Prediction model acquisition method and device, equipment, medium and program product

Abstract

The invention provides a prediction model acquisition method and device, equipment, a medium and a program product, relates to the technical field of artificial intelligence, and can solve the problem of relatively poor accuracy of predicting the customer migration quantity. According to the specific technical scheme, a time sequence data set corresponding to the customer migration number in a first time period is obtained, and the time sequence data set comprises customer migration parameters corresponding to each first unit duration in the first time period; based on customer migration parameters in the time sequence data set, seasonal parameters and model parameters are determined; performing model initialization by using the seasonal parameters and the model parameters, and constructing an SARIMA model; the time sequence data set, the seasonal parameters and the model parameters are used for fitting the SARIMA model, a prediction model is obtained, and the prediction model is used for predicting the migration number of customers with seasonal changes. The method is used for predicting customer migration data volume more accurately.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, medium, and program product for obtaining a prediction model. Background Technology

[0002] In commercial bank operations, branch customer management is a critical task. Typically, customers are tiered based on their Assets Under Management (AUM) to ensure appropriate services and products are provided to customers with different net worths. During this process, a customer's management level is dynamically adjusted according to fluctuations in their AUM. If a customer's AUM fails to meet targets for several consecutive months, a downgrade in management level is triggered.

[0003] To meet the need for refined customer management operations, the number of customers migrating down can be predicted, allowing for proactive decision-making based on the forecast results. Currently, this prediction is typically based on simple rules or Autoregressive Integrated Moving Average (ARIMA) models. Specifically, the simple rule can be a manually set business rule, such as setting a fixed threshold and duration; if a customer's AUM (Amount in Us) does not reach the threshold within that duration, the customer is considered to have a high probability of migrating. The ARIMA model can capture the short-term trends and fluctuations of a customer's AUM and, combined with analysis of the customer's historical AUM, predict customer migration.

[0004] However, predicting customer migration data based on simple rules relies entirely on the customer's historical AUM and cannot consider future trends in AUM; while the ARIMA model can only capture short-term trends and fluctuations in the customer's AUM. This results in poor accuracy in predicting the number of customers migrating. Summary of the Invention

[0005] This application provides a method, apparatus, device, medium, and program product for obtaining a prediction model. By incorporating seasonal components into the processing of time-series data through the prediction model, it can better identify and capture periodic fluctuations in customer AUM. Effectively modeling these seasonal characteristics makes the prediction of customer migration data volume more accurate.

[0006] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: Firstly, a method for obtaining a prediction model is provided. This method includes: acquiring a time-series dataset corresponding to the number of customers migrating down within a given time period, wherein the time-series dataset includes customer migration parameters for each unit of time within that period, and the customer migration parameters include the number of customers migrating down within a unit of time and the asset management scale information corresponding to the number of customers migrating down; then, based on the customer migration parameters in the time-series dataset, determining seasonal parameters and model parameters, wherein the seasonal parameters include P, D, Q, and s, and the model parameters include p, d, and q, where P is the seasonal autoregression order, D is the seasonal differencing frequency, Q is the seasonal moving average order, s is the number of units within a predetermined unit of time, p is the non-seasonal autoregression order, d is the non-seasonal differencing frequency, and q is the non-seasonal moving average order; next, using the seasonal parameters and model parameters, initializing the model to construct a SARIMA model; finally, fitting the SARIMA model using the time-series dataset, seasonal parameters, and model parameters to obtain a prediction model for predicting the number of customers migrating down with seasonal variations.

[0007] Thus, since the prediction model is obtained by fitting the SARIMA prediction model constructed using the aforementioned model parameters and seasonal parameters, this prediction model can incorporate seasonal components when processing time series data, thereby better identifying and capturing periodic fluctuations in customer AUM. By effectively modeling these seasonal characteristics, the prediction of customer migration data volume becomes more accurate.

[0008] In one possible implementation of the first aspect, an initial time series dataset corresponding to the number of customers migrating down within the aforementioned time period can be obtained first. The initial time series dataset includes the initial customer migration parameters corresponding to each unit duration within the time period. Then, a stationarity test is performed on the initial customer migration parameters in the initial time series dataset. Based on the test results of the stationarity test, the aforementioned time series dataset is determined.

[0009] Thus, since the initial customer migration parameters in the initial time series dataset can be tested for stationarity first, and then the time series dataset can be determined based on the test results, it can be ensured that the time series dataset is relatively stable, which is beneficial to improving the accuracy of building the SARIMA prediction model.

[0010] In another possible implementation of the first aspect, if the test result is greater than or equal to the first threshold, the initial customer migration parameters in the initial time series dataset can be differentially processed to obtain the time series dataset; or, if the test result is less than the first threshold, the initial time series dataset can be determined as the time series dataset.

[0011] Thus, since the initial time series dataset can be differentially processed based on the relationship between the stationarity test result and the first threshold to obtain the time series dataset, or the initial time series dataset can be determined as the time series dataset, a time series dataset with better stationarity can be flexibly determined based on whether the initial time series dataset is stationary.

[0012] In another possible implementation of the first aspect, the customer migration parameters in the aforementioned time series dataset can be substituted into the autocorrelation function to calculate Q and q; and the customer migration parameters in the aforementioned time series dataset can be substituted into the partial autocorrelation function to calculate P and p; and the number of differential processing times corresponding to the customer migration parameters in the aforementioned time series dataset can be determined as D and d; and the number of the aforementioned unit durations in 1 year can be determined as s.

[0013] Thus, since Q and q can be determined by the autocorrelation function, P and p by the partial autocorrelation function, D and d by the difference times corresponding to the customer migration parameters, and s by the number of the first unit duration in a year, the model parameters and seasonal parameters can be determined according to the characteristics of each parameter in a corresponding way, so as to ensure that the model parameters and seasonal parameters can be determined quickly and accurately, thereby improving the accuracy of constructing the SARIMA prediction model.

