Based On Neural Network Multi-response Parameters Optimization Method Research Of Complex Process

In the nation's overall development,quality power displays more and more important position,and constructing quality power is regarded as a major policy in the13 th Five-Year Plan.Parameters optimization is an important part in quality improvement.However,production process is complex in the real production process,so parameter optimization usually has the problem of complex multi-response parameters optimization,including multiple quality characteristics,correlation among the responses,highly complex nonlinear and extreme value problem among the response.Therefore,the traditional parameters optimization methods are difficult to apply.At present,the research mainly uses intelligent algorithm to build complex mapping relation between factors and responses,but did not consider response prediction ability,the correlation among the responses the problem.Therefore,improving method of complex multi-response parameters are given in this paper.Research on an improved optimization method based on response surface model,the main purpose is considering the impact of response predicted ability to the results.Response surface method is used to build regression model between factors and responses.Comprehensive weighted regression model is got by weighting response predicted ability index.Then regarding it as the objective function searches the optimal parameter combination in integrated regression model in the regional scale and gives the improved direction of the parameter.The method can optimize parameters,which the result of optimization inclines to stronger predicted ability of response.In view of the complex response of the problems existing in the production process,this paper gives a parameter optimization method based on neural network.Using the method of weighted principal component analysis converts multiple quality indexes to a single quality performance indicators and using neural network model build good mapping model that can make up the lack of response surface method.Then using high generalization ability of the neural network model searches the optimal parametercombination,getting the ideal parameter design with less test data.The results show that this method can largely improve response values.Based on neural network prediction model,this paper studies an improved weighted principal component analysis method.In view of the complex response of the problems existing in the production process,response predicted index is cannot attach to the requirement of calculation by regression model.In this paper,neural network model is used to build nonlinear model and its mean square error is used to calculate response prediction ability index to adjust the weighted principal component analysis for improving the effect of technological parameters optimization.Integrating theory with practice is research feature of this paper.Theory method are based on engineering case.And the case background is practical engineering problems in projects which is the majority source of the test data.The innovation of theory method research in this paper is that multi-response parameters optimization problems of building the regression model make surface fitting degree as response prediction ability index can attach emphasis on the effect of the prediction ability.In a complex response parameters optimization,the surface fitting degree is not ideal that would lead to deviation in the results.Therefore,neural network model is used to establish a good mapping relationship,and its mean square error is regarded as the evaluation index of response prediction ability adding in the weighted principal component analysis to make optimization prefer to the response which has stronger ability to predict.For solving global optimization problems,this paper proposes a global optimization method based on neural network model.Based on the global forecast,the optimization results is iterated step by step.Therefore,parameter optimization problem,highly complicated nonlinear and multiple extreme point,can be solved.