The Short-term Meteorological Element Forecasting Model And Its Application Based On EMD Of Phase Space Reconstruction Of Extreme Learning Machine

Since ancient times, timeliness and accuracy of meteorological science affect all aspects of human life and production, determine the social development and stability. Along with the Development of the times, the human needs of meteorological science rising continuously. But there are more or less different degrees of problems exist in some methods, such as the model accuracy is not high enough, the time for establish models is too long, affected by noise and so on, how to establish a more accurate and effective prediction models is an important research topic in today's society. For this topic, we propose the prediction model based on empirical mode decomposition and phase space reconstruction of extreme learning machine for models establishing with timing sequence data. To de-noising of the data with the technique of empirical mode decomposition, to extraction mode of the data with the technique of phase space reconstruction so that the reconstructed data is more suitable for establish models, and finally to establish models of the reconstructed data with the technique of extreme learning machine. We can test the performance of the model we establish by using these three techniques mentioned above and analyze its advantages.The main contents of this article has the following three points:1) For the problem of noise and the accuracy is not high enough exist in the models for single meteorological elements timing sequence data, we research the method to prediction model based on empirical mode decomposition of phase space reconstruction of extreme learning machine for models establishing. To de-noising of the data with the technique of empirical mode decomposition, reduce the impacts of other uncertainties observational data to provide a more efficient and accurate data for subsequent processing. There is a certain degree of difficulty for establish models with timing sequence data directly and also it is hard to get a success model. In order to restore its dynamic system effectively, we use the technique of phase space reconstruction to research and reproducibility its topology of original dynamic system of the timing sequence data of one-dimensional meteorological single elements, so that the reconstructed data is more suitable for rules extraction and models construction. It can be quickly and effectively guarantee the generalization ability to establish models of the reconstructed data with the technique of extreme learning machine.2) We analysis the shortage of the threshold setting in the algorithm to screening the intrinsic mode function components and propose an improved algorithm to instead by research on the theory of the empirical mode decomposition. Compared with the traditional way, we present a new definition of formula to set dynamic filter threshold for different intrinsic mode function components depend on the related information of the current selected intrinsic mode function components, so that we can select intrinsic mode function components effectively for data reorganization. Dynamic filter threshold can be more effective to determine the critical intrinsic mode function components' rigorous ascription, so that can improve the quality of data after de-noising and overcome the subjectivity.3) We test and analysis the parameters of the prediction model depend on the real temperature elements of one-dimensional time series data, also we compare and analysis the performance between the usual model and the model based on dynamic filter threshold generated of empirical mode decomposition(called improved model).At the same time, we compare and analysis the performance of these two models to several other models of different methods, and find that the performance and the prediction accuracy of the improved prediction model we proposed are better and it can achieve human basic needs for weather forecasting in daily life.Finally, we give some points that can still improve the performance of the models mentioned above, determine the direction for the future research.