CN114117355A - Optimization method, system, equipment and readable storage medium of time-varying-resistance model - Google Patents

Optimization method, system, equipment and readable storage medium of time-varying-resistance model Download PDF

Info

Publication number
CN114117355A
CN114117355A CN202210099881.1A CN202210099881A CN114117355A CN 114117355 A CN114117355 A CN 114117355A CN 202210099881 A CN202210099881 A CN 202210099881A CN 114117355 A CN114117355 A CN 114117355A
Authority
CN
China
Prior art keywords
model
frequency
factor
mixing
factors
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210099881.1A
Other languages
Chinese (zh)
Inventor
肖京
赵盟盟
王磊
李娜
王媛
谭韬
陈又新
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Priority to CN202210099881.1A priority Critical patent/CN114117355A/en
Publication of CN114117355A publication Critical patent/CN114117355A/en
Priority to PCT/CN2022/090222 priority patent/WO2023142288A1/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Development Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Educational Administration (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Biology (AREA)
  • Algebra (AREA)
  • Probability & Statistics with Applications (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Technology Law (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an optimization method, a system, equipment and a readable storage medium of an anti-time-varying model, belonging to the technical field of financial security and comprising the following steps: constructing a mixing model based on the high-frequency factor; checking whether the factor distribution in the mixing model changes; and updating the mixing model on line according to the changed factor distribution. According to the invention, by providing the high-frequency factor mixing model and the self-adaptive modeling framework, the problem of influence of time-varying property of the financial rule on the prediction model is effectively solved, meanwhile, the hysteresis of the risk prediction model is effectively avoided, and the capability of timely adapting of the model along with the change of the financial rule is improved. Based on the framework, the effective prediction of economic and financial targets such as macroscopic economic indexes, mesoscopic industrial landscape, micro enterprise revenue fluctuation and the like can be realized, and dynamic risk early warning can be effectively carried out.

