Yong Hyun Cho

The KIPS Transactions:PartB , Vol. 8, No. 4, pp. 327-334, Aug. 2001
10.3745/KIPSTB.2001.8.4.327,   PDF Download:

Abstract

This paper proposes a regression analysis with high reliability using neural networks of a hybrid learning algorithm. The proposed algorithm is a learning algorithm based on the steepest descent and dynamic tunneling. The steepest descent is applied for high-speed optimization, and the dynamic tunneling is also applied for converging to the global optimum by estimating the new initial weights for escaping the local minimum. And the adaptive principal component analysis (PCA) is also applied for solving the limits of regression model by reducing efficiently the dimension by extracting the features of the given independent variables. The proposed neural networks has been applied to regress the Ammonia producing process (3-independent variable) and the mobile oil cost (10-independent variable). The simulation results show that the proposed models have better performances of the learning and the regression, in comparison with those using the conventional backpropagation-based multilayer perceptron (BPMLP) and the BPMLP using the learning patterns without reducing the dimension by PCA, respectively. And we confirm that the zero mean normalization of learning patterns makes better performances of the regression neural networks.

Statistics
Cite this article