通风管路基温度场神经网络预测模型

ANN PREDICTIVE MODEL FOR PREDICTING TEMPERATURE FIELD EVOLVEMENT WITHIN VENTIDUCT ROADBED

针对青藏铁路沿线冻土特性,对冻土通风管路基温度场约束空间进行了分析、简化。基于系统通风管路基温度场数值仿真分析结果,建立了面向工程设计人员的神经网络预测模型,并开发了相应的优化设计软件系统。该预测模型具有对时间、空间、活动层、布设方案等参数的独立与综合分析能力。同时,对模型的预测效果进行了检验。

Ventiduct roadbed is an active cooling foundation to resist thaw. Many numerical analyses have been conducted to study the effectiveness of the ventiduct layout on the frozen soil foundation, with a lot of data on temperature field evolvement. Obviously, these data are helpful for engineers to optimize the design of ventiduct roadbed. For the maximum uses of the data, the artificial neural network (ANN) method is preferable. Thus, an ANN model needs to be established to predict temperature field evolvement within ventiduct roadbed. In the process of model establishment, the restriction space is predigested with the guidance of expert experiences. The whole predilection consists of three main steps. Firstly, the continuous time region is scattered to dispersed time points. Secondly, the continuous 2D space region is scattered to dispersed space points. Lastly, all the other parameters of scheme design, physical and mechanical properties are retrenched and scattered, such as the ventiduct diameter and spacing, the height of road bank, and the heat exchange parameters of stratum and etc.. The ANN model with improved BP arithmetic is used to predict the temperature field. It is shown that the improved BP neural network can solve ten thousand samples. In the end, an applicable model has been obtained with high efficiency and precision, which can quantitatively predict the temperature field through a continuous period of time. This method, as well as its basic ideas, is significant to predict other quality fields.