Description

Optimization method, system, equipment and readable storage medium of time-varying-resistance model
Technical Field
The invention belongs to the technical field of financial security, and relates to a time-varying resistance model optimization method, a time-varying resistance model optimization system, time-varying resistance model optimization equipment and a readable storage medium.
Background
Economic and financial security is an important component of national security and is also a precondition and an important guarantee of national security. Systematic financial risks are wide and infectious, the risks cover various economic and financial fields, and are transmitted to non-financial enterprises through various financial businesses, and finally become systematic economic and financial crises. The traditional mode for pre-judging economic and financial risks is mainly completed by combining expert experience with statistical analysis, and most of the method defaults to the time invariance of statistical rules. Due to the fact that the economic financial data generally have regular time-varying property, the traditional method cannot effectively solve the problem of inaccurate prediction caused by the regular time-varying property, and therefore the financial risk studying and judging effect is seriously affected.
Disclosure of Invention
The invention aims to solve the problems in the prior art and provides a time-varying-resistance model optimization method, a time-varying-resistance model optimization system, time-varying-resistance model optimization equipment and a readable storage medium.
In order to achieve the purpose, the invention adopts the following technical scheme to realize the purpose:
a method for optimizing a time-varying resistance model comprises the following steps:
constructing a mixing model based on the high-frequency factor, wherein the constructing of the mixing model specifically comprises the following steps: modeling through factor frequency conversion and frequency mixing; the factor frequency conversion is to converge high-frequency factors into low-frequency factors according to a time period; after factor frequency conversion, converting the high-frequency factor into a low-frequency factor, and performing combined modeling with the original low-frequency factor; the mixing modeling is to construct an EMD-MIDADS model based on different spelling factors; the construction of the EMD-MIDADS model specifically comprises the following steps: performing EMD decomposition on the B-sample based on the adaptive node; fusing an EMD model and an MIDAS model;
checking whether the factor distribution in the mixing model changes;
and performing online updating on the mixing model according to the changed factor distribution, wherein the online updating uses Bayesian online learning and regular optimization gradient descent to perform online updating on the model.
The invention further improves the following steps:
the EMD decomposition of the B-sample based on the adaptive node comprises the following steps:
constructing a spline difference value self-adaptive node determination method by utilizing a GCV cross validation method based on a cubic spline regression model with a second-order penalty term;
the spline regression model with the second-order penalty term is as follows:
Figure 775238DEST_PATH_IMAGE001
wherein, S: (x i ) As splinesS(. charpy) at observation pointx i The value of the function (c) of (c),λa penalty factor being a smoothness penalty term;Sthe expression of (dash) is at the interpolation pointε=(ε 1,ε 2,…,ε k ) The cubic spline function obtained by the above-mentioned method,εthe determination standard follows the GCV criterion;nin order to observe the number of samples,yiin order to observe the dependent variable of the sample,xiis an independent variable of the observed sample; s'2(xi) Is the square of the second derivative of the spline function;
GCV is a function of a penalty factor lambda and the number of nodes and represents the relative rate of a residual error term and the complexity of a model;
Figure 311261DEST_PATH_IMAGE002
wherein,εvalue set for interpolation points; tr () represents the trace-finding operation of the matrixIn the calculation, the calculation is carried out,H(λ,ε) Representing that a hat matrix corresponding to the minimum value of the RSS is solved by using a least square estimation expression;
adaptively determining the number and the position of interpolation points by taking the highest fitting degree of the upper envelope line and the lower envelope line as a target; selecting a plurality of interpolation points for a time region with dense local extreme points;
assuming that the upper and lower envelopes of the estimated upper and lower local points are linese u (t) Ande d (t) Then obtain the envelope mean valuem(t) Thereby obtainingh(t)=x(t)-m(t) Judgment ofh(t) Whether or not it isIMFIf not, willh(t) Viewed as ax(t) Continuing the above steps; finally, the rest items are stopped after meeting the stop rule, andx(t) And (4) completing decomposition:
Figure 952458DEST_PATH_IMAGE003
after EMD decomposition, the frequency of the obtained product is differentIMF i (t) The signal, with increasing order, decreases in the vibration frequency in turn, decomposing the residual termr(t) Is a long-term trend term.
The EMD model and the MIDAS model are fused, and the method comprises the following steps:
simultaneously bringing the high-frequency factor and the low-frequency factor into a regression model, and fitting and predicting an EMD decomposed signal:
Figure 638654DEST_PATH_IMAGE004
wherein,τ t is an index of different frequencies after EMD decomposition,m l v()to comprehensively consider the intercept terms of the horizontal and ripple effect models,Sis a high frequency factor
Figure 439120DEST_PATH_IMAGE005
The number of the (c) is,s=1,2,…,S，
Figure 286991DEST_PATH_IMAGE006
is composed ofSThe coefficients corresponding to the respective high-frequency factors,Gfor the reason of the high frequency, it is,ψ k for the weight function corresponding to the high frequency,
Figure 681063DEST_PATH_IMAGE007
and
Figure 905371DEST_PATH_IMAGE008
two parameters to be estimated in the weight function,
Figure 498026DEST_PATH_IMAGE009
intThe current time stamp expressed as a microscopic factor of high frequency,kfor the purpose of its hysteresis order label,k=1,2,…,K l v()，K l v()is the maximum hysteresis order;
Figure 110273DEST_PATH_IMAGE010
is a macroscopic factor of low frequency and is,Nthe number of the macroscopic factors is the number of the macroscopic factors,n=1,2,…N,
Figure 788379DEST_PATH_IMAGE011
is the coefficient corresponding to the exogenous variable,
Figure 754061DEST_PATH_IMAGE012
and
Figure 201223DEST_PATH_IMAGE013
two parameters of the weight function.
The checking whether the factor distribution in the mixing model changes comprises:
determining whether the main part of the mixing model is changed or not through mutation point detection and distribution consistency detection; the mixing model comprises a model body part and a parameter part; the main body part comprises explanatory variables of the model and function assumptions thereof, and the parameter part comprises model hyper-parameters and parameters to be estimated; and after the model main body and the hyper-parameters are determined, estimating the parameters to be estimated according to the observation data.
The mutation point inspection and the distribution consistency inspection are periodically executed according to the observation frequency of the high-frequency data; and when the distribution consistency test is used for determining whether the distribution of the financial factors is changed or not when the mutation points are found by the test, and if the distribution of the financial factors is changed, changing the main body of the model to solve the problem of adaptation of the model.
The distribution consistency check comprises:
when the consistency test does not reject the original hypothesis, the catastrophe point is judged as a local punctate risk, and the model is estimated again after a training set is removed; when the model main body is changed, reconstructing the model main body part, including re-screening the features and changing the function hypothesis of the features; and fusing the estimation result of the new model under the small sample with the estimation result of the original model by using a transfer learning technology.
The main part of the mixing model comprises model factors and model hypotheses;
Figure 984371DEST_PATH_IMAGE014
and
Figure 149773DEST_PATH_IMAGE015
for model factors, linear combinations are model assumptions, weight parameters
Figure 919146DEST_PATH_IMAGE016
、
Figure 220814DEST_PATH_IMAGE017
、
Figure 177794DEST_PATH_IMAGE018
And
Figure 830492DEST_PATH_IMAGE019
is a model hyper-parameter;m l v()、
Figure 137976DEST_PATH_IMAGE020
and
Figure 559730DEST_PATH_IMAGE021
the parameters to be estimated for the model.
The invention also discloses an optimization system of the time-varying resistance model, which comprises the following steps:
the model building module is used for building a mixing model based on the high-frequency factor, and the building of the mixing model specifically comprises the following steps: modeling through factor frequency conversion and frequency mixing; the factor frequency conversion is to converge high-frequency factors into low-frequency factors according to a time period; after factor frequency conversion, converting the high-frequency factor into a low-frequency factor, and performing combined modeling with the original low-frequency factor; the mixing modeling is to construct an EMD-MIDADS model based on different spelling factors; the construction of the EMD-MIDADS model specifically comprises the following steps: performing EMD decomposition on the B-sample based on the adaptive node; fusing an EMD model and an MIDAS model; (ii) a
