The Model Of Thunderstorms Forecast Based On RS-SVM

Thunderstorms are a major hazard in the electronic age. With the development of science and technology, especially the development of meteorological science and technology, the research on thunderstorm forecast is developing, and has also gained a lot of achievement. But this kind of the thunderstorms forecast dealing with small-scale (forecast scale:5km x5km) and short-term (the next3hours) has yet to carry out. Thunderstorm not only belongs to small-scale meteorological phenomena, but also can cause a very serious disaster. Therefore, the disaster prevention and mitigation department is actively carrying out research in this area, small-scale thunderstorm forecasting model is one of the important subjects of meteorological science and technology research.The Rough sets is a very effective tool to deal with uncertain,imperfect, imprecise information, and the support vector machines has better classification accuracy and generalization ability by the balance of the VC dimension and empirical risk. With lightning forecast characteristics, a small scale (forecast scale:5km×5km) short-term thunderstorms forecast (the next3hours) based on rough set theory and support vector machines was proposed. In this paper, the research work mainly includs the following aspects:(1) Data acquisition. The historical LAPS meteorological data and lightning location data were used as our experimental data. Because the lightning samples and non-lightning samples are of the non-equilibrium data, a process method changing the non-equilibrium data into the equilibrium data was presented based on the Huffman tree algorithm.(2) Data discretization. The discretization algorithm discreting the continuous attribute data based on information entropy was introduced. However, because of having too much breakpoints generated by the aogorithm, so we proposed the concept of inconsistency rate and improved the discretization algorithm based on information entropy.The improved algorithm not only improves the efficiency of the algorithm, it also significantly reduces the number of attribute value breakpoints. (3) The generation method of Predictors based on attributes reduction. Firstly, both the method of computing positive domain based on the radix sort and the heuristic function of attribute importance based on the distinguishable degree were described. Secondly, the blind method for the attribute reduction was improved to promote the efficiency of the algorithm. Finally, using the algorithm effective predictors were obtained from the meteorological discretization data.(4) Proposal of thunderstorms forecast models based on rough set and support vector machine, including the following three models:1) the thunderstorms forecast model based on rough sets;2) the thunderstorms forecast model based on SVM;3) the thunderstorms forecast model based on both rough sets and SVM. The proposed thunderstorms forecast models were tested on real data sets and analyzed with respect to the forecast accuracy and stability of the models. The experimental results show that the thunderstorms forecast model based on both rough sets and SVM has better predictionaccuracy and stability than others.At last, give a summary to the whole research work of this paper, including the desirable points and shortcomings, and give future directions.