The main objective of this study is to predict the maximum inter-story drift ratios of steel moment-resisting frame (MRF) structures at different seismic performance levels using feed-forward back-propagation (FFBP) neural network models. FFBP neural network models with varying numbers of hidden layer neurons (5, 10, 15, 20, and 50) were trained to predict the maximum inter-story drift ratios of 5- and 10-story steel MRF structures. The numerical simulations indicate that FFBP neural network models with ten hidden layer neurons better predict the inter-story drift ratios at seismic performance levels for both 5- and 10-story steel MRFs compared to other neural network models.

In this disquisition، has been paid to comparing the performance of static and dynamics neural network by purpose choosing appropriate model in predicting of Tehran Stock Exchange. The data used in this study consists of daily and interval of time 1388/1/5 to 1390/8/30، that Including 616 observation for in sample and out of sample forecasting. Approximately 90% of these observations (556 data) use to estimate coefficients of the model and the rest of them (60 data) use to forecast out of sample. Models are also employed in this research; two stationary neural network models such as fuzzy neural network (ANFIS) and artificial neural network (ANN) and a dynamic regression neural network model (NNARX). The results of this survey indicate that According to Criteria to calculate the forecast error، among Mean squared error (MSE) and root mean square error (RMSE)، Fuzzy neural network model of static، dynamic regression models، neural networks، and finally static artificial neural network models have lowest prediction error، Respectively.

In this paper, the capability of autoregressive static and artificial dynamic neural networks models was compared for forecasting of monthly inflow of Dez reservoir. In previous researches, static and artificial dynamic neural networks models have not been compared for above-mentioned propose. In addition, using artificial neural network model as an autoregressive model is innovation point of this research. Monthly flow data of Dez station in Dez River in years1955 to 2001 is used in this research. Data of 42 former years and 5 recent years are used for Training and testing data set, respectively. Different structure for the static and artificial dynamic neural network models were evaluated by comparing the root-mean-square error (RMSE) of the models. First, static and artificial dynamic neural network models were selected in training phase using data from October 1955 to September 1997. Then, using the selected structures, the monthly forecasted inflow of reservoir was compared with observed data from October 1997 to September 2001. Also, two types of radial and sigmoid activation function and the number of neurons in the hidden layer were investigated in this study. Results showed that the best model to forecast the reservoir inflow is autoregressive artificial neural network model associated with the sigmoid activation function and 17 neurons in the hidden layers. Artificial dynamic neural network model with sigmoid activation function can forecast reservoir inflow for 5 years better than static artificial neural network's model.

The aim of this study was to compare the performance of multiple regression and neural network models to predict the activity of antioxidant enzymes superoxide dismutase (SOD), catalase (CAT), ascorbate peroxidase (APX) and peroxidase (POX) in shoots of wheat (Triticum aestivum), Alvand cultivar in a soil contaminated with cadmium. The treatments consisted of four levels of cadmium (0 (as control), 25, 50 and 100 mg cadmium kg-1 soil), respectively. After 30 days (almost simultaneously with stage of the plant's stem elongation) plant samples were harvested. The following ten different parameters measured included wet and dry weight, chlorophyll a and b, Concentrations of cadmium, copper, iron, manganese, zinc and potassium, respectively. The amount of enzyme activity of SOD, CAT, APX and POX were measured. As next stage, the correlation coefficients between ten parameters and the activity of antioxidant enzymes were determined. The results of multiple regression and neural network models optimized showed that the efficiency of artificial neural network in predicting the activity of SOD and POX enzymes was more of than multiple regression models. Coefficient of multiple determinations (r2) between measured and predicted values of SOD activity for multiple regression and neural network models was 0.76 and was 0.87 respectively. Coefficient of multiple determinations (r2) of POX activity for multiple regression and neural network models was 0.96 and was 0.98 respectively. Coefficient of multiple determinations (r2) between measured and predicted values of CAT activity for multiple regression and neural network models was 0.97 and was 0.98 respectively. In regarding the APX enzyme, coefficients for multiple regression and neural network models was 0.97 and 0.99 respectively. According to the results of this research, generally efficiency of artificial neural network model in predicting the activity of antioxidant enzymes of wheat shoots, under toxicity of Cd was more than the multivariate regression model.

Precise prediction of milk yield is essential for management and modeling of farmer’s income in analysis of cost-benefit. Such that, accurate prediction of future records can decrease recording time. Early estimation of bull breeding value using partly records of parity can lower generation interval and increase selection intensity and genetic progress. Linear model is the most commonly used modeling method in research on different field of science. Artificial Neural Network (ANN) is based on artificial intelligent that uses working principles of human brain. Ease of application and power to model complex functions and relationships is factor of wide use of Artificial Neural Network. Revolutionized use of artificial neural network modeling in different aspects of science in the last two decades, is indication of successful application of this powerful technique to solve wide range of problems in different scientific issues. Therefore, object of current research is prediction milk yield of different parity milk production of dairy cattle using production of first recorded parity and artificial neural networks modeling. Furthermore, results of artificial neural network model compared with linear model. In current research, two sequential records of 2460 dairy cattle of a herd were investigated. Pedigree of used data set contained 2517 individuals. Data divided into two sub data of training and testing, to fitting Artificial Neural Network model. Artificial Network model learned the relationship between output and inputs of training data set. Adequacy parameters of the model investigated using model predicted outputs of testing data set and original outputs of the data. Network structure with the beast adequacy parameters were used for Artificial Neural Network model. Finally, linear model was fitted and compared with artificial neural network model. The best structure of Neural Network had 8 inputs, 4 neuron at first hidden layer, 2 neuron at second hidden layer and output of milk production that inputs were breeding value of average milk yield of first recorded parity, parity, sire group, age at first registered parturition, number of records for each parity and mean, minimum and maximum of recorded days in milk for each parity. The used artificial neural network model, predicted the parity milk production with RMSE and R2 of 7.94 and 0.625, respectively. R2 and RMSE of considered linear model was 0.39 and 26.63, respectively. The applied model of artificial neural network appropriately predicted the subsequent parity production using precedent parity data. This research indicated that use of artificial network model can be beneficial for decreasing recording period for dairy cattle genetic evaluation specially in sire evaluation and will decrease generation interval. The results showed that incomplete data can be used for genetic evaluation using artificial neural network model. Comparison of the results with past reports indicated that use of effective inputs for milk production can increase accuracy and precision of the ANN model.

