جستجوی مقالات مرتبط با کلیدواژه "چند لایه پرسپترون" در نشریات گروه "عمران"

Concrete is the most widely used material in the construction industry of the country. Porous concrete has always been the focus of researchers due to its special properties such as high permeability. The present article has determined the design of porous concrete mix with specific strength based on the laboratory program and modeling by using artificial neural network. The experiment program in this research started with 200 mixing plans in the field of porous cylindrical concrete and preparing a database and it was modeled by using different types of multilayer perceptron neural network (MLP). Evaluation indices including regression coefficient and mean square error have been used to check the efficiency of artificial neural network structures. The results showed that the regression coefficient in most of the structures used was above 85% and in the neural network with the optimal structure it was 95%. Also, the evaluated error index in the neural network with optimal structure is determined to be 0.121.

Optimization is one of the most important issues in civil engineering. In this research, artificial neural network (ANN) has been used to optimize the structure of steel structures. The multilayer neural network of perceptron, one of the most widely used neural networks, has been used. Different structures of artificial neural network with different number of hidden layers and number of different neurons have been modeled to achieve the best architecture of artificial neural network model. Neural network models have acceptable success in the optimization process. Dimensional properties of structures have been used in all these models. Models used for optimization have four input parameters, one output parameter is used. A large set of structural models has been used as a database. In the perceptron neural network, networks with different architecture with one and two hidden layers have been used to determine the most accurate network. Extracting and presenting the relationships governing a neural network model gives the user more confidence in using such models, thus facilitating the application of such models in engineering work.

Increasing traffic volume and creating traffic nodes in interurban and urban traffic networks will reduce the efficiency of the traffic network and the desired routes. Anticipating and discovering these traffic nodes as soon as possible can help solve the problem and streamline traffic flow. Artificial neural networks have shown that they are able to perform very well on their own learning capabilities. The main objective of this research is to predict and automatically detect traffic nodes using the intelligent neural network model and compare the model's performance with other models. By using educational data, the artificial neural network can be trained so that it can detect the desired output and perform predictive prediction on target data. The research method is to predict the network architecture from three input parameters and one output parameter. In this research, three types of artificial neural networks are used to predict and automatically detect traffic nodes. The data used in this research is the actual information of the traffic control center of Tehran-Karaj Freeway on a daily, weekly and monthly basis. Initially, the multilayer perceptron of artificial neural network was used, and another neural network used in this research was neuro-fuzzy network and eventually the neural network of the radial-based function was used to examine the success of the two previous networks. Based on the results, the efficiency and accuracy of different models based on the best and most comprehensive set of evaluation indicators were used to evaluate the performance of each of the models as well as to compare their efficiency with each other. Finally, the perceptron model was introduced with optimal efficiency. Comparison of the results of the traffic flow predicted values ​​with the measured values ​​in reality shows that the proposed model satisfies the traffic flow satisfactorily.

The traffic nodes are one of the most traffic-flow problems have been known to freeways. Discover the traffic nodes and fix are one of the main arteries, authorities and researchers in concerns has been. So presenting a suitable model for the detecting of nodes and taking necessary measures to accelerate the traffic-fiowing has certain importance. The purpose of the present research is forecasting and finding of automatically nodes by using neural network capabilities. Research methodology in this study is using three kinds of artificial neural network in order to discover the outcome of auto traffic nodes. That includes some laminated prspetron neural network, neural network of radial basis function and neural network neuro-fuzzy. The data which used in this research are actual data that obtained from control center of Tehran-Karaj freeway traffic daily, weekly and monthly. Finally the results which based on the evaluation indicators show that the perceptron network with three entrance parameters with two hidden layers of 15 neural in hidden layers has the best performance at comparing to the other models that optimize performance as the model which introduced.The research application is reducing the damages and problems which their causes are traffic nodes.