A model checking module for checking whether the factor distribution in the mixing model changes;
and the model updating module is used for updating the mixing model on line according to the changed factor distribution, and the model is updated on line by combining Bayesian online learning and regular optimization gradient descent.
A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the method as described above when executing the computer program.
A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method as described above.
Compared with the prior art, the invention has the following beneficial effects:
according to the method, firstly, strong timeliness factors are introduced, a mixing factor system is constructed to resist time-varying property, meanwhile, a financial risk distribution pattern recognition model is researched and developed, and finally, an online self-adaptive modeling framework is formed, so that the model can be adapted in time along with financial rules. The method ensures that the financial risk identification model can resist the time-varying characteristics of financial rules, reduces the probability of model failure, and fundamentally improves the accuracy of model prediction. According to the invention, by providing the high-frequency factor mixing model and the self-adaptive modeling framework, the problem of influence of time-varying property of the financial rule on the prediction model is effectively solved, meanwhile, the hysteresis of the risk prediction model is effectively avoided, and the capability of timely adapting of the model along with the change of the financial rule is improved. Based on the framework, the effective prediction of economic and financial targets such as macroscopic economic indexes, mesoscopic industrial landscape, micro enterprise revenue fluctuation and the like can be realized, and dynamic risk early warning can be effectively carried out.
Drawings
In order to more clearly explain the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention, and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart of the optimization method of the anti-time-varying model of the present invention.
FIG. 2 is a block diagram of an optimization system of the anti-time-varying model of the present invention.
FIG. 3 is an overall logic diagram of the present invention.
FIG. 4 is a schematic diagram of mixing data according to the present invention.
FIG. 5 is a diagram illustrating a financial distribution pattern recognition technique according to the present invention.
FIG. 6 is a schematic diagram of the online learning method of the present invention.
FIG. 7 is a diagram of an adaptive learning framework in accordance with the present invention.
Fig. 8 is a schematic diagram of the income distribution of the residents in the united states before and after the financial crisis in 2008.
FIG. 9 is a graph showing GDP trends and predictions in Zhongshan City.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.
In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. are used for indicating the orientation or positional relationship based on the orientation or positional relationship shown in the drawings or the orientation or positional relationship which is usually arranged when the product of the present invention is used, the description is merely for convenience and simplicity, and the indication or suggestion that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and thus, cannot be understood as limiting the present invention. Furthermore, the terms "first," "second," and the like are used merely to distinguish one description from another, and are not to be construed as indicating or implying relative importance.
Furthermore, the term "horizontal", if present, does not mean that the component is required to be absolutely horizontal, but may be slightly inclined. For example, "horizontal" merely means that the direction is more horizontal than "vertical" and does not mean that the structure must be perfectly horizontal, but may be slightly inclined.
In the description of the embodiments of the present invention, it should be further noted that unless otherwise explicitly stated or limited, the terms "disposed," "mounted," "connected," and "connected" should be interpreted broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
The invention is described in further detail below with reference to the accompanying drawings:
referring to fig. 1, the embodiment of the invention discloses an optimization method of an anti-time-varying model, comprising the following steps:
s1, constructing a mixing model based on the high-frequency factor;
modeling by factor frequency conversion and mixing. The factor frequency conversion is to converge high-frequency factors into low-frequency factors according to a time period; after factor frequency conversion, the high-frequency factor is converted into a low-frequency factor, and the low-frequency factor can be combined with the original low-frequency factor for modeling. The mixing modeling is used for constructing an EMD-MIDADS model based on different spelling factors.
Constructing an EMD-MIDADS model, which comprises the following steps:
performing EMD on a B-sample based on an adaptive node, specifically:
constructing a spline difference value self-adaptive node determination method by utilizing a GCV cross validation method based on a cubic spline regression model with a second-order penalty term;
the spline regression model with the second-order penalty term is as follows:
Figure 684681DEST_PATH_IMAGE022
wherein, S: (x i ) As splinesS(. charpy) at observation pointx i The value of the function (c) of (c),λa penalty factor being a smoothness penalty term;Sthe expression of (dash) is at the interpolation pointε=(ε 1,ε 2,…,ε k ) The cubic spline function obtained by the above-mentioned method,εthe determination standard follows the GCV criterion;nin order to observe the number of samples,yiin order to observe the dependent variable of the sample,xiis an independent variable of the observed sample; s'2(xi) Is the square of the second derivative of the spline function;
GCV is a function of a penalty factor lambda and the number of nodes and represents the relative rate of a residual error term and the complexity of a model;
Figure 559096DEST_PATH_IMAGE023
wherein,εvalue set for interpolation points; tr () represents the trace-finding operation of the matrix,H(λ,ε) Representing that a hat matrix corresponding to the minimum value of the RSS is solved by using a least square estimation expression;
adaptively determining the number and the position of interpolation points by taking the highest fitting degree of the upper envelope line and the lower envelope line as a target; selecting a plurality of interpolation points for a time region with dense local extreme points;
assuming that the upper and lower envelopes of the estimated upper and lower local points are linese u (t) Ande d (t) Then obtain the envelope mean valuem(t) Thereby obtainingh(t)=x(t)-m(t) Judgment ofh(t) Whether or not it isIMFIf not, willh(t) Viewed as ax(t) Continuing the above steps; finally, the rest items are stopped after meeting the stop rule, andx(t) And (4) completing decomposition:
Figure 732589DEST_PATH_IMAGE024
after EMD decomposition, the frequency of the obtained product is differentIMF i (t) The signal, with increasing order, decreases in the vibration frequency in turn, decomposing the residual termr(t) Is a long-term trend term.
Fusing an EMD model and an MIDAS model, specifically:
simultaneously bringing the high-frequency factor and the low-frequency factor into a regression model, and fitting and predicting an EMD decomposed signal:
Figure 946532DEST_PATH_IMAGE025
wherein,τ t is an index of different frequencies after EMD decomposition,m l v()to comprehensively consider the intercept terms of the horizontal and ripple effect models,Sis a high frequency factor
Figure 180068DEST_PATH_IMAGE026
The number of the (c) is,s=1,2,…,S，
Figure 604096DEST_PATH_IMAGE006
is composed ofSThe coefficients corresponding to the respective high-frequency factors,Gfor the reason of the high frequency, it is,ψ k for the weight function corresponding to the high frequency,
Figure 315700DEST_PATH_IMAGE027
and
Figure 649729DEST_PATH_IMAGE008
two parameters to be estimated in the weight function,
Figure 54166DEST_PATH_IMAGE009
intThe current time stamp expressed as a microscopic factor of high frequency,kfor the purpose of its hysteresis order label,k=1,2,…,K l v()，K l v()is the maximum hysteresis order;
Figure 965490DEST_PATH_IMAGE010
is a macroscopic factor of low frequency and is,Nthe number of the macroscopic factors is the number of the macroscopic factors,n=1,2,…N,
Figure 480785DEST_PATH_IMAGE011
is the coefficient corresponding to the exogenous variable,
Figure 669321DEST_PATH_IMAGE012
and
Figure 244659DEST_PATH_IMAGE028
two parameters of the weight function.
S2, checking whether the factor distribution in the mixing model changes;
