Abstract:
Many real word product or process design problems involve multi response problems which have stochastic nature. This paper proposes a hybrid approach involved genetic algorithm and artificial neural network methodology to solve these problems. Usually, in these problems the relationship between responses and independent variables is indeterminate; therefore to generate required input data we are interested to use a method to approximate this relationship. Artificial Neural Network (ANN) is a methodology employed in this research to evaluate linear and nonlinear relationship between variables. We model the statistical multi response problem by three different multi objective decision making (MODM) techniques. Moreover, four different genetic algorithms are proposed in which four pairwise multiple comparisons statistical tests are used to control the random nature of the problem. Finally, the performance of the proposed methodology is demonstrated using a tow way analysis of variance (ANOVA) for a numerical example and the results are compared statistically..
Machine summary:
"Objective function evaluation After producing the new chromosomes by crossover and mutation operators and approximating the responses using suitable artificial neural network, we evaluate each chromosome by fitness functions described in section 3 and use the result to reproduce a new generation.
Table (1) is here After statistical normalization of the observation, the architecture of the neural network for each response is determined to generate the required input data based on the number of hidden layers, activation function and the number of neurons in each layer.
First in terms of four test procedures and three modeling method used in this research, using MATLAB computer software, the algorithms run 30 times, each time changing its parameters in their corresponding ranges and obtained the responses values for each algorithm.
7. Conclusions In processes with stochastic nature and data sets showing nonlinear and complex relationships between control and response variables and complicated optimization models, to overcome the limitation and weakness of the methods which are used before in this cases, in this paper a new approach based on a neural network, three modeling techniques, and four different genetic algorithms was proposed to solve multiple response statistical optimization problems.
The neural network approach generated the required input data and the MODM techniques modeled the problem and the four different structures of genetic algorithm optimized the model to find the levels of the control factors.
Employing Artificial Neural Network into Achieving Parameter Optimization of Multi Response Problem with Different Importance Degree Consideration."