The Research And Application Of Assistant Decision-making System Based On Many Intelligent Computing Methods

Enterprise's financial crisis predicts is the non-linear prediction, there is a complicated association decision relation between each influence factor, and the data in reality are continuous, it is very difficult to be used in the categorized machine to study directly. Just the characteristic and complexity of the financial early warning problem itself, make us very difficult to carry on modeling to it with the traditional statistics analytical method. After analyzing the characteristic of the early warning problem, we merged many kinds of soft computing methods to construct the prediction model. The overall course is as follows: Firstly, take consistency level of decision, average information entropy and degree of discretization as evaluation criteria of the result of discretization. Then utilize the overall search of genetic algorithm to find the optimized cut points. After discretization by the optimized cut points, train the BP neural network with the samples in training set. When finishing the training, we use the BP neural network to predict financial crisis of listed company.Main work of the thesis:1. Have improved calculation model of the consistency level in Rough Set. Standard calculating model of the consistency level is very complex. Especially when we search for the optimum cut points with the genetic algorithm, each change of cut points set will involve calculating consistency level again, it influence the efficiency of algorithm badly .Hence we construct "Rough Set Matrix of cut points" and use the number of different type pairs which the cut point can distinguish to represent the consistency level. We also adopt the tactics of space exchange time, create view which ranks the recorders order by attribute for every condition attribute. So we can omit the operation of arrange in an order, in this way the complexity of the algorithm can be reduced O(n· log n). And view is only logic index of metadata, so space that consume even quite little.2. Have proposed a new kind of optimization GA based on Rough Set and information of entropy to make the condition attributes discrete. Have adopted 3 elicitation information, such as consistency of decision table, information entropy and figure of cut points, etc. It makes the result of discretization received finally more rational. So it can be maximum assurance the consistency of the decision table is not destroyed and consistent after discretization. And because of considering...