determining whether the main part of the mixing model is changed or not through mutation point detection and distribution consistency detection; the mixing model is divided into a model main part and a parameter part; the main body part comprises explanatory variables of the model and function assumptions thereof, and the parameter part comprises model hyper-parameters and parameters to be estimated; and after the model main body and the hyper-parameters are determined, estimating the parameters to be estimated according to the observation data.
The mutation point inspection and the distribution consistency inspection are periodically executed according to the observation frequency of the high-frequency data; when the distribution consistency test is used for determining whether the distribution of the financial factors is changed or not when the mutation points are found through the test, if so, the main body of the model needs to be changed to solve the problem of adaptation of the model.
The distribution consistency check comprises:
when the consistency test does not reject the original hypothesis, the catastrophe point is judged as a local punctate risk, and the model is estimated again after a training set is removed; when the model main body is changed, reconstructing the model main body part, including re-screening the features and changing the function hypothesis of the features; and fusing the estimation result of the new model under the small sample with the estimation result of the original model by using a transfer learning technology.
The main part of the mixing model comprises model factors and model hypotheses;
Figure 643279DEST_PATH_IMAGE014
and
Figure 962265DEST_PATH_IMAGE015
for model factors, linear combinations are model assumptions, weight parameters
Figure 5307DEST_PATH_IMAGE016
、
Figure 485967DEST_PATH_IMAGE017
、
Figure 43987DEST_PATH_IMAGE018
And
Figure 228981DEST_PATH_IMAGE019
is a model hyper-parameter;m l v()、
Figure 392109DEST_PATH_IMAGE020
and
Figure 309250DEST_PATH_IMAGE021
the parameters to be estimated for the model.
S3, updating the mixing model on line according to the changed factor distribution;
online updating of the model is performed using bayesian online learning in conjunction with regularized optimized gradient descent followthe regularized leader.
As shown in fig. 2, an embodiment of the present invention discloses an optimization system for an anti-time-varying model, including:
the model building module is used for building a mixing model based on the high-frequency factor;
a model checking module for checking whether the factor distribution in the mixing model changes;
and the model updating module is used for updating the mixing model on line according to the changed factor distribution.
As shown in fig. 3, the invention discloses an optimization method of a time-varying resistance model for resisting regular time-varying, which comprises the following specific steps:
1. introducing high-frequency factor to carry out mixing modeling
First, experts in the financial industry build theoretical and empirical systems for specific scenarios. For example, in a credit default scenario, where experts consider the business' operating conditions to be the most critical factors for the risk of default, the risk of credit may be low if the snap rate is kept above 100%; in the stock price risk identification scene, the experts consider that the stock price is increased after the bankruptcy reorganization. For different industries, experts also provide a prior index for pre-judging the industrial landscape situation in advance. However, since generation of these important factors depends on acquisition of data and information extraction from different sources, structures, and frequencies, the frequencies of the factors may vary greatly.
On one hand, factors with strong real-time performance need to be screened and factor synthesized by combining expert experience and a statistical method on the basis of internet big data. The internet big data comprises a large amount of real-time updated data which can be collected and analyzed by days. For example, a large number of expert perspectives can be extracted from microblogs, snowballs, and twitter and processed into a daily update of investor emotional factors using information perception and extraction techniques. For example, the recruitment information, the bidding information, and the patent declaration information can be used to synthesize growth factors for the technology companies. In addition, information such as the attention condition, price and sales volume of the goods in the manufacturing industry or the service industry can be obtained from browsing, evaluating and purchasing records of the e-commerce platform. This information forms a high frequency brand interest index for the enterprise.
The business development condition of the enterprise is the essential reason for the occurrence of the punctiform risk, and is very important for describing the development condition of the enterprise. For example, the business and risk tolerance of an enterprise is affected by the industry chain, the macro economic environment, and the monetary environment. Generally, the development condition of an industrial chain is reflected by utilizing the market ratio of upstream and downstream, the scenic condition of macroscopic economy is reflected by utilizing the increment and acceleration of important industries such as industry and the like, and the currency environment is reflected by utilizing the increment and acceleration of the scale of social financing M2. The factors are typically low frequency factors that are updated monthly, quarterly, or even annually. Although the long-term trend and stability law is accurately depicted, the method has serious hysteresis in a risk prevention scene.
On the other hand, under the condition of introducing high-frequency factors, the invention simultaneously excavates effective information of different frequency factors by using a mixing modeling method. The frequency mixing modeling technology is realized by two methods, namely factor frequency conversion and direct frequency mixing modeling.
The factor frequency conversion method is to converge high-frequency factors into low-frequency factors according to time periods. After factor frequency conversion, the high-frequency factor is converted into a low-frequency factor, and the low-frequency factor can be combined with the original low-frequency factor for modeling. The factor frequency conversion technology is widely applied in the financial field, the model of the factor frequency conversion technology has strong anti-noise capability, and the factor frequency conversion technology has better robustness in explaining and predicting long-term laws. On the other hand, however, this factor conversion method inevitably loses some information of the high frequency factor.
As shown in fig. 4, the solid line in fig. 4 is continuously observed high-frequency data of the day, and the point is lunar low-frequency data obtained by frequency conversion by a weighted average method according to a lunar window. Therefore, the variation trend of the lunar low-frequency data has obvious hysteresis compared with the high-frequency data. The downward trend is discovered from the monthly low frequency data only after the high frequency data has been down trended for a period of time. In addition, there is a clear appearance of abnormal local minimum peaks in the high frequency data of the second year, but the low frequency data has lost the perceptibility of the local peaks. Therefore, on one hand, a common frequency conversion method is improved, and a spline MIDAS sampling frequency conversion method is provided. On the other hand, an improved EMD-MIDADS model is provided on the basis of the MIDAS, and a mixing model is constructed on the basis of factors with different frequencies. The two methods are used in a fusion mode, real-time updated high-frequency factor and low-frequency factor combined modeling is achieved, and the problems that a traditional prediction model is serious in hysteresis and insufficient in prediction accuracy are solved.
The EMD-MIDADS model consists of the following two parts.
(1) EMD decomposition method based on self-adaptive node B-sample
Compared with the traditional fluctuation rate decomposition model, the EMD model has wider application range and more stable result. The EMD model can process decomposition of any type of time sequence, and has good performance in processing non-stationary and non-linear financial event sequence decomposition, particularly aiming at a financial event sequence with a mutation point.
In order to improve the effectiveness of the traditional EMD decomposition method constructed based on the cubic spline fitting value, the invention constructs a spline difference value self-adaptive node determination method by utilizing a GCV cross validation method based on a cubic spline regression model with a second-order penalty term. Cubic spline fitting is the key of EMD decomposition, and after other parameters of a spline basis function are determined, the selection of difference points directly determines the fitting effect.
The spline regression model with penalty term is as follows:
Figure 354566DEST_PATH_IMAGE022
wherein S: (x i ) For the spline estimate, λ is the smoothness penalty term. S (A), (B)x i ) The determination criteria, determined by the interpolation points, follow the GCV criteria.
The GCV is a function of a spline difference penalty term lambda and a node, represents the relative rate of a residual error term and the complexity of a model, and fully embodies the balance of the fitting effect and the robustness of the model
Figure 77671DEST_PATH_IMAGE029
