شبکه های عصبی مصنوعی

Artificial neural networks (ANN) have been widely used for both prediction and classification tasks in many fields of knowledge; however, few studies are available on animal science. The objective of this study was to prediction of breeding values of weaning weight in Markhoz goats based on the Mixed Model Equation (MME) and Artificial Neural Networks (ANNs). Quality control and calculation of descriptive statistics was performed using the GLM procedure of the SAS statistical package. The pedigree file included 5541 kids produced by 261 bucks and 1616 does. In the first step, genetic evaluations and Best Linear Unbiased Prediction (BLUP) of breeding values for weaning weight was computed with the mixed model equations, animal model by DMU program. Later, unique dataset was introduced to the ANN models by the R statistical program. A variety of models including, multilayer perceptron (MLP), radial basis function (RBF) and Support Vector Regression (SVR) were used to build the neural models. The artificial neural networks were trained and several networks (different hidden layers and nodes/ neurons) were evaluated. In artificial neural networks, the data were randomly divided to two parts (75% training and 25% for test/validation). Best architecture was selected according to the mean square of error and correlation. Correlation between true breeding value (BVs predicted by MME) and estimated breeding value (BVs predicted by ANNs) for MLP, RBF and SVR models were 0.72, 0.49 and 0.73, respectively. Analysis of farm data showed that the MLP and SVR models have higher performance than RBF for prediction of breeding values or ranking of individuals.

This study was conducted to investigate the effects of mixer added fat, crude protein and conditioning temperature on the pellet durability index, and electrical energy consumption during feed production using computational modeling tools. A total of 192 broiler feed samples with different levels of mixer added fat and crude protein in feed components and different conditioning temperatures to determine the pellet durability index, modified pellet durability index and electrical energy consumption during feed production were used. Multiple linear regression and artificial neural network were used to analyze data. Both models had the ability to predict the value of the pellet durability index, modified pellet durability index and the electrical energy consumption during feed production; but the prediction accuracy of the artificial neural network model was higher than that of the multiple linear regression model for all three outputs. Optimization was done using the artificial neural network model, and in these calculations, in order to achieve the highest possible level of pellet physical quality and the lowest possible level of electrical energy consumption, the crude protein amount was 20-20.5% and the conditioning temperature was predicted to be 85 C. However, the amount of fat was predicted to be 1% for the highest amount of pellet physical quality and 4% for the lowest amount of electrical energy consumption during production. In practical conditions, this model can help in more accurate prediction of electricity consumption and the quality of produced feed in order to achieve the optimal situation in feed production factories.

Production systems in bio-Industries rest on resources management and conversion of various forms of energy. This research is investigated and modelled the energy of broiler production in a non-parametric form. The studied samples consist of 70 broiler productions in Ardabil province which were randomly selected from statistical society of the region. In this study, the equivalent energy consumption was estimated in the mentioned industry and the energy indices were calculated, then the equivalent amount of output energy and performance of system were modelled and estimated, using the artificial neural network models. Based on the obtained results, the total equivalent energy of input and output in broiler production were calculated as 153.79 and 27.45 GJ per 1000 birds, respectively. The most consumable input energy in the broiler production of region belongs to the fossil fuel with 61.48% of the total equivalent energy. Based on the results of artificial neural network, the best structure for modelling of energy consumption of broiler production was estimated 5-14-2 structure with five inputs, one hidden layer with 14 neurons and one output layer with two output parameters. The determination coefficient of the best weight combination for estimating the equivalent energy of broiler and manure production for testing data were obtained 93% and 91%, respectively and for validation data were obtained 98% and 95%. In assessing the effectiveness of inputs on the outputs, the fossil fuel showed the highest sensitivity among the production inputs that reveals the needs for revision of the energy resources more than ever.

In this study, the relationships between live body weight at slaughter, carcass characteristics (weights of hot carcass, cold carcass, cold half-carcass, lean meat, fat, bone and fat-tail) and body measurements (body length, chest girth, hip width and shoulder width) were investigated in Lori-Bakhtiari sheep. Two methods of linear regression and artificial neural network (ANN) in prediction of live weight and carcass characteristics were compared.
The data of 58 male lambs were collected at the Breeding Station of Lori-Bakhtiari sheep in Shahrekord city of Chaharmahal and Bakhtiari province, Iran. The lambs were weaned at 90 ± 5 days of age and were fattened in three groups of 60, 80 and 100 days. At the end of fattening period, body measurements and after slaughtering and skinning, all carcass characteristics were measured. A total of 696 records were used for prediction of carcass characteristics by using regression models and ANN. The possibility of predicting carcass characteristics and live weight at slaughter through body measurements was evaluated by choosing the most suitable regression model based on the value of the determination coefficient. The selective regression models also were fitted by artificial neural network then; these two methods were compared based on the value of the determination coefficient and the mean square errors.
The estimated phenotypic correlation coefficients between carcass characteristics and live weight at slaughter were generally positive and relatively high (0.58 to 0.99). The phenotypic correlation coefficients between carcass characteristics and body measurements were estimated between 0.29 and 0.69.
Between all body measurements, the highest correlations are observed between chest girth and carcass characteristics and live body weight. Based on relatively high phenotypic correlation coefficients in this research, prediction of carcass characteristics and live body weight can be possible with high accuracy by using body measurements. The results show that, to prediction of live body weight of Lori-Bakhtiari male lambs, body length, chest girth and hip width must be included in the regression model to obtain an accuracy of 79 percent. The regression equation including live body weight and heart girth could predict hot carcass and cold carcass weights with an accuracy of 97 and 96 percent, respectively. Live body weight alone explains the 94 and 72 percent of the variation of half-carcass and fat-tail weight, respectively. Moreover, the results show that the accuracy of ANN model for prediction of some carcass traits was more than regression models.
Based on results of this research, prediction of carcass characteristics and live weight is possible with relatively high accuracy by using some phenotypic body measurements. Also, results indicate that the artificial neural network technique was much better capable of predicting weight and carcass traits in Lori-Bakhtiari sheep compared to linear regression equations. The results of this study can be useful due to the importance of carcass traits in determination of genetic potential and regulation of animal breeding programs for high meat production.

