CN-122022203-A - Economic prediction method, system, electronic equipment and medium based on Zhou Du month quarter big data

CN122022203ACN 122022203 ACN122022203 ACN 122022203ACN-122022203-A

Abstract

The invention relates to the technical field of economic metering modeling, in particular to an economic prediction method, an economic prediction system, electronic equipment and a medium based on big data of the quarters of a circumference month; the method comprises the steps of obtaining multi-frequency multi-source economic time series data, executing frequency mixing alignment pretreatment on the multi-source economic time series data according to preset frequency mixing data processing rules to generate a frequency mixing data set, constructing a connection model through a predefined modularized framework and a connection algorithm based on the frequency mixing data set, training the connection model through a parameter estimation algorithm, inputting the multi-source economic time series data obtained in real time into the trained connection model, executing high-frequency prediction, and outputting real-time prediction results. By the method, the technical problem that the prediction accuracy is insufficient when the existing frequency mixing dynamic factor model is used for processing multi-frequency mixing data is solved, and the capability of carrying out high-frequency, accurate and real-time monitoring on macroscopic economic trend is improved.

Inventors

CHEN YANZUO
WEN FAN
DONG DANHUANG
XU CHENBO
WU SHUHONG
FENG YE
ZHANG YIHONG

Assignees

国网浙江省电力有限公司经济技术研究院

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. An economic prediction method based on big data of the quarter of the circumference and the month is characterized by comprising the following steps: Acquiring multi-frequency multi-source economic time series data; performing a mixing alignment pre-process on the multi-source economic time series data according to a preset mixing data processing rule to generate a mixing data set, wherein the mixing alignment pre-process comprises aligning data with different frequencies on a time scale; Constructing a connection model through a predefined modular architecture and a connection algorithm based on the mixing dataset, wherein the modular architecture is configured to carry out information transfer interconnection on a Zhou Du month prediction module and a month quarter prediction module by adopting the connection algorithm; training the connection model through a parameter estimation algorithm according to the observation data in the historical time window so as to iteratively update model parameters and state estimation of the connection model; inputting the multisource economic time series data acquired in real time into a trained connection model, performing high-frequency prediction, and outputting a real-time prediction result of a preset economic index.
2. The method of claim 1, wherein the multi-source economic time series data comprises Zhou Du times, month times, and quarter times data, wherein performing the mix-alignment pre-process on the multi-source economic time series data according to a pre-set mix data processing rule to generate a mix data set comprises: performing frequency conversion processing on the Zhou Du frequency data to generate corresponding month frequency conversion data, wherein the frequency conversion processing comprises calculation of month average value or month accumulated value; performing a stabilization process on the month frequency conversion data including seasonal fluctuations or periodic fluctuations and the original month frequency data to generate stabilized month data, the stabilization process including calculating a homonymous growth rate; Performing a time alignment and missing value tagging process on the smoothed month data and the quarter frequency data, generating the mixed data set, the time alignment and missing value tagging process including unifying all data to the same time stamp sequence and identifying unpublished observations.
3. The method of claim 1, wherein the constructing a join model with a predefined modular architecture and join algorithm based on the mixed dataset comprises: constructing the Zhou Du month prediction module, wherein the Zhou Du month prediction module is configured to adopt a mixed data sampling model and a machine learning variable selection algorithm to generate a high-frequency prediction result of a month economic variable based on the circumference data and month data in the mixed data set; Constructing the quarter-of-month prediction module, wherein the quarter-of-month prediction module is configured to adopt a mixing dynamic factor model to generate a prediction result of a quarter economic index based on month data and quarter data in the mixing dataset; And establishing an information transmission path between the Zhou Du month prediction module and the month quarter prediction module, wherein the information transmission path is configured according to the connection algorithm and is used for transmitting the month economic variable high-frequency prediction result generated by the Zhou Du month prediction module to the month quarter prediction module as partial input information.
4. The method of claim 1, wherein training the joined model by a parameter estimation algorithm to iteratively update model parameters and state estimates of the joined model based on observed data within a historical time window comprises: Performing state estimation on the state space representation of the connection model by adopting a Kalman filtering algorithm, wherein the state estimation comprises a state prediction step and a state updating step; Iteratively solving model parameters of the joint model by combining a desired maximization algorithm and the Kalman filtering algorithm, wherein the iterative solving process comprises a desired step and a maximization step; In the expectation step, calculating a conditional expectation and covariance matrix of a state vector in the state space representation by the Kalman filtering algorithm based on a current parameter estimation value; in the maximizing step, updating a model parameter estimation value of the connection model based on a conditional expectation and covariance matrix of the state vector; And repeating the expected step and the maximizing step until the model parameter estimated value meets a predefined convergence condition, so as to obtain the trained connection model.