And adaptively determining the number and the position of interpolation points by taking the highest fitting degree of the upper envelope and the lower envelope as a target. For time regions where the local extrema are dense, the method may accommodate selecting a plurality of interpolation points. On the other hand, the method is not influenced by the length of the waveform, and the number of the nodes can be adjusted in a self-adaptive manner when a signal is selected for decomposition in a certain period of time.
Assuming that the upper and lower envelopes of the estimated upper and lower local points are linese u (t) Ande d (t) Then obtain the envelope mean valuem(t) Thereby obtainingh(t)=x(t)-m(t) Judgment ofh(t) Whether or not it isIMFIf not, willh(t) Viewed as ax(t) And continuing the steps. And finally stopping after the rest items meet the stopping rule. The above steps are rightx(t) The decomposition is completed:
Figure 892044DEST_PATH_IMAGE030
after EMD decomposition, the frequency of the obtained product is differentIMF i (t) The signal, with increasing order, decreases in the vibration frequency in turn, decomposing the residual termr(t) The term is a long-term trend term used for explaining fixed financial laws.
(2) Fusion of EMD model and MIDAS model
In order to predict information of different frequencies of financial time sequences and finally obtain effective financial rules, the invention brings high-frequency factors and low-frequency factors into a regression model simultaneously, and performs fitting and prediction on EMD decomposed signals.
Figure 183348DEST_PATH_IMAGE031
Wherein,
Figure 715960DEST_PATH_IMAGE026
is a microscopic factor of the high frequency,
Figure 977177DEST_PATH_IMAGE032
is a macroscopic factor of low frequency and is,τ t is an index of different frequencies after EMD decomposition. Compared with the traditional MIDAS model, the model introduces high-frequency and low-frequency characteristics into the explanatory variables, thereby well avoiding the hysteresis of the model on the local short-term fluctuation prediction and ensuring the robustness of the model on the long-term trend prediction.
2. Investigating the distribution test method to determine whether the factor distribution has changed significantly
The mixing model is divided into a model body part and a parameter part. The main body part comprises explanatory variables of the model and function assumptions thereof, and the parameter part comprises model hyper-parameters and parameters to be estimated. After the model body and the hyper-parameters are determined, the parameters to be estimated can be estimated according to the observation data. The adaptation of the adaptive model to the financial distribution is primarily influenced by the model body and secondarily influenced by the parameters to be estimated. The invention determines whether the main structure of the model is changed or not through mutation point detection and distribution consistency detection.
The mutation point test and the distribution consistency test are periodically executed according to the observation frequency of the high-frequency data. When the distribution consistency test is used for determining whether the distribution of the financial factors is obviously changed or not when the mutation points are found through the test, if so, the main body of the model needs to be changed to solve the adaptation problem of the model. Of course, some distributions change more slowly and no mutation points must occur. Against this practical background, the present invention simultaneously studies distribution inspection methods that have high robustness and are sensitive to distribution change recognition. And when the original hypothesis is not rejected by the consistency test (the original hypothesis is that the distribution does not change significantly), the mutation point is judged as a local punctate risk, and the model is estimated again after the mutation point is removed from the training set. When the model body is changed, the model body part is reconstructed, and the function hypothesis of the feature is rescreened and changed. Then, the method utilizes the transfer learning technology to fuse the estimation result of the new model under the small sample with the estimation result of the original model, and solves the problems of cold start and less samples during the estimation of the new model. On the other hand, the invention researches the data sampling problem under a new scene, so that the estimation result of the new model meets the application requirement in a short time.
The main part of the above mixing model is divided into factors and model hypotheses.
Figure 911635DEST_PATH_IMAGE026
And
Figure 373840DEST_PATH_IMAGE010
for model factors, linear combinations are model assumptions, weight parameters
Figure 128170DEST_PATH_IMAGE016
、
Figure 449868DEST_PATH_IMAGE033
、
Figure 238832DEST_PATH_IMAGE018
And
Figure 606360DEST_PATH_IMAGE019
is a model hyper-parameter.m l v()、
Figure 113565DEST_PATH_IMAGE020
And
Figure 982163DEST_PATH_IMAGE021
the parameters to be estimated for the model.
When the distribution of the financial factors is not changed, the main body and the hyper-parameters of the model are not changed. But often followed by a change in the distribution of financial factors prior to a systematic crisis outbreak.
The invention provides a mutation point detection method and a distribution comparison method based on a fusion estimation method. The conventional distribution test is based on the difference of distribution functions or the variation trend of time series. However, this method requires a long time of history data, and is sensitive to differences in the middle period of the history data and poor in the end period of the history data. This results in significant hysteresis in the verification, i.e., when a change in financial distribution has occurred, the verification method cannot sense the change in time. The method provided by the invention is stable in high-order derivative estimation, and can accurately detect the mutation point based on the high-order derivative estimation. Meanwhile, the function comparison method provided by the invention can compare the overall and local differences of a plurality of distribution functions simultaneously, and compared with the traditional distribution function overall difference based on K-S statistics and K-L distance measurement, the model is sensitive to the local differences, and the distribution test result is more stable.
3. Research and development of integration of multiple online learning technologies and realization of real-time updating of models
When the financial factor distribution is not obviously changed, the main body part of the model does not need to be adjusted, and the model can be updated in real time through an online learning technology. And Bayesian online learning is used for Bayesian online learning and regularized optimized gradient descent Follow and regularized filter combination to update the model online, so that the retraining cost of the model is saved, and the robustness of model estimation is improved.
As shown in fig. 5, the above fusion estimation, change point inspection, distribution inspection method and threshold model are linked together, so that the model can be adapted in time when the distribution changes.
The principle of the invention is as follows:
when the financial distribution is changed, the distribution test result prompts whether the model main body is changed, at the moment, the threshold model is started, and the model parameters and the function relationship hypothesis are changed, so that the model is adapted in time. The framework realizes the timely adaptation of the model when the distribution is changed, but the parameters to be estimated in the adapted model need to be estimated according to the latest data. According to the method, on one hand, the robustness of new model estimation is improved by utilizing transfer learning, and on the other hand, a novel importance sampling method is provided to supplement training data in a new scene after the rule is changed, so that the cold start problem of the adaptive model is solved. The rejection value is set in the importance degree sampling method, and a resampling technology is applied to obtain more effective (high-richness and consistent with theoretical distribution) samples under the condition that the total sampling frequency is ensured to be certain. The invention solves the resampling problem under the condition by using an RR-RC-SIS sampling algorithm and a Gibbs sampling algorithm. The specific process is as follows:
the RR-RC-SIS sampling algorithm is an improved algorithm for RC-SIS (sampling rejection-sequence importance sampling), and the selection of the sampling rejection check point combines redrawing (redrawing) and resampling (resampling) sampling methods on the basis of the RR-RC-SIS sampling algorithm. The reliability of the checkpoint is first evaluated by redrawing samples and the less reliable checkpoint is replaced with a new sample. The generated samples are then optimized by resampling, thereby preserving the sample's characteristics. The specific algorithm is as follows:
1. assume that the checkpoint sequence is: 0<t 1<t 2<…<t k<…<t l ≤dAnd a threshold sequence oft 1,t 2,…,t k,…,t l 。
2. For each check pointt kOrder sequencex tk Are all based on threshold pointsc k And weightx tk A rejection sampling is performed.
3. If it is notx tk Pass the check pointt kSet its weight tow tk Then performing standard sequence importance sampling; if it is notx tk Failed checkpointt kTo what is neededt kRedrawing sampling (redrawing) until newx tk Can pass the checkpoint.
