| In large-scale and complex Multidisciplinary Design Optimization system problems, as for great amount design variables, numbers of response value, different kinds of constraints and complex design space’s exist, it is hard to creat and solve optimization model to get result, and it usually couldn’t get the best optimization result. Then approximation algorithm is rightly to solve the problem like those the previously mentioned.Approximation method has an alias, Agent mode, it could simulate the part or all design space to get the explicit expression which is cannot be known directly, through one or more analysis of the subject data. It is able to get the approximation relation to deal with the multidisciplinary coupling problem, and then be applied in optimization process. At the same time, it is convenient to exchange the information when using approximation function, and is good for reducing the error of value.As for the ability of multidisciplinary crossing learning approaching, Artificial Neural Networks is widely applied in many areas. Traditionally, the Artificial Neural Networks has four basic mode:the forward feedback mode, the backward feedback, self-organization mode and random network mode. It is widely applied in the signal processing and pattern recognition, but for most forward feedback network, there is a common weak point that is the extremely nonlinear between network and parameter, the parameter in learning is likely to fall into partly MIN.RBF (radial basis function) neural network applies a very effective method for this kind of multi-layer forward feedback network. At first, RBF has simple layer network with only three layers, at the second it can avoid the complicated calculations like BP neural network, so, enhancing the learning ability, at the last, it can solve the partly Min problem, besides RBF neural network can replace any BP approaching neural network. with DOE technology, RBF algorithm is better and better, and widely applied. |
PDF Full Text Request