5. The method of claim 1, wherein inputting the multi-source economic time series data acquired in real time into a trained linkage model, performing high-frequency prediction, and outputting a real-time prediction result of a predetermined economic index, comprises: Performing the mixing alignment preprocessing on the multi-source economic time series data acquired in real time to generate a real-time mixing data set; Inputting the real-time mixing data set into the trained connection model, and updating the state estimation of the connection model through an incremental updating mechanism, wherein the incremental updating mechanism comprises state prediction and state updating of newly added data based on a Kalman filtering algorithm; Calculating a real-time predicted value of the predetermined economic index through a measurement equation of the connection model based on the updated state estimation; and outputting a real-time predicted value of the preset economic index, wherein the preset economic index comprises a total national production value in a quarter and a key monthly economic variable.
6. A method according to claim 3, wherein prior to building a joining model with a joining algorithm through a predefined modular architecture, the method further comprises: Defining a component of the modular architecture and an interconnection relationship, wherein the component comprises the Zhou Du month prediction module, the month quarter prediction module and the connection algorithm, and the interconnection relationship prescribes that the output of the Zhou Du month prediction module is used as part of input of the month quarter prediction module; configuring a first function specification for the Zhou Du month prediction module, wherein the first function specification comprises a combination mode of the mixed data sampling model and the machine learning variable selection algorithm, and the function requirement of high-frequency prediction on month economic variables is clear; Configuring a second function specification for the quarter prediction module, wherein the second function specification comprises a function requirement for adopting the mixing dynamic factor model and definitely predicting a quarter economic index; and setting information transmission rules for the connection algorithm, wherein the information transmission rules comprise the format, the content and the triggering conditions of the Zhou Du month prediction module for transmitting information to the month quarter prediction module.
7. The method of claim 4, wherein the convergence condition includes a variation of a model parameter estimate being less than a first predetermined threshold, or an increment of a logarithmic likelihood function value corresponding to the desired maximizing algorithm being less than a second predetermined threshold, the repeating the desired step and the maximizing step until the model parameter estimate meets a predefined convergence condition, the obtaining the trained joined model comprising: After the maximization step of each round is completed, calculating the variation between the model parameter estimation value of the current iteration round and the model parameter estimation value of the previous iteration round, and calculating the log likelihood function value corresponding to the current iteration round; Judging whether the variation of the model parameter estimation value is smaller than the first preset threshold value or whether the increment of the log likelihood function value is smaller than the second preset threshold value; If the variation of the model parameter estimation value is smaller than the first preset threshold value or the increment of the log likelihood function value is smaller than the second preset threshold value, judging that the convergence condition is met, terminating the iteration process, and taking the model parameter estimation value of the current iteration wheel as the final parameter of the connection model to complete the construction of the trained connection model.
8. An economic prediction system based on big data of the quarter of the month of the week, comprising: the multi-frequency data acquisition module is used for acquiring multi-frequency multi-source economic time series data; the mixed data alignment processing module is used for executing mixed alignment preprocessing on the multi-source economic time series data according to a preset mixed data processing rule so as to generate a mixed data set, wherein the mixed alignment preprocessing comprises the steps of aligning data with different frequencies on a time scale; The connection model construction module is used for constructing a connection model through a predefined modularized framework and a connection algorithm based on the mixing data set, and the modularized framework is configured to conduct information transfer interconnection on the Zhou Du month prediction module and the month quarter prediction module through the connection algorithm; The connection model training module is used for training the connection model through a parameter estimation algorithm according to the observation data in the historical time window so as to iteratively update model parameters and state estimation of the connection model; And the multi-frequency economic index prediction module is used for inputting the multi-source economic time series data acquired in real time into the trained connection model, executing high-frequency prediction and outputting a real-time prediction result of the preset economic index.
9. An electronic device, comprising: and a memory communicatively coupled to the at least one processor; Wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.
10. A non-transitory computer readable storage medium storing computer instructions, wherein the computer instructions are for causing a computer to perform the method of any one of claims 1-7.