Calving interval is sought to be substantially affected by management and environmental effects. Using intellectual algorithms of machine learning methods to investigate complex systems are growing and these algorithms could be assumed as right approach to analysis dairy cattle industry data. Artificial neural networks which are part of artificial intelligence, gain from these algorithms. The objective of this study was to find the association between productive and reproductive traits with calving interval over different course of inseminations. It was assumed that extracting this association may improve the management of this trait.
In this research, the productive and reproductive traits from FOKA, an agriculture and animal husbandry, associated to Isfahan Vahdat Cooperative was used. The records were calving interval (day), the length of dry off period (day), number of insemination, total milk production (kg), total fat production (kg), total protein production (kg), adjusted milk production (kg), adjusted fat production (kg) and adjusted protein production (kg). The data were due to 15465 cows in which their parturition date spanned between 1368 and 1393. The data dimensions were reduced using elimination of high level of correlation among variables and principal component analysis before undertaking artificial neural network modeling. For each insemination, we learned a neural network. We used coefficient of determination and root mean square error to investigate the efficiency of neural network. The linear and nonlinear relationship among input variables with calving interval was measured using maximal information criterion.
The similar values of coefficient of determination and root mean square error due to neural network in different insemination obtained which would indicated that there is almost identical associations between calving interval input variables. Higher values of root mean square error obtained in neural network which learned based upon principal component analysis of independent data than elimination one. Therefore, it was concluded that elimination approach is suitable choice in this context. The maximal information criterion showed almost an identical association between calving interval and input variables over different inseminations. In this way, days in milk and milk production showed high amount of correlation with calving interval.
The results indicated that coefficient of determination and root mean square error of different inseminations was similar. Therefore, it is hardly expected that the association between calving interval and independent variables to be different over different inseminations. Based upon these results, it was elucidated that independent variables used in this study could fairly adequately predict the calving interval. Also, we could say that management – environmental conditions over different inseminations had minuscule effect on the associations of calving interval with independent variables.

With using multiple linear regression (MLR), can simultaneously analyses several different variables, but to get the desirable results from the MLR, the samples must be much and accurate. Therefore, this method has high sensitivity and may cause errors in results. In addition, to use this method, the variable must have normal distribution and modification follow from a linear relationship. Artificial Neural Network (ANN) technique is used to solve a wide range of problems in science and engineering, particularly for some areas where the mathematical modeling methods fail. Nowadays, the ANNs are one of the most powerful modeling techniques to model complex nonlinear, multidimensional function relationships without any prior assumptions about the nature of the relationships. Artificial neural network models are different from mathematical modeling approaches in their ability to learn relationships between dependent and independent variables through the data itself rather than assuming the functional form of the relationships. A well trained ANN can be used as a predictive model for a specific application. The prediction by a well-trained ANN is normally faster than the mathematical models. Several authors have shown greater performances of ANN as compared to regression models. An ANN model can predict multiple dependent variables based on multiple independent variables, where a mathematical model is only able to predict one dependent variable at a time. Therefore, this study was designed to evaluate the prediction of production performance of laying hens using the neural networks and nonlinear regressions.

Review the four consecutive, information were obtained from a laying hen farm. Data mining methods include: three-layer perceptron neural network, four-layer perceptron neural network, radial basis function (RBF) neural network and multiply linear and nonlinear regression. In linear model, the variables of age flock, month of production, feed intake have been considered as the predictor variable and production (percent and egg mass production and feed conversion ratio) have been considered as the response variable. Three steps were taken to select an optimal ANN model. The first step was to determine the best number of hidden layers, number of neurons in each hidden layer, and activation function. The best models were selected on the basis of training and prediction accuracy. The second step was to work with the selected models to find the optimum epoch size. The third step was to find the optimum learning rate and momentum values. The evaluating method for selecting the optimal ANN was based on the minimization of deviations between predicted and measured values.

The aim of this study is to obtain an ANN model with minimum errors in training and testing. Nonlinear regression models were compared with Neural Network models. All the models are compared using the coefficient of determination (R2) and Mean Absolute Error (MAE). The results showed that the artificial neural networks compared the regression models and between different artificial neural networks the RBF model had better curve fitting for laying hen production performance indicators included; egg production (% / b/ d), egg mass (g/ b/ d) and feed conversion ratio in front of age and this fact shows that even for spiral data artificial neural network works well. Therefore, we can use these models for complex situations.

The obtained results revealed that the ANN model may efficiently be fitted into the laying hen production performance include percentage and egg mass and feed conversion ratio of hen flocks. Results showed that the method of radial basis function (RBF) neural networks acts better than other models in predicting the production performance of laying hens. So we can conclude RBF model performed better predict laying hen performance.