Decomposition Combination Prediction Method Of Interval Time Series And Its Application To Daily Extreme Temperature

After the issue of global warming was raised,relevant national departments have been participating in climate control actions.However,high temperature heat waves and droughts caused by global warming continue to occur,affecting people's daily travel and daily life.In addition,these extreme weathers have caused considerable impacts in various industries.For agricultural production,the loss of land resources and other natural resources caused by extreme weather will also worsen the current climate warming situation,and thus As a result,extreme weather has become more frequent and the losses caused by it have further expanded,thus forming a vicious circle.Climate warming has always been a global problem,and it is also the most urgent problem to be solved.In recent years,relevant national departments have been attaching great importance to it.In 2019,the No.1 document of the Central Committee put forward the requirement of "building a modern meteorological service system for agriculture" in the discussion on agricultural foundation and the supply of agricultural products.The China Meteorological Administration clarified the current abnormal weather and climate situation and the severity of the situation at the meeting on monitoring,forecasting and early warning services for disastrous weather held on May 24,2021.As an important indicator in the meteorological system,temperature has a certain degree of influence on the supply of agriculture and agricultural products.Not only that,extreme temperature is used as a temperature indicator,when the extreme temperature in a certain area exceeds or falls below a certain threshold.They are all early warning phenomena of some natural disasters,which shows that predicting extreme temperatures can help people to prevent and reduce disasters in areas such as daily travel and agriculture.Secondly,extreme temperature data often has a certain periodicity,and also exhibits the characteristics of nonlinearity and volatility.Therefore,using the form of sequence decomposition and reorganization to obtain useful information within the data,and then using the combined prediction method to predict the sequence after decomposition can improve the prediction accuracy,so as to better use the extreme temperature indicators for early warning and other functions.In this paper,the daily extreme temperature interval time series is taken as the analysis object.Firstly,the interval time series is divided into central series and radius series.The central series and radius series of the daily extreme temperature interval are divided by the complementary integrated empirical mode decomposition algorithm(ceemd).For several IMF components formed after decomposition,the IMF is reorganized by using the difference of component mean value,and the mean value of each IMF component is different,Sum up the IMF that can reflect the similar internal information of the data to obtain the high-frequency data and low-frequency data of the central sequence,the high-frequency data and low-frequency data of the radius sequence respectively(the res components of the two sequences are used as the trend item data respectively);Then the exponential smoothing method,long-term and short-term neural network method(LSTM)and support vector regression(SVR)are used to predict the high-frequency,low-frequency and residual sequences of center sequence and radius sequence respectively;Then,according to the center series and radius series divided by interval time series,the preference coefficient is introduced,and the cosine of vector angle is used as a correlation index to construct the fixed weight coefficient combination prediction model;As an error index of this paper,the sum of squares of logarithmic error is used to construct the fixed weight coefficient combination prediction model,and the fmincon function in the optimal toolkit of MATLAB is used to calculate the optimal weight coefficient of the fixed weight coefficient combination prediction model;Then,according to the characteristics of different prediction accuracy of single methods at different time points,IOWGA operator is introduced to regroup the single prediction values at each time point.The vector angle cosine variable weight coefficient combination prediction model is still constructed from the perspective of correlation index,and the variable weight coefficient combination prediction model of logarithmic error square sum is constructed from the perspective of error index,Similarly,the fmincon function in the optimization toolkit of MATLAB is used to calculate the optimal weight coefficient of the two variable weight coefficient combination prediction models;Finally,different models are used in the empirical analysis of Hangzhou from January 1,2011 to June 30,2021.The predicted values of the central series and radius series are restored to obtain the predicted values of the daily extreme temperature interval time series,and the predicted results are summarized.The different models are compared by using the model evaluation indexes of the inter regional time series(SSE,MSE,MAE,MAPE,MSPE).After empirical application,it is found that the daily extreme temperature interval series has obvious periodicity,and some evaluation indexes predicted by the combined prediction model based on correlation index or error index are almost better than the three single prediction methods selected in this paper;In the application of combined forecasting model,the evaluation indexes of the combined forecasting model with variable weight coefficient constructed in this paper are less than those of the combined forecasting model with fixed weight coefficient,and the combined forecasting model with variable weight coefficient can effectively reduce the prediction error.