基于进化神经网络的深基坑时空效应分析

ANALYSIS OF TIME-SPACE EFFECT OF DEEP-FOUNDATION PIT BASED ON GA-ANN MODEL

深基坑时空效应分析中,常采用数值模拟及拟合经验公式等方法进行研究,因深基坑时空效应的影响因素复杂,各种因素难以量化,常规的分析方法适应性不强。而目前时空效应分析中,一般只考虑时间或空间效应的影响,尚未建立同时考虑时空效应的人工智能分析模型。对于处理高度非线性的复杂问题,神经网络有独到的优势;基于进化算法的神经网络还具有全局优化功能。在详细分析了时空效应影响因子的基础上,尝试建立了利用进化算法分步优化神经网络结构和初始权值的深基坑时空效应分析模型。经与BP模型、网络结构及权值同步优化模型对比分析表明,分步优化模型可提高网络的辨识精度及分析效率,更符合实际情况。

Methods such as numerical simulation and experience function imitated and so on were adopted to study on time-space effect of deep foundation pit. Because of complexity in influence factors and difficulty in measure of various factors, normal analysis methods were not strong in adaptability. At present, only time or space effect was considered in analysis of time-space effect, and analysis method of time-space effect considered simultaneously based on artificial intelligence was not established. It is shown that high non-linear complex problems can be solved easily by artificial neural network as well as the method of ANN（artificial neural network） based on GA（genetic arithmetic） has optimizing function in the whole field of solution. The st/ucture and weight value of the network are optimized in-phrase in GA-ANN model, and the weight value of the network and influence factors of problem are two kinds of different qualities. When the two kinds of variables are regarded as input variables to train the network, it may lead to chaos of the network, thus it will decrease identification precision of the network. The model of time-space effect on deep foundation pit is established that the structure and weight value of the network can be optimized by stages with GA based on labor of influence factors. The comparative analysis with the BP（back propagation） model and the GA-ANN model optimized in-phrase shows the model optimized by stages can improve identification precision and analysis efficiency of the network, thus it will be closer to the fact.