應用軟性計算於時間序列預測之研究

Study on Time Series Forecasting Problem by Applying Soft-Computing

王彥文(Yen-Wen Wang)

指導教授 ： 張百棧

元智大學/工程學院/工業工程與管理學系/博士(2007年)

https://doi.org/10.6838/YZU.2007.00014

摘要

在一般的預測問題中，時間序列預測佔了相當重要的一部份，許多時間序列的預測模型也因此而被發展出來，一般來說這些方法可以簡單的區分為統計預測模型以及軟性計算應用預測模型。軟性計算技術已經發展多年，與傳統統計方法不一樣的地方是它可以容忍一些干擾存在於我們所蒐集的資料當中，包含：不精確的資料、具不確定性的資料、不明確的資料等等，因此本研究將利用軟性計算來建構一個混合預測方法，並將該方法應用在預測的問題上。本研究第一部份將利用基因演算法來優化類神經網路的結構，第二部分將利用改良的模糊規則觀念來修正之前的模型，利用相似度的觀念將相似的資料合併，如此，該混合預測方法將能夠節省運算時間、並準確的預測。在最後的實驗結果章節中，我們可以發現該方法的平均誤差僅有2.11%，若是比較其它許多預測模型，都具有相當的優勢，因此我們可以成功的應用該混合預測方法於預測的問題上。

關鍵字

軟性計算 ； 模糊規則 ； 演化式神經網路 ； 銷售預測

並列摘要

Time series forecasting is one of the important problems in the time series analysis. Many different methods have been developed in this field, and these methods can be distinguished into statistic methods and soft-computing methods roughly. Soft-computing method has been developed for many years. Unlike the traditional statistic methods, it tolerates the interference of the time series data, such as imprecision, uncertainty, partial truth, and approximation. The goal of this research is to develop a hybrid method and successfully applied this method in the forecasting problem. In the first part of this research, we will provide a Soft-computing based hybrid method to improve the structure of neural network. The second part of the research will use the rule-based model to gather the data as rule set on the similar tendency in order to reduce the solving time and increase the accuracy of the forecasting method. The experimental results reveal that the MAPE for our hybrid method is 2.11% which is the best compared to others. In summary, our hybrid method can be applied practically as a sales forecasting tool in the PCB industry.

並列關鍵字

Soft-Computing ； Fuzzy Rule ； Evolving Neural Network ； Sales Forecasting

參考文獻

Barbounis, T.G. and J.B. Theocharis, “Locally Recurrent Neural Networks for Long-term Wind Speed and Power Prediction,” Neurocomputing, 69, 4-6, 466-496, Jan. 2006.

Bayhan, G. M., and M. Bayhan, “Forecasting Using Autocorrelated Errors and Multicollinear Predictor Variables,” Computers and Industrial Engineering, 34(2), 413-421, 1998.

Cao, L.J., and F.E.H. Tay, “Support Vector Machine with Sdaptive Parameters in Financial Time Series Forecasting,” IEEE Transactions on Neural Networks, 14(6), 1506 – 1518, 2003.

Chai, M.L., S. Song, and N.N. Li, “A Review of Some Main Improved Models for Neural Network Forecasting on Time Series,” Intelligent Vehicles Symposium, 866-868, 6-8 Jun. 2005.

Chang, T.C. and S.J. Lin, “Gray Relation Analysis of Carbon Dioxide Emissions from Industrial Production and Energy Uses in Taiwan,” Journal of Environmental Management Vol. 56, Issue: 4, Aug., 1999, 247-257.