a Covariance Matrix Adaptation Evolutionary Strategy. Our experiments revealed that the Evolutionary Strategy is very well suitable to forecast time series of natural data and computer generated sequences, but less good in forecasting financial time series. 2 |
| File Format | |
| Publisher Date | 2009-01-01 |
| Access Restriction | Open |
| Subject Keyword | Evolutionary Algorithm Evolutionary Strategy Non-linear Approach Evolutionary Strategy Natural Data Autoregressive Moving Average Model Genetic Algorithm Covariance Matrix Adaptation Evolutionary Strategy Universiteit Leiden Opleiding Informatica Investigating Combined Model Forecast Time Series Overall Architecture Forecasting Time Series Good Solution Neural Network Many Investigation Different Origin Time Series Hybrid Model Evolutionary Programming Third Current Mainstream Financial Time Series Test Sequence |
| Content Type | Text |
