损伤位置分步识别法中参数选择、样本简化等问题的探究

对于多层多跨的框架结构，构件数量众多，直接确定损伤位置十分困难。基于神经网络的损伤位置分步识别法是先构建一个网络来确定结构损伤层的位置，然后针对每层再分别构建一个网络，确定损伤层中损伤柱的编号，于是一个复杂的神经网络被分解简化为几个简单的子网络，这样可以根据模态参数的性质合理选择损伤识别指标，并降低训练样本的数量，从而提高了网络的识别精度。结合分步识别法，建立了详细的多层多跨框架结构数值模型，对模态参数的性质和选择，如何减少训练样本．样本的构造方法等影响识别的因素做了初步探究。

Choice of Parameter and Decrease of Stylebook with the Method of Damage Location Hierachical Identification for Complex Structure

It is difficult to identify the damage location directly for the high frame structure which contains too many components. A method of damage location hierachical identification based on artificial neural networks for high frame structure is introduced. The main idea of this method can be described as follows: Firstly, a network should be designed to identify the damaged story. And then many networks are designed for any story to identify the damaged columns. In this way, a complex identification network is decompounded into some simple sub-networks. Meanwhile more rational identification index can be choose according to the character of modal parameter for each network. Therefore, amount of the train stylebook in hierachical identification are less than that of convention identification and the precision of identification is enhanced. With hierachical identification,a detailed high frame structure model is founded and choice of parameter, decrease of stylebook, method of stylebook make are studied.