文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > AN INTELLIGENT PREDICTION METHOD FOR SHORT-TERM TIME SERIES FORECAST ON ENGINEERING EDUCATI
侵权投诉

AN INTELLIGENT PREDICTION METHOD FOR SHORT-TERM TIME SERIES FORECAST ON ENGINEERING EDUCATI

Abstract ? Traditional methods usually encounter the problem in which the predicted results cannot reach a satisfactory need because the overshooting prediction value made by the forecasting model cause a big residual error at the turning points where the

AN INTELLIGENT PREDICTION METHOD FOR SHORT-TERM TIME

SERIES FORECAST ON ENGINEERING EDUCATION

Bao Rong Chang 1, Shiou Fen Tsai 2

Abstract Traditional methods usually encounter theproblem in which the predicted results cannot reach asatisfactory need because the overshooting prediction valuemade by the forecasting model cause a big residual error atthe turning points where the peak or valley observed valuesoccurred. Therefore, this study introduced a intelligentprediction algorithm (including two types) utilized for theapplications of non-periodic short-term time series forecast.This algorithm actually is a hybrid model, combing a greyprediction model and a cumulative least squared linearprediction model, with the technique of automaticallycompensating a possible overestimated predicted value by apotential damped predicted value for those predicting pointshaving extreme peak or valley value. The verification of thisstudy is also tested successfully in three experiments whoseare stock price index, economy growth rate forecasts, andtyphoon moving trace. Furthermore, the results of thisintelligent algorithm also concluded that the proposed oneachieved the best accuracy of predicted values in theseexperiments when compared with other five traditionalforecasting models discussed in the experiments.

Index Terms Grey prediction model, Cumulative least squared

linear prediction model.

prediction accuracy. Thus, the GM(1,1|a) model is oftenutilized in the short-term forecast for years. Although theGM(1,1|a) model has the advantage of simple and fast topredict the future output, the precision is also still arguablein many papers [9][10] since the overshooting predictionvalue made by the forecasting model cause a big residualerror at the turning points where the peak or valley observedvalues occurred. However, 3 points cumulative least squaredlinear model introduced in this study might output andamped predicted value at the turning points. Therefore, aproposed intelligent algorithm combing both aforementionedmethods to introduce a compensation method for solving theproblem of overshooting and damped predicted results so asto improve the prediction accuracy. The verification of thisstudy is also tested successfully in three experiments inwhich involves two types of applications, non-momentuntype and momentum type, respectively.

TRADITIONAL MODELS

Grey Prediction Model

A prototype of grey prediction model GM(1,1|a) isintroduced in the grey system theory 1982 [8].

Step 1: accumulated generating operation once (1-AGO)

x

(1)

INTRODUCTION

Typically, six traditional models- Simple exponential, Holt-Winters smoothing, Regression method, Causal regression,Time series method, and Box-Jenkins [1][2][3] have beenemployed into many scientific or economic applications.Basically, the forecast is obtained through means ofextrapolating a value at the next time instant according to theforecast model’s equation [4][5]. However, some of modelsas shown above, e.g. Holt-Winters smoothing, Regressionmethod, and Box-Jenkins, usually require many observeddata for fitting their models to obtain better approach [5][6]so that these models probably do not be suitable for short-term forecasting case because of only a few data available.Secondly, some of models mentioned above, e.g., Holt-Winters smoothing, Regression method, and Time seriesmethod, are applicable for the specified situation where datadistribution characterized the wave form with regularvariation or normally distributed for a certain system [4][7],for example, the seasonal or cyclical data series. Contrarily,the GM(1,1|a) model introduced in [8] just need a few datafor model construction implied the simple and reasonable12

(k)=

∑x

j=1

k

(0)

(j),k=1,2,...,n

(1)

the original sampled data that is a nonnegative

sequence

Step 2: finding developing coefficient and controlcoefficient by using grey difference equation

x(0)(k)+az(1)(k)=b,k=2,3,...,n(2)

z(1)(k)=αx(1)(k)+(1 α)x(1)(k 1),0≤α≤1

z(1)(k)

x(0)(k):

(3)

: the background value

a=∑

n

x(0)(k)

∑

n

z(1)(k) (n 1)

(n 1)

=

∑

k=2

z(1)(k)2 (z(1)(k))2

∑x

n

(0)(k)z(1)(k)

∑

k=2

(4)

Bao Rong Chang, Dept. of Electrical Engineering, Cheng Shiu Institute of Technology, Kaoshiung, Taiwan brchang@cc.csit.edu.tw

Shiou Fen Tsai, Dept. of International Trades, Kao Yuan Institute of Technology, Kaoshiung, Taiwan sftsai@cc.kyit.edu.tw

International Conference on Engineering Education

1

August 18–21, 2002, Manchester, U.K.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email