Temperature Time Series Prediction In Two Cities Of Gansu Province

Temperature time-series data is the first-hand information for predicting the future trend of temperature,and provides an important basis for preventing natural disasters caused by temperature.Changes in temperature will have a significant impact on crops,water resources,tourism and so on.With the continuous development of computer computing power,some intelligent learning algorithm like neural network have been successfully applied in the field of meteorological research,and good prediction results have been obtained.However,at the present stage,there are still some problems in the prediction of time series applied neural network,such as determining the number of neurons in each network layer and the connection threshold between each network layer,etc.,which directly affect the accuracy of the prediction results.To solve the above problems,this paper adopts autocorrelation coefficient,genetic algorithm?GA?and collective empirical mode decomposition?EEMD?for optimization,so as to establish an effective temperature prediction model and conduct temperature prediction.This paper mainly takes the daily average temperature,daily minimum temperature and daily maximum temperature of dunhuang city and yuzhong county in gansu province as the research object.The main work is as follows:?1?When establishing the temperature prediction model suitable for the two places,the number of nodes in each network layer of the neural network was determined.In order to avoid the information redundancy and irregularity of the input layer data,when determining the number of neurons in the input layer m,the autocorrelation coefficient of time series data is taken into account.In the case that the autocorrelation coefficient does not change much,that is R_i-1-R_i?l,write down the autocorrelation order at this time,and set m=i.Through the verification of temperature time series data of the two places,when l=0.1,the prediction effect of the model is the best.Then,according to Lippmann RP formula,the number of hidden layer neurons n=i+1;?2?For the optimization of neural network weight threshold,genetic algorithm was used to find the optimal parameters;?3?Select the optimal neural network model.The test data set is input into the trained model for network testing,and the optimal neural network model is determined according to the error matrix obtained.The BP neural network?GABP?optimized by genetic algorithm was found to be effective in temperature prediction of two cities.?4?The eemd-gabp model was further constructed by using the GABP of the optimal model?3?.First,EEMD is used to decompose the temperature time series data to obtain the eigenmode function imf_i?i=1,...k?and then used GABP to respectively predict the imf_i,and finally added up the predicted results of k imf_i to obtain the final predicted value.?5?Compared the predicted results before and after decomposition?GABP and EEMD-GA-BP?,the model was selected again,and the temperature was predicted in the later stage.By comparing the temperature predictions of the two places,the model evaluation index values are obtained,and it is found that the EEMD-GA-BP prediction model will be more accurate.The EEMD-GA-BP model was selected to predict the temperature on March 6,2019.