Initialization of multilayer feedforward neural networks to approximate nonlinear functions in engineering mechanics applications
This paper presents a heuristic initialization methodology for designing multilayer feedforward network networks in modeling nonlinear functions in engineering mechanics applications.In this and previous studies that this work is built upon,the authors do not presume to provide a universal method to approximate arbitrary functions,rather the focus is given to the development of a rational and unambiguous initialization procedure that applies to the approximation of nonlinear functions in the specific domain of engineering mechanics.The applications of this exploratory work can be numerous including those associated with potential interpretation of the inner workings of neural networks,such as damage detection.The goal of this study is fulfilled by utilizing the governing physics and mathematics of nonlinear functions and the strength of sigmoidal basis function.A step-by-step graphical procedure utilizing a few neural network prototypes as “templates to approximate commonly seen memoryless nonlinear functions of one or two variables is developed in this study.Decomposition of complex nonlinear functions into a summation of some simpler nonlinear functions is utilized to exploit this prototype-based initialization methodology.Training examples are presented to demonstrate the rationality and efficiency of the proposed methodology when compared with the popular Nguyen-Widrow initialization algorithm.Future work is also identified.
Multilayer feedforward neural networks initialization function approximation engineering mechanics
J.-S. Pei Eric C. Mai
The University of Oklahoma,Norman,Oklahoma,USA
国际会议
The World Forum on Smart Materials and Smart Structures Technology(SMSST07)(2007年世界智能材料与智能结构技术论坛)
重庆·南京
英文
2007-05-01（万方平台首次上网日期，不代表论文的发表时间）