4. In generatingxIn the process of (1), if checking the pointtIs redrawn, and then the check point is checkedtResampling (resampletechnique) is performed. If the new sequence passes the checkpoint, the weight of the sequence is reset tow tk Andc k is measured. Then, all the weights are weighted againw tk And
Figure 625634DEST_PATH_IMAGE034
multiplying to generate new weight
Figure 429642DEST_PATH_IMAGE035
And then performing a round of sequence importance sampling 1-4.
5. And obtaining the enterprise sampling data with the required quantity after the N times of steps.
Gibbs sampling is primarily used for sampling of multidimensional data, assuming data dimensions ofnUsing conditional probabilities between features, from the conditional probabilities
Figure 158564DEST_PATH_IMAGE036
Obtained by intermediate sampling
Figure 502958DEST_PATH_IMAGE037
Multiple sampling ofnThe set of dimensional samples is the sample set. The specific algorithm is as follows:
1. smooth distribution of inputπ(x 1,x 2,…,x n ) Setting the number of transitions N and the number of samples M
2. Random sampling initial state
Figure 328831DEST_PATH_IMAGE038
3. Cycle t =0 to N + M-1
3.1 from the conditional probability distribution
Figure 100478DEST_PATH_IMAGE039
Middle sampling to obtain sampleBook (I)
Figure 519958DEST_PATH_IMAGE040
3.2 from the conditional probability distribution
Figure 668043DEST_PATH_IMAGE041
Sampling to obtain a sample
Figure 348423DEST_PATH_IMAGE042
…
N from the conditional probability distribution
Figure 290971DEST_PATH_IMAGE043
Sampling to obtain a sample
Figure 932168DEST_PATH_IMAGE044
Wherein the distribution is smoothπ(x 1,x 2,…,x n ) Indicates that after infinite transfer, stay in (x 1,x 2,…,x n ) Probability of state. The index with the joint distribution in the simulation individual case can be sampled through the Gibbs sampling algorithm.
As shown in fig. 6, when the model structure does not change significantly, the model needs to be updated in real time according to the latest high-frequency data. Therefore, the invention develops an FTRL + ADF online learning framework, and realizes the real-time update of the model by using the calculated amount with low cost.
Based on the technology, the invention constructs an adaptive learning framework based on a mixing model, and the framework is shown in FIG. 7.
The self-adaptive modeling framework is based on a mixing model, whether the main structure and the main parameters of the model are changed or not is judged through a distribution inspection method, and the model is updated in real time at the minimum cost through the combination of an online learning technology and a fusion estimation technology. The self-adaptive modeling framework formed by alternately using the distribution inspection and the online learning can better resist the time-varying property of financial rules, and the timeliness and the precision of mutation risk prediction are improved while the long-term trend prediction robustness is improved.
According to the invention, by providing the high-frequency factor mixing model and the self-adaptive modeling framework, the problem of influence of time-varying property of the financial rule on the prediction model is effectively solved, meanwhile, the hysteresis of the risk prediction model is effectively avoided, and the capability of timely adapting of the model along with the change of the financial rule is improved. Based on the framework, the effective prediction of economic and financial targets such as macroscopic economic indexes, mesoscopic industrial landscape, micro enterprise revenue fluctuation and the like can be realized, and dynamic risk early warning can be effectively carried out.
The invention provides computer equipment. The computer device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. The processor realizes the steps of the above-mentioned method embodiments when executing the computer program. Alternatively, the processor implements the functions of the modules/units in the above device embodiments when executing the computer program.
The computer program may be partitioned into one or more modules/units that are stored in the memory and executed by the processor to implement the invention.
The computer device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The computer device may include, but is not limited to, a processor, a memory.
The processor may be a Central Processing Unit (CPU), other general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc.
The memory may be used to store the computer programs and/or modules, and the processor may implement various functions of the computer device by running or executing the computer programs and/or modules stored in the memory and calling data stored in the memory.
The computer device integrated modules/units, if implemented in the form of software functional units and sold or used as separate products, may be stored in a computer readable storage medium. Based on such understanding, all or part of the flow of the method according to the embodiments of the present invention may also be implemented by a computer program, which may be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method embodiments may be implemented. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer memory, Read-only memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, etc. It should be noted that the computer readable medium may contain content that is subject to appropriate increase or decrease as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, computer readable media does not include electrical carrier signals and telecommunications signals as is required by legislation and patent practice.
Example 1:
before the 2008 loan crisis outbreak, the income distribution of the population entering the credit market has changed significantly, and the income distribution is illustrated schematically in fig. 8 below. The income distribution has obvious left shift in the density function of the income distribution of the crowd entering the credit market during the outbreak of the financial crisis. Indicating a significant drop in income for this segment of the population. The results of the distribution consistency test (K-S) show that the revenue distribution function is significantly different from the normal year. In this case, the mixing model is already unable to adapt to the new situation, and therefore the subject or hyper-parameters of the model need to be changed.
Example 2:
the new crown epidemic situation outbreaked in 2020, the traditional prediction model can effectively capture the factor rule and characteristics when the economy runs stably, and the accurate prediction is completed. However, when the system encounters such a major accident and a long time after the accident, the traditional model cannot capture the time variation of the factor and the law which influence the economic operation because of drastic change, so that the prediction effect is greatly reduced. The reason is that the factor systems used in the traditional modeling are always monthly frequency or even quarterly frequency, and the low-frequency index is enough to enable the model to adapt to the slow economic change rule when the economy runs stably. And due to the lag characteristic of the release time of the economic indicators, the low-frequency data is not easy to monitor and flexibly update the to-be-predicted indicators in time. As shown in table 1, a low frequency factor system is often used when a prediction model is built for the total production value of a region.
Figure 883943DEST_PATH_IMAGE045
However, when a new crown epidemic comes suddenly, most of the fields of urban economy suffer from huge impact, and economic indexes fluctuate dramatically. At the moment, the traditional low-frequency factor index and the model cannot be adapted to a new economic rule in time, so that the main body and parameters of the model are updated. At this time, high-frequency indexes (such as a cycle frequency, even a day frequency and the like) are required to be introduced, and the economic change condition is monitored and updated in real time; on the basis of establishing a high-frequency factor system, the self-adaptive frequency mixing and online learning technology provided by the project is used, the model main body and parameters are updated through the technology, and the new economic rule time-varying property in a special period is resisted in time, so that the accurate research and judgment and the timely updating of the low-frequency macroscopic index are realized.
Figure 418830DEST_PATH_IMAGE046
In addition, in order to resist against the time-varying economic law, the method also needs to rely on expert experience adjustment and guidance, and the influence period and degree of the emergency are judged in advance through the study of learning and historical specific similar events based on the law of a knowledge base, so that the model prediction result is finally corrected. FIG. 8 shows the prediction result after combining the high frequency factor and expert experience, which shows that the method can effectively overcome the time-varying property of the economic rule in time and accurately predict the trend of the economic index.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes will occur to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. A method for optimizing a time-varying resistance model is characterized by comprising the following steps:
