Prediction Of Precipitation Modeling Based On Multicellular Gene Expression Programming And Time-frequency Analysis Methods

Precipitation is an important climatic phenomenon that is formulated under the influence of many factors and it is an important part of atmospheric cycle.With the intensifying of global warming and the deepening of various unreasonable human activities,precipitation mutations occurs from time to time,and its regularity is becoming increasingly more elusive.Accurate and timely prediction of precipitation provides reference for the development and management of regional water resources and the prevention of flood disasters as well as human's daily activities and production plans.However,the precipitation data are nonlinear and instability under the influence of different factors,and it is more or less accompanied by certain noise in the process of collecting precipitation data,which makes it difficult to predict the rainfall.The effect of traditional prediction method is difficult to meet the public expectations.As humans attach more importance to meteorological research,as well as the development of the theory and technology in data mining and intelligent computing,more and more technologies of intelligent computing and data mining are applied to precipitation prediction and better performance is achieved than traditional algorithms.At present,the Neural Network method and Support Vector Machine(SVM)algorithm widely used in precipitation prediction can effectively describe the complex relationship between the various factors of rainfall data.However,it is difficult to unify the structure and parameter selection of the neural network method and SVM algorithm,which requires programmer's experience to choose and determine,and they need very huge computational amount,which is not conducive to the training and learning of large-volume samples;the Gene Expression Programming(GEP)algorithm with powerful regression analysis ability directly predicts the rainfall and directly affects the rainfall sequence itself,the time-frequency components of the complex sequence and the time-frequency differences of different rainfalls has not received much attention,which seriously reduces the effect of rainfall prediction.In the field of practical engineering,time-frequency analysis methods are extensively used for the processing of nonlinear and instability data.Empirical mode decomposition(EMD)and wavelet analysis are currently time-frequency analysis methods for effectively processing nonlinear and non-stationary data and eliminating noise.This paper combines Multicellular gene expression programming algorithm(MC_GEP)with strong symbolic regression analysiscapabilities,with EMD and wavelet analysis to predict precipitation data.A precipitation prediction algorithm EMGEP2 RP based on EMD and MC_GEP is proposed,and the other precipitation prediction algorithm WTGEPRP based on wavelet-based analysis and MC_GEP is proposed.The proposed algorithms are able to improve the effectiveness and accuracy of precipitation prediction model,the time-frequency analysis of precipitation data was followed by GEP modeling and prediction.The main research work of this thesis is summarized as follows:(1)First,the precipitation prediction algorithm EMGEP2 RP is proposed based on MC_GEP and EMD.Using EMD method decomposed the original data signal with the features of complexity and hard-to-extract rule is adaptively into the form of the sum of multiple smooth and easy-to-observe components,on the basis of the extremum characteristic scale of the data signal as a measure.Followed by being respectively tested with the real precipitation data from Zhengzhou,Nanning and Melbourne,Australia,through the MC_GEP sliding window modeling fitting and prediction.RMSE and MAE are used as evaluation indicators.The results of simulation experiment show that the fitting and prediction results of EMGEP2 RP algorithm are better than SVM,MC_GEP,BP and GEP.The prediction accuracy of the algorithm is improved.(2)Second,the precipitation prediction algorithm WTGEPRP is proposed based on MC_GEP and wavelet analysis.Wavelet analysis is an effective method for time-scale analysis.Its window size will not change,while the time and frequency windows can be changed.It can decompose the original rainfall data into smaller frequency sums,which is convenient for observing the signal in a more subtle way.Similarly,the precipitation data of Zhengzhou,Nanning,and Melbourne,Australia are used in the simulation experiment.The experiments was carried out on all the decomposition layers of the four different wavelet bases.The results show that when the proper wavelet base and decomposition layers are selected for WTECPRP,the evaluation indicators of fitting and prediction,i.e.RMSE and MAE,by WTECPRP algorithm are significantly smaller than those by EMGEP2 RP algorithm,showing better fitting and prediction results,and higher prediction accuracy than EMGEP2 RP algorithm...