基于人工神经网络的高层建筑结构选型

【作者】 郑浩；

【导师】 王全凤；

【作者基本信息】 华侨大学 ， 结构工程， 2001， 硕士

【摘要】 高层建筑结构初步设计是一个涉及面广、综合性强的工作，需要多方面的知识和丰富的工程设计经验，目前国内外尚无理论对其进行充分的研究。本文介绍了一种基于人工神经网络的方法来进行高层结构体系的选择，它充分运用了神经网络高度的非线性、高度的容错性和鲁棒性、自学习、实时处理等特点。本文研究结果表明，利用人工神经网络方法可以解决高层建筑结构选型问题。 本文首先分析高层建筑主要结构型式的特点以及适用范围，提取了高层建筑结构选型的主要控制因素，并以此建立了基于人工神经网络的高层结构选型的数学模型。探讨了BP人工神经网络隐层个数和隐层单元节点数的选取问题，确定了BP神经网络的隐层结构。利用确定的BP网络结构，分别采用传统的BP算法、改进的带动量自适应学习率BP算法，以及L—M算法，分析、解决高层结构选型问题。从本文研究可得出结论，普通的BP算法无法适应土木工程中大规模的数据结构，应采用改进的L—M算法，该算法收敛较普通BP算法快10~2～10~3倍，精度高，能够较好地解决土木工程中的高层建筑结构选型问题。本文还提出使用径向基函数神经网络，该网络的应用在土木工程领域尚未见有关文献公开发表。经本文研究表明，径向基函数神经网络运算速度较普通BP算法快10~3～10~4倍，并且精度高，应用径向基函数神经网络可以高效、高质地完成高层建筑结构选型任务。更多还原

【Abstract】 In the early stage of the design process, the design of tall building is a complex work. It needs various knowledge and professional experience for the structural design. A way concerned about the choice of structural styles is put forward based on artificial neural network in this paper. The qualities of the ANN, high- nonlinear ,high- permissibility of error and high- robustness, self- adaptability, online work, and so on, are adequately used in the research. From the research we know that the method based on ANN can solve the problem on choice of structural styles.First of all, the main characters of tall building are analyzed, and the main factors that can dominate choice of structural styles are picked out, and the mathematics model of the choice of structural styles based on ANN is established. Then, the question on how to choose the numbers of the hidden-layer and neuron is discussed. Two kinds of ANN, BP neural network and radial basis function neural network, are respectively used in the paper. Three kinds of BP algorithm, traditional BP algorithm, improved self-adaptive learning algorithm with momentum, and L-M algorithm, are discussed and compared in the course of analysis. It is concluded that traditional BP algorithm is not suitable for the complex data structure in the field of civil engineering, and improved L-M algorithm whose running-speed is 102~ ~ times faster than traditional BP algorithm, can deal with the problem on choice of structural styles very well. RBFNN is also introduced and used in this paper, the application of which has not been published in the field of civil engineering yet. And from the research, it is concluded that RBFN7N whose running-speed is l0~-10~ times faster than traditional BP algorithm is more efficient than BP neural network. By using RBFNN can solve the problem on choice of structural styles efficiently and correctly.更多还原