constructing a mixing model based on the high-frequency factor, wherein the constructing of the mixing model specifically comprises the following steps: modeling through factor frequency conversion and frequency mixing; the factor frequency conversion is to converge high-frequency factors into low-frequency factors according to a time period; after factor frequency conversion, converting the high-frequency factor into a low-frequency factor, and performing combined modeling with the original low-frequency factor; the mixing modeling is to construct an EMD-MIDADS model based on different spelling factors; the construction of the EMD-MIDADS model specifically comprises the following steps: performing EMD decomposition on the B-sample based on the adaptive node; fusing an EMD model and an MIDAS model;
checking whether the factor distribution in the mixing model changes;
and performing online updating on the mixing model according to the changed factor distribution, wherein the online updating uses Bayesian online learning and regular optimization gradient descent to perform online updating on the model.
2. The optimization method of the time-varying resistance model according to claim 1, wherein the EMD decomposition of the B-samples based on the adaptive nodes comprises:
constructing a spline difference value self-adaptive node determination method by utilizing a GCV cross validation method based on a cubic spline regression model with a second-order penalty term;
the spline regression model with the second-order penalty term is as follows:
Figure 722946DEST_PATH_IMAGE001
wherein, S: (x i ) As splinesS(. charpy) at observation pointx i The value of the function (c) of (c),λa penalty factor being a smoothness penalty term;Sthe expression of (dash) is at the interpolation pointε=(ε 1,ε 2,…,ε k ) The cubic spline function obtained by the above-mentioned method,εthe determination standard follows the GCV criterion;nin order to observe the number of samples,yiin order to observe the dependent variable of the sample,xiis an independent variable of the observed sample; s'2(xi) Is the square of the second derivative of the spline function;
GCV is a function of a penalty factor lambda and the number of nodes and represents the relative rate of a residual error term and the complexity of a model;
Figure 537319DEST_PATH_IMAGE002
wherein,εvalue set for interpolation points; tr () represents the trace-finding operation of the matrix,H(λ,ε) Representing that a hat matrix corresponding to the minimum value of the RSS is solved by using a least square estimation expression;
adaptively determining the number and the position of interpolation points by taking the highest fitting degree of the upper envelope line and the lower envelope line as a target; selecting a plurality of interpolation points for a time region with dense local extreme points;
assuming that the upper and lower envelopes of the estimated upper and lower local points are linese u (t) Ande d (t) Then obtain the envelope mean valuem(t) Thereby obtainingh(t)=x(t)-m(t) Judgment ofh(t) Whether or not it isIMFIf not, willh(t) Viewed as ax(t) Continuing the above steps; finally, the rest items are stopped after meeting the stop rule, andx(t) And (4) completing decomposition:
Figure 828623DEST_PATH_IMAGE003
after EMD decomposition, the frequency of the obtained product is differentIMF i (t) The signal, with increasing order, decreases in the vibration frequency in turn, decomposing the residual termr(t) Is a long-term trend term.
3. The optimization method of the time-varying resistance model according to claim 1, wherein the fusing of the EMD model and the MIDAS model includes:
simultaneously bringing the high-frequency factor and the low-frequency factor into a regression model, and fitting and predicting an EMD decomposed signal:
Figure 361235DEST_PATH_IMAGE004
wherein,τ t is an index of different frequencies after EMD decomposition,m l v()to comprehensively consider the intercept terms of the horizontal and ripple effect models,Sis a high frequency factor
Figure 888031DEST_PATH_IMAGE005
The number of the (c) is,s=1,2, …,S，
Figure 556910DEST_PATH_IMAGE006
is composed ofSThe coefficients corresponding to the respective high-frequency factors,Gfor the reason of the high frequency, it is,ψ k for the weight function corresponding to the high frequency,
Figure 19116DEST_PATH_IMAGE007
and
Figure 39024DEST_PATH_IMAGE008
two parameters to be estimated in the weight function,
Figure 103932DEST_PATH_IMAGE009
intThe current time stamp expressed as a microscopic factor of high frequency,kfor the purpose of its hysteresis order label,k=1,2, …,K l v()，K l v()is the maximum hysteresis order;
Figure 892897DEST_PATH_IMAGE010
is a macroscopic factor of low frequency and is,Nthe number of the macroscopic factors is the number of the macroscopic factors,n=1,2,…N,
Figure 526003DEST_PATH_IMAGE011
is the coefficient corresponding to the exogenous variable,
Figure 502050DEST_PATH_IMAGE012
and
Figure 308331DEST_PATH_IMAGE013
two parameters of the weight function.
4. The method of claim 1, wherein the checking whether the distribution of factors in the mixture model changes comprises:
determining whether the main part of the mixing model is changed or not through mutation point detection and distribution consistency detection; the mixing model comprises a model body part and a parameter part; the main body part comprises explanatory variables of the model and function assumptions thereof, and the parameter part comprises model hyper-parameters and parameters to be estimated; and after the model main body and the hyper-parameters are determined, estimating the parameters to be estimated according to the observation data.
5. The optimization method of the time-varying resistance model according to claim 4, wherein the mutation point test and the distribution consistency test are periodically performed according to the observation frequency of the high-frequency data; and when the distribution consistency test is used for determining whether the distribution of the financial factors is changed or not when the mutation points are found by the test, and if the distribution of the financial factors is changed, changing the main body of the model to solve the problem of adaptation of the model.
6. The method of optimizing a model of resistance to temporal variations according to claim 5, wherein the distribution consistency test comprises:
when the consistency test does not reject the original hypothesis, the catastrophe point is judged as a local punctate risk, and the model is estimated again after a training set is removed; when the model main body is changed, reconstructing the model main body part, including re-screening the features and changing the function hypothesis of the features; and fusing the estimation result of the new model under the small sample with the estimation result of the original model by using a transfer learning technology.
7. The method of claim 5, wherein the body portion of the mixture model comprises model factors and model hypotheses;
Figure 279699DEST_PATH_IMAGE014
and
Figure 880444DEST_PATH_IMAGE015
for model factors, linear combinations are model assumptions, weight parameters
Figure 812628DEST_PATH_IMAGE016
、
Figure 157022DEST_PATH_IMAGE017
、
Figure 982895DEST_PATH_IMAGE018
And
Figure 754542DEST_PATH_IMAGE019
is a model hyper-parameter;m l v()、
Figure 174022DEST_PATH_IMAGE020
and
Figure 322107DEST_PATH_IMAGE021
the parameters to be estimated for the model.
8. A system for optimizing a time-varying resistance model, comprising:
the model building module is used for building a mixing model based on the high-frequency factor, and the building of the mixing model specifically comprises the following steps: modeling through factor frequency conversion and frequency mixing; the factor frequency conversion is to converge high-frequency factors into low-frequency factors according to a time period; after factor frequency conversion, converting the high-frequency factor into a low-frequency factor, and performing combined modeling with the original low-frequency factor; the mixing modeling is to construct an EMD-MIDADS model based on different spelling factors; the construction of the EMD-MIDADS model specifically comprises the following steps: performing EMD decomposition on the B-sample based on the adaptive node; fusing an EMD model and an MIDAS model;
a model checking module for checking whether the factor distribution in the mixing model changes;
and the model updating module is used for updating the mixing model on line according to the changed factor distribution, and the model is updated on line by combining Bayesian online learning and regular optimization gradient descent.
9. A computer arrangement comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1-7 when executing the computer program.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.
CN202210099881.1A 2022-01-27 2022-01-27 Optimization method, system, equipment and readable storage medium of time-varying-resistance model Pending CN114117355A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202210099881.1A CN114117355A (en) 2022-01-27 2022-01-27 Optimization method, system, equipment and readable storage medium of time-varying-resistance model
PCT/CN2022/090222 WO2023142288A1 (en) 2022-01-27 2022-04-29 Method and system for optimizing time-variation resistant model, and device and readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210099881.1A CN114117355A (en) 2022-01-27 2022-01-27 Optimization method, system, equipment and readable storage medium of time-varying-resistance model

