Research On Complex Simulation Data Analysis Method Based On Data Mining

Simulation data analysis is the process of collecting and analyzing simulation data purposefully to become information or knowledge, and is the key issue in simulation area. The aim of it is to understand and improve the system preferably, There is not an effective method of simulation data analysis for dealing with the complex simulation data whose scale is large, dimensionality is high and relation is complex in existence,therefore, Data Mining technique is applied to the simulation data analysis using a predictive modeling method based on Support Vector Machine to provide solutions for the problem of plan optimization in large range and data trend forecast.The dissertation's background and purposes are expatiated. The difficulties in current data analysis are proposed. The history and status in the home and abroad of data analysis methods are analyzed. The task, development history and research progress of Data Mining are summarized. And the contents of the paper are given.The requirement for the data analysis of simulation system experiments is proposed based on the compare between simulation experiments and traditional experiments. The characteristics (large amounts, high dimensionality, a few sample and strong uncertainty) are analyzed based on the characteristics of the simulation data. According to the characteristics, two kinds of data pretreatment methods are proposed to meet different requirements of data analysis. One is method of reduction of attributes, which can solve the problem that cannot be settled by high dimension data. And the other one is outlier detection method that can solve the problem of outlier influencing reliability of regression model.Key techniques of simulation data analysis based on Data Mining are further researched on basis of the shortcomings of traditional data analysis. Support Vector Machine is adopted as the core theory of predictive modeling analysis to simulation data. Classification model and regression model are applied respectively to real simulation system, and this method could solve the problems to a large extent that plan optimization in large range and data trend forecast are difficult to achieve.The optimization algorithm of predictive model is realized. Grid search and Genetic Algorithm are adopted respectively to optimize parameter based on Cross Validation thoughts. The shortcomings of the two algorithms are pointed out after comparing the optimization results of the two algorithms. A hybrid optimization algorithm that applies the two algorithms combined is proposed to solve the problem of low computing speed of grid search optimization in wide range and overfitting that probably exists in result of Genetic Algorithm optimization.