A Preprocessing Method And Intelligent Model Based On Grey Prediction

The rise of artificial intelligence makes the neural network (NN), support vector machine (SVM), fuzzy system and other methods who have been widely used for prediction. However, artificial intelligence based approach who requires a big training data and relative long training time makes that it is very difficult to carry on the economic forecast because of the highly nonlinear and randomness of economic time-series resulting that the time-series is highly irregular and extremely flexible. The introduction of grey system theory and single-step prediction method (scroll mechanism) can solution this problem well because of that the grey system model only needs a small, or discrete, or incomplete training data set to construct the prediction model, and the single-step prediction method also overcomes the limitations of grey prediction which is caused by the chaotic data. This master’s degree thesis which good involves the particle swarm intelligent algorithm (PSO) into the scrolling grey prediction model is good that this new grey rolling model can optimize automatically its parameters according to the data trend change instead of the traditional method which is fixed parameters, and significantly enhances its applicability, accuracy and intelligence. Meanwhile, this thesis proposes a new data preprocessing model which is good proper the grey forecast model, stems from data preprocessing theory of data mining but is different from the existing data preprocessing method which is for coping the massive data, and especially is proper the small and exponential growth data. In its preprocessing process, it always continues that the more recent data is more effective to the predicting results.With the purpose of gradually in-depth study algorithms and the studying subject of the economic data of tertiary industry, this paper first measures the data whether is high noise, unstable, nonlinear and the model’s applicability whether is weak by a benchmark of the predicting results of the standard grey prediction model-GM (1,1) model. Then the rolling grey model-RGM (1,1) model and particle swarm grey model-PGM (1,1) model as the reference were employed by the paper. Thus, a intelligent rolling grey model-PRGM (1,1) model which searches the most single-step optimize parameter by nested iteration in the constrained space, inherits the good convergence and easy program of PSO, also has a good robustness and gets a better prediction in the experiments whether on the actual experimental data or on the data preprocessed was proposed in this paper.Some unexpected economic events (such as the financial crisis) possible result the economic data serious fluctuations, nonlinear and no regularity in a certain time period and make the traditional preprocessing method which does not applied to cope such a small sample and strict time series of economic data, then an exponential preprocessing model-EP model came into being. This EP model established by iteration select the data points which can satisfy the certain condition in all sample data, finally the preprocessing results were obtained. It is similar a feedback system when we give the criteria which is first ensuring the error of most recent time series is small. The experiments show that the EP method generally increases the performance of four grey models.The efficient and intelligent prediction method could better help the policy-maker accurately and timely grasp the socio-economic development and also provide important technical support for macroeconomic planning. A useful attempt has been done by experiment at economic data in this paper, but it is not limited to the field of economic forecast, this idea with the application of the theory is significance for the real problem.