Publications (1)

Publication Number Publication Date
CN114117355A true CN114117355A (en) 2022-03-01

Family

ID=80361704

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210099881.1A Pending CN114117355A (en) 2022-01-27 2022-01-27 Optimization method, system, equipment and readable storage medium of time-varying-resistance model

Country Status (2)

Country Link
CN (1) CN114117355A (en)
WO (1) WO2023142288A1 (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116340796A (en) * 2023-05-22 2023-06-27 平安科技（深圳）有限公司 Time sequence data analysis method, device, equipment and storage medium
WO2023142288A1 (en) * 2022-01-27 2023-08-03 平安科技（深圳）有限公司 Method and system for optimizing time-variation resistant model, and device and readable storage medium
CN117670413A (en) * 2023-12-13 2024-03-08 中教畅享科技股份有限公司 Market crowd behavior-based market prediction method

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117674199B (en) * 2024-02-01 2024-04-23 西安热工研究院有限公司 Power system frequency modulation method and system of super-capacitor coupling lithium battery

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120150570A1 (en) * 2009-08-20 2012-06-14 Ali Samad-Khan Risk assessment/measurement system and risk-based decision analysis tool
CN108876106A (en) * 2018-05-14 2018-11-23 华南理工大学 A kind of macro Micro dynamic mixing β estimation method merging macro economic policy factor
CN109101731A (en) * 2018-08-15 2018-12-28 电子科技大学 Time-varying reliability sequence optimisation method based on B-distance
WO2021195660A1 (en) * 2020-03-27 2021-09-30 Markov Processes International Inc. System and method for estimating multifactor models of illiquid, stale and noisy financial and economic data
CN114117355A (en) * 2022-01-27 2022-03-01 平安科技（深圳）有限公司 Optimization method, system, equipment and readable storage medium of time-varying-resistance model

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023142288A1 (en) * 2022-01-27 2023-08-03 平安科技（深圳）有限公司 Method and system for optimizing time-variation resistant model, and device and readable storage medium
CN116340796A (en) * 2023-05-22 2023-06-27 平安科技（深圳）有限公司 Time sequence data analysis method, device, equipment and storage medium
CN116340796B (en) * 2023-05-22 2023-12-22 平安科技（深圳）有限公司 Time sequence data analysis method, device, equipment and storage medium
CN117670413A (en) * 2023-12-13 2024-03-08 中教畅享科技股份有限公司 Market crowd behavior-based market prediction method

Also Published As

Publication number Publication date
WO2023142288A1 (en) 2023-08-03

Similar Documents

Publication Publication Date Title
CN114117355A (en) Optimization method, system, equipment and readable storage medium of time-varying-resistance model
CN112131212A (en) Hybrid cloud scene-oriented time sequence data anomaly prediction method based on ensemble learning technology
Chen et al. Multi-step-ahead crude oil price forecasting using a hybrid grey wave model
Longo et al. A neural network ensemble approach for GDP forecasting
CN103186575B (en) A kind of clustering method of sensing data and system
Hopp Economic nowcasting with long short-term memory artificial neural networks (LSTM)
Saltelli et al. Climate models as economic guides scientific challenge or quixotic quest?
Assaad et al. Predicting the price of crude oil and its fluctuations using computational econometrics: deep learning, LSTM, and convolutional neural networks
US20210397956A1 (en) Activity level measurement using deep learning and machine learning
CN114580545A (en) Wind turbine generator gearbox fault early warning method based on fusion model
Zhang et al. Analysis and Application of Grey‐Markov Chain Model in Tax Forecasting
Du et al. Research on stock forecasting based on random forest
Huang et al. A decomposition‐based multi‐time dimension long short‐term memory model for short‐term electric load forecasting
CN116149895A (en) Big data cluster performance prediction method and device and computer equipment
Azizi et al. Intermittent Demand Forecasting Using LSTM With Single and Multiple Aggregation
Kotenko et al. Formation of Indicators for Assessing Technical Reliability of Information Security Systems
Ruhal et al. A Comparative Study Of Statistical Methods And Machine Learning Approaches For Stock Price Prediction
CN115221024A (en) Database performance short-time early warning method, model training method, device and equipment
Alam Predicting revenues and expenditures using artificial neural network and autoregressive integrated moving average
Shukla et al. Sentiment analysis of international relations with artificial intelligence
Qureshi et al. A comparative analysis of traditional SARIMA and machine learning models for CPI data modelling in Pakistan
Jackson et al. Machine learning for classification of economic recessions
Scaffidi Domianello et al. Smooth and Abrupt Dynamics in Financial Volatility: The MS‐MEM‐MIDAS
CN115222113B (en) CNN-LSTM-based railway area freight volume short-term prediction method
Alamsyah et al. Comparison of Predictions Foreign Currency Exchange Rates (USD, JPY, EUR, GBP, CHF, CAD) to Rupiah (IDR) using Artificial Neural Network and Long Short-Term Memory

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20220301