Application of the Improved BP Neural Network in Ship and Ocean Engineering
Huo-ming ZHANG,Zhi-lin SUN,Ming-zheng GAO
DOI: https://doi.org/10.3969/j.issn.1007-7294.2010.06.006
2010-01-01
Abstract:Artificial neural network (ANN,in short),as a computation model which has the ability to process the questions with high-nonlinear mapping relations,has a wide as well as promising application future in engineering.In the numerical value forecast area,it need not predefine the mathematical model of the sample data collection.Only by means of learning the sample data can it be used for numerical forecast.In this paper,the elementary back propagation neural network (BPNN,in short) is introduced and some useful improvement measures arc presented to solve the problems including the slowly learning velocity,namely the neural network (NN,in short) is difficult to be convergent,and the so-called "platform effect" which means that the exercising error will keep unchanged for many times of iterations in practical application.The improved BPNN algorithm with three layers is presented in detail.In order to verify the improved BP forecast algorithm,the famous XOR problem and a high nonlinear numerical forecast problem with the matrix constructed by 0 and 1 is selected to test the algorithm while satisfactory results are obtained.Finally,combining with two practical problems in ship and ocean engineering,the method of how to use the improved BPNN in numerical value forecast and the problems which should be paid attention to are discussed and some helpful advices are put forward.Based on the outcome of the examples,it indicates that the numerical values forecast ability of the improved BP neural network has the significance to be promoted in ship and ocean engineering.