Research On Cost Estimation System Based On Neural Network And Similarity Analysis

As the core element of product competitiveness, the cost of products has a vital impact on the survival and development of magnetic materials enterprises. However, many difficulties exist in prediction of production data, such as processing yield. Considered this situation under magnetic materials industry background, we have a research on forecasting of gross specification, processing yield, similarity weights optimization problems and forward their own solutions, and then build a Web-based application for this system.This article includes:(1) To solve the problems of forecasting gross specification during estimation procedure, this article researches the relationship between product specification and slice num, then proposes a method based on BP neural network to build a model between them, then to forecast slice num. Finally we have a test based on current data and test results showed the algorithm achieved a high accuracy.(2) To solve the problem of forecasting production processing yield, we analyze the relationship between product features and the processing yield, and propose a similarity analysis method based on CBR, finally test this method on current data and test results showed the algorithm got a significant improvement on yield predication.(3) To solve the problem of similarity weights optimization, we propose a method based on genetic algorithm, and showed weight optimization results after implementing the method, finally test on current data and test results showed optimized method got an improvement on original method of yield predication.(4) Finally, we summarize all achievements of the dissertation. After the deployment of the system, the enterprise reduces the production cost and enhances competitiveness. The method proposed in this article deserved to promote to other companies in the same industry. We also analyze the defects and give the further work in the future.