The Forecasting Model Of Sand-Dust Storm Based On FNN

At present, the technology of artificial intelligence and pattern recognizing has been widely applied in various fields. The implementation of extracting feature and forecasting sand-dust storm, which use the technology of artificial intelligence and pattern recognize is presented in this paper. At first,starting from the character of sand-dust storm and non-sand-dust storm samples reflected on four physical fields, we choose representative samples used in modeling. By studying the result of principal component analysis (PCA), a scheme to integrate features is designed. "gross features" which give attention to all the principal components are formed hereby. Thus, 40 dimensions features of samples are extracted again by the method of integrating features, 10 dimensions features of modeling samples are formed at last, which are more reasonable than ever. Second, the sand-dust storm forecasting model based on fuzzy nerve net (FNN) is researched in succession. By means of model optimized in several aspects such as FNN topology, parameters and compilation of the samples, a more reasonable forecasting result is acquired by the FNN model.It is pointed that the primary factor in influencing the right forecast ratio is lots of non-typical samples among training samples by researching the forecasting result based on FNN again. Then, by clustering, the region of non-typical samples is founded, so does the statistical modeling based on non-typical samples. A method to adjust the degree subjecting to sand-dust storm has been designed at the same time, which gives attention to the forecasting result of FNN and the degree of non-type. Then samples are forecasted again using statistical model (second stage) based on the FNN model (first stage), some samples which are forecasted wrong at first stage are corrected at second stage, and the statistical model has little effect on samples which are forecasted right at first stage. Compared with the forecasting result of NN based on 40 dimensions from literature [5], the right ratio of sand-dust storm is improved from 60% to 73.3%, CSI is improved from 25.9% to 38.7% by the model combined FNN with statistical model pointed in this paper, the result is better than ever. At last, some problems encountered in constructing forecast system are summarized in the paper. Some resolutions to these problems such as reconstructing features, integration NN, combining expert system and NN are presented at the same time, as well as the realizing frame of dynamical forecasting system.