In this paper, an adaptive cruise control system is designed that is controlled by a neural network model. This neural network model is trained with data resulting from the simulation of a multi-objective nonlinear predictive adaptive cruise control system. For this purpose, first, an adaptive cruise control system was designed using the concept of model predictive control based on a nonlinear model to maintain the desired speed of the driver, maintain a safe distance with the car in front, reducing fuel consumption and increasing ride comfort. Due to the time-consuming computations in predictive control systems and the consequent need for powerful and expensive hardware, it was decided to use the extracted data from the simulation of this designed cruise control system to train a neural network model and use this model to achieve control objectives instead of the predictive controller. Using the neural network model in the cruise control system, despite a significant reduction in computation time, the control objectives were well achieved, and in fact a combination of model predictive controller accuracy and neural network controller speed was used.

Traffic flow on urban highways influenced by various factors and constantly changing from one state to another. Therefore, knowledge of the future traffic flow condition is remarkably important. This paper analyzing volume data to predict future traffic flow. To predict traffic flow we can use different statistical or computational intelligence methods, such as artificial neural networks.Using of artificial neural network model with crude signal input is not better than simple statistical models. As the traffic signal has high noise, in order to access the details of signal we use wavelet transform to turned it into several Frequency sections. These sections are considered as an input to the artificial neural network model. Flow forecasting with wavelet transform-artificial neural network model is more accurate than simple artificial neural network model.

In this study, long-term memory and dynamic behavior of daily flow time-series of Khorramabad River, which its basin is mountainous and has urban land use, is investigated by Hurst exponent. The Hurst exponent of runoff signal of Khorramabad River during 1991-2014 period was obtained as 0.8. This value shows long-term memory and nonlinear, dynamic signal of this river’s runoff. By applying neural network and wavelet transforms, the rainfall-runoff time-series of this river was simulated. In this respect, by taking the time-series of rainfall and rainfall-runoff as input to the artificial neural network and wavelet-neural network hybrid, four models including: 1) rainfall, neural network, 2) rainfall-runoff, neural network, 3) rainfall, wavelet-neural network and 4) rainfall-runoff, wavelet-neural network were developed. In the hybrid models of wavelet-neural network, time-series of rainfall and runoff were decomposed to high-frequency and low-frequency sub-signals. Results of evaluating the accuracy and efficiency of the four models showed that the wavelet–neural network model correctly simulated the runoff behavior with the best efficiency at 99% confidence level. Comparison of the results of wavelet–neural network model to the neural network model, using Morgan-Granger-Newbold, showed significant superiority of the first model. Also, results of evaluating signal error of the four implemented models, using two tests of Von-Neumann and Buishand test, showed that there is a significant substitution point in the signal error of the neural network model and signal of rainfall-runoff model. Therefore, existence of very different monthly and periodical fluctuations in 1991-1998 and 1999-2014 in the behavior of rainfall-runoff leads to reduction of efficiency and precision coefficient of neural network model. While, in the hybrid model of wavelet-neural network, allocation of relative weight to each sub-signal, has effectively reduced the short-term, average and long-term fluctuations in modeling error.

In this research the comparative prediction of Iran''s banking system (included 14 banks) was carried out by using econometric and artificial neural network models. Accordingly، at first، by using the Kohonen neural network model، the considered banks were divided into two categories of high performance and low performance groups and then using the output of Kohonen neural network model، financial proportions and Panel Data econometric model، the performance of Iran''s banking system was estimated for the period 2004-2010 and finally by using models evaluation criteria، the performance of Panel Data and ANN models was compared. The results of Kohonen neural network model indicated that from 14 considered bank، 4 banks belong to high performance group and 10 banks are belong to low performance group. Also the results of Panal Data estimations showed that “capital income/total income «portion has the lowest and “cash/total deposits»، has the haighes effect on the Iran''s banking system. Finally the results of models comparison stated that the ANN model outperforms the Panel Data model to predict the performance of Iran''s banking system.

This paper is aimed to compare stock index forecasting using hybrid models based on Genetic Algorithm (GA) and Harmonic Search (HS) with Artificial Neural Network (ANN). The most relevant technical indicators as inputs and the optimal number of neurons in hidden layer of Artificial Neural Network designated by metaheuristics including Genetic Algorithms and Harmony Search. Daily price index of Tehran Stock Exchange from 21 December 2012 to 21 December 2015 applied to predict and test stock index. The accuracy of forecasting of three models including Regular Artificial Neural Network model, hybrid neural networks based on GA and hybrid neural networks based on HS is evaluated by the prediction error. The results show that the accuracy of prediction in Metaheuristics models such as Genetic Algorithms and Harmony Search in the test period is higher than normal Artificial Neural Network. Also prediction by hybrid neural network model based on harmony Search during the test period compared to hybrid Artificial Neural Network model based on Genetic Algorithm is more accurate.