Research On Time Series Prediction Based On SVM

As a new kind of machine learning method, support vector machines algorithm is based on rigorous statistical theory, and it does well in both pattern recognition and regression function. From the early 1990s of the last century to now, the algorithm has developed greatly during just more than 10 years, and continues to receive widespread attention of scholars both at home and abroad.Support vector regression is based on the principle of structural risk minimization, both the fitting and the complexity of training samples are considered, and it has better fitting effect. However, the parameters select in the model is very important, and it will directly affect the generalization effect. At present, there is no generic parameters selection method. In view of this problem, the existing method of support vector machine parameters selection was studied, and the parameters selection method of support vector regression algorithm based on the ant colony algorithm was presented. As a global probabilistic selection algorithm, ant colony algorithm creates a limited size of worker groups, and it searches the optimal solution by mutual cooperation between the ants. Then the improved support vector regression algorithm was adapted for time series prediction model. Taking the grain production of the country for example, the prediction model was established, trained and forecasted. Finally, the predicted results was analyzed and compared, and it certificated that the parameters selection method based on the ant colony algorithm is effective. The searching capabilities of overall situation robust of ant colony algorithm make some contribution to improve prediction accuracy rate of time series prediction model based on support vector machines.