Optimization Model, Based On Support Vector Regression Meteorological Stations

Weather forecast not only facilitates daily travel, but also through long-term accumulation and statistics, it is processed to gain acceptance for research basis of agricultural, industrial, transportation, military, etc. Meteorological observation stations are the main source of modern weather forecasting data, but in reality the establishment of meteorological observation stations and maintenance costs are very high. In order to cut the costs, within a essential range reducing number of meteorological observation stations and simultaneously ensuring sufficient amount of information are essential. The quantity of annual precipitation information loses small, which requires us to reduce the number of stations while ensuring precipitation information is not lost.For the problems of meteorological observation stations which belongs to the field of meteorological observations, this thesis based on precipitation data to set up the regression analysis of relationship between meteorological observation stations and the goodness of fit, and also to establish the regression equation. The precipitation of the deletion observation stations is forecasted. Then combining BP neural network, the Mean Impact Value (MIV) method is used to study how to utilize neural network to select variables. With using BP neural network variables selection method, the observation stations which effect the deleted observation stations are filtered. For the deleted observation stations, the selection results are shown that the main factors of the regression analysis model and the BP neural network model are basically the same, but in the case of reducing the observation stations, compared with the regression model, BP neural network model can reflect more details of the importance of the observation stations. The results provide a guarantee to the optimality of reducing the observation stations.Finally, regression forecasting model based on BP neural network variables selection-support vector machine is set up. This model is based on support vector machine theory, which is a regression algorithm-support vector regression. This algorithm firstly choose the parameters, then depending on the decision-making model the equation is solved, which extend the theory of general support vector machines and applications. The samples were predicted and compared with the prediction results received by principal component regression model and BP neural network variables selection model. It is confirmed that the prediction effect of support vector regression prediction model performs better. The results reflect that Support Vector Regression techniques have the advantages of excellent generalization ability, global optimization, shortly training time and highly forecast accuracy.