改进的BP神经网络在船舶与海洋工程中的应用研究(英文)

Application of the Improved BP Neural Network in Ship and Ocean Engineering

人工神经网络作为一个具有高度非线性映射能力的计算模型,在工程中具有广泛的应用前景。在数值预测方面,它不需要预选确定样本的数学模型,仅通过学习样本数据即可以进行预测。文章介绍了BP神经网络,并针对实际应用中收敛速度慢,平台效应等问题对网络进行了改进并优化,详尽地给出了改进的三层BP神经网络数值预测算法。为测试该算法,选用了著名的XOR(异或)问题和和一个高度非线性的0-1矩阵预测问题对其进行了验证。计算结果表明文中算法能给出令人满意的精度。最后结合船舶与海洋工程的两个实际问题,探讨了利用改进的BP神经网络进行数值预测的方法和应该注意的问题,并给出了一些有益的建议。实践表明,文中给出的改进的BP神经网络数值预测算法值得在船舶与海洋工程中加以应用并推广。

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 are 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.