【作者】 马祥森；

【导师】 史治宇；

【作者基本信息】 南京航空航天大学 ， 工程力学， 2004， 硕士

【摘要】 利用模态参数进行结构破损诊断是国内外研究的热点和难点。本文提出了基于改进型BP神经网络进行结构破损诊断的方法。BP网络由于具有强大的映射能力、容错性和鲁棒性等优点，非常适合解决破损诊断这类反问题。但随着研究的深入，BP网络在应用中遇到了两个主要问题：(1)难以确定网络结构和初始值；(2)易陷入局部最小解。 针对BP网络的不足，本文提出了一种基于遗传算法(GA)-BP网络的混合技术进行结构破损诊断的方法。该方法采用实数编码的遗传算法优化BP网络的结构及初始参数，从而提高了网络的精度。对比遗传BP网络与传统BP网络对三个仿真算例的识别结果，遗传BP网络的稳定性更好，精度更高，对噪声有很强的鲁棒性，是一种准确有效的结构破损诊断方法。更多还原

【Abstract】 Structural damage detection based on the measured modal data is the prevalent and difficult issue at present. A structural damage detection method based on modified Back-Propagation neural network(BPNN) is presented in this thesis. Owing to the advantages of BP neural network (traditional BP neural network), i.e., non-linear, tolerance and robust, it has played a very important role in structural damage detection. However, Because of the limitation of itself, traditional BP neural network encounters two main problems in practice: (1) lack of systematic means in designing network’s structure and original parameters; (2) convergence to a local minimum.In this thesis, an improved BP algorithms based on genetic algorithm(GA)-BP neural network combined technology has been proposed to solve the two problems mentioned above. The genetic algorithm coding in the real number is engaged to optimize the structure and original parameters of BP neural network, so the network can learn the training patterns more accurately. Three numerical simulations have showed that the GABP has a better stability, precision and robustness than the traditional BP neural network, and is the reliable and accurate methods in structural damage detection.更多还原