Fuzzy Logic Approach - Computational Intelligence in Time Series Forecasting - page 168

mcgtry3.m: f uzzy-forecast of Mackey-G l ass-Series

1.5

1

0.5

0

550 600 650 700 750 800 850 900 950 1000

time(Second)-(nMF=51 & Training Data =1 to 500)

mcgtry 3 .m: fuzzy pre d iction errors f or Mackey-G l ass Chaotic S eries)

0.2

0.15

0.1

0.05

0

-0.05

-0.1

-0.15

-0.2

550 600 650 700 750 800 850 900 950 1000

time(Second)-(nMF=51 & Training Data= 1 to 500)

Figure 4.4(e). Forecasting chaotic series with fuzzy predictor with n = 51 GMFs

Table 4.1. Forecasting performance comparisons for various fuzzy predictors with 500 to
1000 input data sets (evaluation data) for Mackey-Glass chaotic series

Sl. No. No. of GMFs SSE RMSE

1. 17 0.1252 0.0224,

2. 27 0.0389 0.0125

3. 37 0.0255 0.0101

4. 51 0.0164 0.0081

conventional approaches, like the linear predictive method and Gabor polynomial
method, etc . Here, we apply the fuzzy predictor to forecast the future values of
Mackey-Glass chaotic time series data that were directly obtained from MATLAB
version 5.2 as “mgdata.dat” (M ATLAB , 1998).
From the numerical data of the chaotic series, a fuzzy logic system capable of
forecasting the future values of the above time series was developed. For this
purpose, neglecting the first 100 transient data points of the chaotic series, with the
remaining data 1000 rows of the XIO matrix have been built, out of which the first
500 rows (training data) are used for rules generation and the remaining 500 rows