Research On The Multidimensional Dynamic Data Mining Technology And Its Application In Incomplete Information Modeling

Multi-dimensional dynamic data processing and analysis has become one of important research topics which is applicable to data processing, dynamic data modeling and visual modeling, with the expansion of the scientific research and engineering applications. There is usually a variety of factors that lead to incomplete information in reality, such as the lack of information, the data object mechanism knowledge incompletion or part of characteristics loss in multidimensional space. Therefore, the research in multi-dimensional dynamic data processing and the utilization of incomplete information modeling has important theoretical and practical values.In the field of the oilfield developing application background which includes the research of well group relationship between injection and production, and the prediction of developing index, using the technology of data mining and nonlinear system modeling. Thinking of the properties of multi-dimensional dynamic data, this thesis has researched the data mining models and mining algorithm which have the space characteristics and the process characteristics in nonlinear dynamic system, and integrity and consistency processing in incomplete information data based on soft-sensing techniques. This thesis also has implemented effective mining in multi-dimensional dynamic data, incomplete information modeling and simulation of evolution law in nonlinear dynamic system, put forward a more complete theoretical framework of the multi-dimensional dynamic data mining system, and built the describing model of multi-dimensional dynamic information. The data mining models in this thesis are intelligent dynamic data mining models including time sequence mining model based on SVR, serial input and output PNN mining model, multi-aggregation PNN mining model and radial basis PNN mining model. The completing data algorithms in this thesis are based on multiple statistical methods, Kriging method and data filtering technology. By using the phase space reconstruction, the training sample of intelligent mining model set is constructed. The nerve network training algorithm is come from an IQPSO and the gradient descent algorithm. It effectively implements the optimization of the nerve network initial weight, threshold value and hidden nodes number. Finally achieve satisfactory results of applications with the incomplete information modeling achieved by the mining modeling. Through the applications of developing index prediction method in oil-field and 2D/3D modeling of the seepage law in well group, which both are typical applications with incomplete mechanism information in oil-field development, the effectiveness of the methods and techniques in this thesis are validated.