Research Of Dynamic Forecasting Method And Implementation Technique Based On Support Vector Machines

Dynamic forecasting is a common problem in project fields and scientific research. In application, many systems can be considered as complex time-varying problems. As some problems lack of prior theory and prior knowledge, and inner transformation and the interaction between environmental factors are complex, they are difficult to be described and analyzed accurately by deterministic models, for instance, the performance degradation prediction of aircraft engine, the PID control of chemical production, and the productivity prediction of oilfield development, etc. The support vector machines can import in doing not knowing or the output variable is completed on the premise of room relation nonlinearity build a model, make use of the support vector machines to build the systematic entering/ output model, be to approach an ability owing to the regression.Recently statistical learning theory has received considerable attention proposed based on small sample data, which is an important supplement and development of traditional statistics. Support Vector Machines (SVM) algorithms based on the foundations of statistical learning theory show excellent learning performance. It has virtue of self-contained theory, global optimization, good generalization ability, independence of algorithm complexity on feature space dimension, etc and has been successfully extended from basic classification tasks to regression, density estimation, novelty detection, etc. Unlike traditional methods, which minimize the empirical training error, SVM make use of the structure risk minimization principle,which may bring on a good generalization performance. Additional advantages of SVM can be appreciated in comparison to neural networks. For SVM there are only a small number of tunable parameters and training amounts to solving a convex quadratic programming problem hence giving solutions that are global, and usually unique.The paper comprehensive and systematically elaborate the present situation of application research and introduced theory background, basic thought and learning algorithm of SVM synoptically, etc. Support Vector Machines for classification problems are called Support Vector Classification (SVC), while for regression problems are Support Vector Regression (SVR). The connection of these two models is deduced by the paper, and then it introduces SVR into dynamic forecasting and solves the problem of nonlinear combination forecasting using SVR. The experiment of oilfield production prediction proves that combination forecasting based on SVR has better dynamic forecasting ability than forecasting based on SVR directly and traditional forecasting methods. In the end, PSVR model, with time-varying functions as its inputs, based on orthogonal basis expansion was proposed by the paper and it converts the signal forecast problem of time-varying function space into regression problem in high dimensional real-vector space via isomorphic principle of function space and high dimensional vector space and solves the regression analysis problem of time-varying function effectively by dint of traditional SVM method. PSVR broadens traditional SVM's restrict to inputs and widens application field of SVM.