Neuro-fuzzy Approach - Computational Intelligence in Time Series Forecasting - page 244

Trnfmimo1:SSE-Vs-Epoch

300

250

200

150

100

50

50 100 150 200 250 300

No of Epochs

T rnfm im o 1 : N e uro -fuzzy o utp ut-V s-A ctua l

150

100

50

0

-5 0

-1 0 0

-1 5 0

0 500 1000 1500

Tim e

Figure 6.7(a). Training performance of Takagi-Sugeno-type multi-input multi-output neuro-
fuzzy network with n = 4 inputs, m = 3 outputs, M = 15 fuzzy rules and 15 GMFs for short-
term forecasting of electrical load time series when trained with proposed backpropagation
algorithm. Backpropagation algorithm training parameters: K = 0.0005, J = 0.5, mo = 0.5,
maximum epoch = 300, training (pre-scaled) data = 1 to 500 rows of XIO matrix, initial SSE
= 324.6016 (with random starting parameter), final SSE = 23.8580, data scaling factor =
0.01.

sim fm im o .m : N e uro -fuzzy o utp ut vs. A c tua l

150

100

50

0

-5 0

0 500 1000 1500 2000 2500 3000 3500

tim e

sim fm im o .m : N e uro -fuzzy p re d ictio n e rro r

150

100

50

0

-5 0

-1 0 0

-1 5 0

0 500 1000 1500 2000 2500 3000 3500

tim e

Figure 6.7(b). Forecasting performance of Takagi-Sugeno-type multi-input multi-output
neuro-fuzzy network with n = 4 inputs, m = 3 outputs, M = 15 fuzzy rules and 15 GMFs for
short-term forecasting of electrical load when trained with proposed backpropagation
algorithm. Data 1 to 1500 correspond to training data and data 1501 to 3489 ( i.e. row 501 to
1163 from XIO matrix) represent the forecasting performance.