复杂系统判别决策的降维、建模方法与LAD理论研究

This project studied the academic mechanism and connection of LAD and CART. The equivalent condition between LAD support set and decision tree was proved, and a new algorithm of LAD decision was proposed. Moreover, LAD decision tree was generalized to multi-state discrimination analysis. The project also studied dynamic analysis method of cube time series data table. Based on defining a mathematic criterion of whole dimension reduction, a modeling method of decision tree of cube time series data table was proposed, and the dynamic rule of discriminating pattern changing with the time was studied. Furthermore, a new dimension reduction method was proposed to realize dynamic pattern discrimination of high-dimensional mixed data by combining CART and self-organizing map, which depending on the artificial neural network. The method was already applied to the analysis of flood disaster pattern and financial condition, and then the rationality and validity of the method were proved. In fact this discriminating method as an important tool can be widely applied to economic and social analysis fields.

本项目拟采用国际合作方式，全面沟通LAD与CART的内在联系，将CART以及一些成熟的统计椒ㄒ氲絃AD理论研究，全面系统地扩展LAD的理论范畴和适用功能，建立多状态系统降维虢５腖AD理论体系，研究对时序立体数据表进行动态判别决策分析的理论方法。本项目的实现，将为复杂系统的判别决策提供具有更加广泛应用价值的理论方法。