Comparison of neural networks prediction and regression analysis (MLR and PCR) in modelling nonlinear system
Different methods for modelling nonlinear system are investigated in this paper. Neural network (NN) techniques, multiple linear regression (MLR) and principal component regression (PCR) are applied to two nonlinear systems which are sine function and distillation column. For the sake of studying th...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Published: |
2007
|
| Online Access: | http://journalarticle.ukm.my/2585/ http://journalarticle.ukm.my/2585/ |
| Summary: | Different methods for modelling nonlinear system are investigated in this paper. Neural network (NN) techniques, multiple linear regression (MLR) and principal component regression (PCR) are applied to two nonlinear systems which are sine function and distillation column. For the sake of studying these three distinctive methods, all the data taken is from simulation which is then be seperated into training, testing and validation. Among those different approaches, the NN approach based on the nonlinear prediction technique gives a very good performance in for both case studies. It is also shown that MLR model suffers from glitches due to the collinearity of the input variables whereas PCR model shows good result in the prediction output. As a conclusion, the NN methods exhibit a consistent result with least sum square error (SSE) on the unseen data compared to the other two technique |
|---|