Determining the bearing capacity of piles is an important issue that always Geotechnical engineers focus on. Effect of factors such as environmental dissonance of soil which contains a pile، pile implementation، pile gender and its shape make correct estimation of bearing capacity difficult. Pile load testing as a reliable method could be used in various stages of analysis، design and implementation of piles to determine theaxial bearing capacity of piles. On the other hand، pile load testing، despite high accuracy، imposes high cost and long duration for development projects and it causes limitations in this experiment. Thus acceptance of numerical analysis at geotechnical studies is increasing. The modeling using artificial neural networks is the method that is based on previous data and don’t need to simplify and improve the high reliability coefficient. In this study serious models of multi-layer perceptron neural network، one of the most commonly used neural networks، was used. Network design and factors influencing its behavior in this issue has been studied as a summary. In this study، artificial neural networks are used for prediction of bearing capacity of driven steel piles in sandy soil، in all models four parameters are used as input data which are length and diameter of the pile، the coefficient of elasticity and internal friction angle of soil and the bearing capacity of piles is used as output data. Models have reasonable success in predicting the bearing capacity of piles. In order to evaluation of networks، the different indices such as RMSE، MAE، MAXAE and SDAE were used. To increase the accuracy of predicting bearing capacity، for the network training stage the real tests that has been done at the geotechnical studies of dry dock area hormozgan by POR Consulting Engineers were used. Acording to (Because we) need of more data for training and testing network، several tests on pile bearing capacity، in smaller dimensions were performed in the laboratory. The sixty tests have been performed on piles with various length (35، 40، 45 and 50 cm)، various diameters (20، 25 and 32 mm) and different relative compacted sandy beds (50، 60، 70، 75 and 80%). To perform these tests the device of pile bearing capacity، made in university of TarbiatModarres، was used. Models based on neural networks، unlike traditional models of behavior don’t explain effect of input parameters on output parameters. In this study، by the sensitivity analysis on the optimal structure of introduced models in each stage it has been somewhat trying to response this question.

In-situ tests play important role in any geotechnical investigation. Pressuremeter test can be considered one of the most important in situ tests.This test can be considered one of the most important in situ tests in Geotechnical Engineering. This test is capable to properly estimate deflection parameters of soil. Three types of pressuremeters exist based on their placement in the boreholes: Predrilled pressuremeters (P.D.P), Self-boring pressuremeters (S.B.P) and Push-in pressuremeters (P.I.P). The Predrilled pressuremeters (P.D.P) have been used in this project. Based on expansion of a cylinder that is placed inside the borehole the pressure-volume variation during testing is recorded. In this research, the results of approximately 500 conducted Pressuremeter tests on the soils by Pajohesh Omran Rahvar Ltd (2006-2007) are employed. The number of tests decreased to 400 due to lack of accuracy and also high changes in the range of Pressuremeter modules. The tests have been carried out on the soils of Northwest Iran (Tabriz), South Iran (Kharg Island) and Northeast Iran (Mashhad). The Pressuremeter instrument used is menard pre-boring. Conducted tests in accordance with ASTM-D4719 represented acceptable accuracy. In the current paper, three types of Artificial Neural Network (ANN) are employed in interpretation of pressuremeter test results. First, multi layer perceptron neural network, one of the most applicable neural networks, is used. Then, neuro-fuzzy network, combination of neural-phase network is employed and finally radial basis function, a successful network in solving nonlinear problems, is applied. Neural network models showed prosperity to interpret Pressuremeter test. Soil physical and compaction properties are used in all these models. The applied models own 5 input parameters and 1 output parameters. Hidden layers with different number of neurons are tested in both one and two layers networks so as to select the most proper network architecture. It has been shown that a three-layer perceptron with differential transfer functions and sufficient number of neurons in hidden layer can approximate any nonlinear relationship. Consequently, one hidden layer is used in the present study. The neural network toolbox of MATLAB7.4, a popular numerical computation and visualization software, is used for training and testing of the MLPs. Transfer functions of networks are selected by trial and error. A large complex of carried out tests on the extensive range of fine and course grained soils is used as database. In order to determine the most exact network in the perceptron neural network, some networks with different architecture are employed. Of all neural network models, multi-layer perceptron neural network proved to be the most effective. However, other applied networks have shown favorable performance. Finally, different models have been compared and network with the most outstanding performance is determined. In order to evaluate the capability of model generalization, the performance of mentioned network against inexperienced data has been compared with empirical results. Contrary to conventional behavioral models, models based neural network do not demonstrate the effect of input parameters on output parameters. This research is response to this need through conducting sensitivity analysis on the optimal structure of proposed models. Also, derivation of governing equation for neural network model give more assurance to user to employ such models and consequently this facilitates the application of models in the engineering practices.