Application Of Data Mining And Neural Network To Civil Engineering

Data mining (DM) aims at drawing implied and useful information/knowledge from massive incomplete, noisy, blurry, and stochastic real data; while neural network is a frequently used tool for DM. This thesis addresses how to apply the DM technique and neural network to civil engineering.The choice of the value of the training error of a neural network is a challenging problem. To our knowledge, the majority of existing methods for this problem are based on the simple trial process. With application to civil engineering, this thesis advises a new method for determining the training error of a neural network by taking into account the stochastic characteristic of the training samples. In this method, the confident interval of true value corresponding with actual measured value is calculated. As for a certain value of the training error, if the predicted value of a neural network model lies in true value confident interval, it is believed that the true value of a neural network output variable be obtained, and the training process of a neural network is over. Thus, an initial value of the training error can be determined, and it is half the confident interval of true value. Experimental results show that the proposed method can decrease the time overhead required by the training of a neural network.In the past, the predicted results produced by neural network models were often evaluated by using the relative error has been widely applied to this problem. But it is no longer fitted for random training samples. Therefore, based on mathematics statistics theory, this thesis presents a new evaluation method for evaluating these predicted results by introducing the notion of the correction rate of prediction. Then, the confident interval of correction rate of the predictions is deduced. Indeed, it can give an indication as to the likely future of a neural network model. When applied to civil engineering, experimental results justify that this evaluation method is useful, and its conclusion is close to engineering actualities.A set of software for high performance concrete fabrication is also developed with Delphi. enhance the production efficiency of high performance concrete fabrication. The software supplies some important fuction blocks, such as data management, neural network training, neural network chocking, performance predicting and costs calculating. A practical test show that the software is actually useful, and has engineering value, and it can enhance the production efficiency of high performance...