The Research On Fusion Strategies Based On Grey Model And Support Vector Machine

In order to develop economic and social, prediction is necessary for us to make decision for business and research, which is based on the study of research data and statistic data to discover the regulation of things'development and changes in future, so it has been a new hotspot in the field of machine learning. With the more complex object and higher precision, there are some blind spots in the single model and it could not meet people's requirements. Combining several predicting methods as a whole would let the model uses the advantages and avoids the disadvantages of each method, this will increasing the model's predicting accuracy comparing to a single predicting method.This paper introduced the theory and modeling process of grey model (GM) and support vector machine (SVM). Through the comparison and analysis on GM and SVM, we confirmed the possibility of integrating the above predicting models as a whole, and proved that new predicting models have higher precision.The first one is combination optimization model. The core concept of it was to minimize the mean square error, and then used the optimal combination weights to combine GM (1,1) model and SVM model. We got the combination prediction as the final prediction, so as to improve the prediction precision.The second model is a grey support vector machine based on grey relational analysis (GRA). The first method was to use GRA to extract the main factors of model, and then took these main factors as the input factors to SVM model. The second method was based on the first one, it was to accumulate the main factors, and then inputted the accumulating generation operator (AGO) to SVM model.Applying the combination model into two practical problems, the experiments proved that the model was effective and accurate, and had good robustness. The grey support vector machine is applied to the urban environment pollution, the experiment results indicated the increasing of predicting accuracy and generalization ability of the new model, superior to the support vector machine.