Comparison Between The Neural Network And Adaptive Neuro-Fuzzy Inference System For The Optimal Design Of Reinforced Concrete Beams

• Main Authors: Yang, Renpei, 楊人霈
• Other Authors: Yeh, Jiinpo
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/66235375874377152650

碩士 === 義守大學 === 土木與生態工程學系 === 100 === This paper first works on the optimal design of reinforced concrete beams using the genetic algorithm. Given conditions are the span, dead and live loads, compressive strength of concrete and yield strength of steel. Single tensile reinforcement and No. 3 vertical stirrups are adopted. The strength requirements of the maximum positive and negative moments and shear as well as the service requirement of deflection are considered. The constraints are built based on the local reinforced concrete design code and the objective function is the total cost of the steel and concrete. There are 108 sets of optimal data found from the genetic algorithm, which will be divided into three groups: the training set, validation set and test set. The training and validation sets are used to train the neural network and the adaptive neuro-fuzzy inference system. The test set will be substituted into the trained the neural network and the adaptive neuro-fuzzy inference system to examine the efficiency of the network by drawing the scatter plots and calculating the correlation coefficients of the target and network output. The reinforced concrete beams designed in this paper is two-span continuous beams. The selected neural networks are the feedforward backpropagation network and the adaptive neuro-fuzzy inference system, whose input vector consists of the span, width and effective depth of the beam, dead load, compressive strength of concrete and yield strength of steel. The objectives are the optimal positive and negative steel ratios and cost. After substitution of the test data into the trained model, it reveals that the feedforward backpropagation network only needs 6 neurons in the hidden layer to achieve excellent prediction results. The correlation coefficients of the positive and negative steel ratios and cost can reach as high as 0.9992, 0.9980 and 0.9999, respectively. To train the adaptive neuro-fuzzy inference system, the data don’t need to be normalized like the feedforward backpropagation network. First of all, the technique of subtractive clustering is used to group the data. When the cluster radius is set to be 1.4, the prediction results of the test data will be the best with the correlation coefficients of the positive and negative steel ratios and cost being 0.9983, 0.9983 and 0.9996, respectively. This paper confirms that the above two prediction models can effectively predict the positive and negative steel ratios and cost of two-span continuous reinforced concrete beams.