Designs and Applications of Fuzzy Synthetic Evaluation Models Based on Parallel Algorithms
|Keywords||Decision Support System Fuzzy Synthetic Evaluation BP Neural Networks Genetic Algorithms Parallel Algorithms|
The Decision Support System (DSS) is a computer application system that helps decision makers to make scientific decisions in the form of man-machine interaction, and it uses computers as the tool and applys leadership science as well as decision -making related theories and methods to provide various information for decision makers. DSS consisits of data bases, model bases, knowledge bases, method bases and corresponding management systems. In the process of decisiong making, model bases play a very important role. And a key step for successfully designing DSS is choosing an efficient and resonable model base. However, models in DSS generally are abundant and changing frequently. Aiming at this problem, this paper mainly studies the design and the application of the optimal models in synthetic evaluation DSS.The existing synthetic evaluations are Analytical Hierarchy Process(AHP), Principal Component Analysis (PCA), Data Envelopment Analysis(DEA), Fuzzy Synthetic Evaluation(FSE), Artificial Neural Network (ANN) and evaluation model s that utilize two or more methods above. At present, FSE and ANN are two widely used synthetic evaluation methods. On the basis of further studying the above evaluation models, a fuzzy synthetic evaluation model based on parallel genetic neural networks is proposed by introducing parallel thought. Specifically, basic theories of synthetic evaluations and DSS are first introduced; then FSE, NN, GA and Parallel Algorithm (PA) are studied; in what followed is a fuzzy synthetic model based on parallel genetic neural networks that is established according to demands and features of enterprise synthetic evaluations. This model is based on the fuzzy analytic synthetic evaluation model and uses BP network to determine the fuzzy weight vector and the global search capability of GA to find the network’s initial optimal weights and thresholds to speed up the training process and to overcome some shortcomings of BP algorithm, such as sensitivity to initial values and local optimum. But, GA itself is a time consuming algorithm, so we introduce PA to speed up its convergence by taking into consideration of GA’s internal parallelism. At last, the performance and evaluation results of this model are made a comparison to counterparts of fuzzy analytic synthetic evaluation model, BP neural network based fuzzy synthetic evaluation model and genetic neural network based fuzzy synthetic evaluation model.The example shows that taking into consideration both time efficiency and evaluation results, parallel genetic neural network based fuzzy synthetic evaluation model is more reasonable.