Description

Economic prediction method, system, electronic equipment and medium based on Zhou Du month quarter big data Technical Field The invention relates to the technical field of economic metering modeling, in particular to an economic prediction method, an economic prediction system, electronic equipment and a medium based on big data of the circumference and the quarter. Background Along with the increasing demand of macroscopic economic regulation on real-time performance and accuracy, the high-frequency prediction and monitoring of domestic total production value (Gross Domestic Product, abbreviated as GDP) and important monthly economic variables has become the focus of attention of government decision-making departments, financial institutions and the public society, and has a key meaning for enhancing the scientificity and timeliness of macroscopic regulation. GDP is used as a quarternary index, the data release has inherent lag, and the traditional prediction method mainly relies on monthly and quarternary Frequency mixing data to construct a Frequency mixing dynamic factor model (Mixed Frequency-Dynamic Factor Model for short)) To enable real-time prediction of the periodicity or higher frequency of the GDP. In the field of macro-economic prediction, a mixed-frequency dynamic factor model has become one of the mainstream technologies. For example, reference 1 (application publication number CN 117455025A) discloses a macroscopic economic real-time prediction method for solar power data, and the method improves timeliness of GDP prediction by introducing solar power data and performing parameter estimation and real-time update based on a mixed dynamic factor model. The comparison document 1 drives data update through an economic calendar, and carries out frequency mixing processing on the date index, so that the problem of data lag is relieved to a certain extent. However, the technology still has obvious limitations that on one hand, focusing on the integration of solar power data, the model cannot utilize wider high-frequency information sources due to insufficient covering Zhou Du-frequency high-frequency data (such as traffic logistics, climate information and the like), and on the other hand, the existing technology has the defects that the technology is convenient to use and has high-frequency information sourcesThe model is limited to a monthly and quarterly mixing framework, and is lack of systematic integration of Zhou Du, monthly and quarterly frequency data, so that effective high-frequency prediction of important monthly variables (such as industrial increment values, fixed asset investment, social consumer product retail and import and export and the like) is difficult to realize. Currently, macroscopic economic prediction platforms based on larger scale weekly, monthly, quarterly mixed data are not yet mature. The existing method has the defects in the aspects of data frequency diversity, model adaptability and prediction coverage, and especially cannot give consideration to multivariate collaborative prediction and global economic state perception under the background of high-frequency data permeability improvement, so that the comprehensiveness and reliability of a prediction result are restricted. Therefore, the conventional mixing dynamic factor model has a technical problem of insufficient prediction accuracy when dealing with multi-frequency mixing data. Disclosure of Invention Aiming at the defects or shortcomings, the invention provides an economic prediction method, a system, electronic equipment and a medium based on the large data of the circumference and the quarter, which can solve the technical problem that the existing frequency mixing dynamic factor model has insufficient prediction precision when the existing frequency mixing dynamic factor model is used for processing the multi-frequency mixing data. The invention provides an economic prediction method based on big data of a circumference and a month and a quarter, which comprises the following steps: Multiple frequency multi-source economic time series data are acquired. A mixing alignment pre-process is performed on the multi-source economic time series data according to a preset mixing data processing rule to generate a mixing data set, wherein the mixing alignment pre-process comprises aligning data with different frequencies on a time scale. Based on the mixed dataset, a joining model is constructed by a predefined modular architecture and a joining algorithm, the modular architecture being configured to interconnect the Zhou Du month prediction module and the month quarter prediction module by information transfer using the joining algorithm. Training the connection model through a parameter estimation algorithm according to the observed data in the historical time window so as to iteratively update model parameters and state estimation of the connection model. Inputting the multisource economic time series data acquir