[0014] In another possible implementation of the first aspect, after obtaining the above prediction model, the number of customers migrating down within a historical time period can be processed using the prediction model to obtain the predicted number of customers migrating down within a future time period.

[0015] Thus, after obtaining the prediction model, the number of customers migrating in the future period can be obtained by processing the number of customers migrating in the historical period through the prediction model. Therefore, the prediction model can be used to predict the number of customers migrating, making the prediction of the amount of customer migration data more accurate and convenient.

[0016] In another possible implementation of the first aspect, after obtaining the predicted number of customers migrating to a certain future time period, a residual analysis can be performed on the predicted number of customers migrating to that time period and the actual number of customers migrating to that time period; and if the analysis result of the residual analysis is greater than or equal to a threshold, and the preset model evaluation index is greater than or equal to another threshold, the SARIMA model is reconstructed.

[0017] Thus, after predicting the number of customers migrating to lower-level cities in the future, the prediction model can be evaluated by residual analysis of the predicted and actual values, as well as by model evaluation metrics. If the residual analysis result is greater than or equal to the second threshold and the model evaluation metric is greater than or equal to the third threshold, the SARIMA model can be reconstructed, thereby ensuring that the SARIMA model used always has good performance.

[0018] Secondly, a predictive model acquisition device is provided, comprising: an acquisition module for acquiring a time series dataset corresponding to the number of customers migrating down within a first time period, wherein the time series dataset includes customer migration parameters corresponding to each first unit of time within the first time period, and the customer migration parameters include the number of customers migrating down within a first unit of time and the asset management scale information corresponding to the number of customers migrating down; a determination module for determining seasonal parameters and model parameters based on the customer migration parameters in the time series dataset, wherein the seasonal parameters include P, D, Q, and s, and the model parameters include p, d, and q, where P is the seasonal autoregression order, D is the seasonal differencing frequency, Q is the seasonal moving average order, s is the number of first unit of time within a predetermined unit of time, p is the non-seasonal autoregression order, d is the non-seasonal differencing frequency, and q is the non-seasonal moving average order; a construction module for initializing the model using the seasonal parameters and model parameters to construct a SARIMA model; and a fitting module for fitting the SARIMA model using the time series dataset, seasonal parameters, and model parameters to obtain a predictive model, which is used to predict the number of customers migrating down with seasonal variations.

[0019] Thirdly, an electronic device is provided, the method comprising: a memory and at least one processor. The memory is communicatively connected to the processor. The memory is used to store computer program code, the computer program code including computer instructions. When the processor executes the computer instructions, it causes the electronic device to perform the method as described in the first aspect and any possible implementation thereof.

[0020] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions. When executed by a processor, these computer instructions are used to implement the method described in the first aspect and any possible implementation thereof.

[0021] Fifthly, embodiments of this application provide a computer program product that, when run on a computer or executed by a computer's processor, implements the method described in the first aspect and any possible design thereof. The computer may be the electronic device described in the third aspect and any possible implementation thereof.

[0022] It is understood that the beneficial effects achieved by the prediction model acquisition device described in the second aspect, the electronic device described in the third aspect, the computer-readable storage medium described in the fourth aspect, and the computer program product described in the fifth aspect can be referred to as the beneficial effects in the first aspect and any possible implementation thereof, and will not be repeated here. Attached Figure Description

[0023] Figure 1 A flowchart of a method for obtaining a prediction model is provided for some embodiments of this application; Figure 2 A flowchart of a method for obtaining a prediction model is provided for some embodiments of this application; Figure 3 A flowchart of a method for obtaining a prediction model is provided for some embodiments of this application; Figure 4 A flowchart of a method for obtaining a prediction model is provided for some embodiments of this application; Figure 5 A flowchart of a method for obtaining a prediction model is provided for some embodiments of this application; Figure 6 A flowchart of a method for obtaining a prediction model is provided for some embodiments of this application; Figure 7 A flowchart of a method for obtaining a prediction model is provided for some embodiments of this application; Figure 8 A schematic diagram of the structure of a predictive model acquisition device provided for some embodiments of this application; Figure 9 This is a schematic diagram of the structure of an electronic device provided for some embodiments of this application. Detailed Implementation

[0024] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0026] The technical solutions provided in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data, comply with relevant laws and regulations and do not violate public order and good morals.

[0027] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0028] The following explains some concepts and terms involved in the prediction model acquisition method, apparatus, device, medium and program products provided in the embodiments of this application.

[0029] AUM: Refers to the total client assets managed by a bank or financial institution. It is an important indicator for measuring the size of client assets and is often used as one of the key criteria for client management level.

[0030] Customer management tiers: This refers to the method by which banks classify and manage customers based on indicators such as asset size and contribution. Customer management tiers are typically divided into categories such as high-net-worth customers, medium-net-worth customers, and ordinary customers.

[0031] Client downgrade: This refers to the process where a client's client management level is lowered because they fail to meet the corresponding AUM requirements within a specified time. For example, a high-net-worth client may be downgraded to a medium-net-worth client if their AUM fails to meet the requirements for several consecutive months.

[0032] The Seasonal Autoregressive Integrated Moving Average (SARIMA) model is a time series analysis model that incorporates seasonal factors to handle time series data with trends and seasonal fluctuations. The SARIMA model performs predictions and modeling by processing historical data through autoregression, differencing, moving averages, and seasonal components.

[0033] Autocorrelation Function (ACF): A function used to measure the correlation of time series data at different time lags.

[0034] Partial Autocorrelation Function (PACF): A function used to measure the relationship between a time series and its own lagged values, eliminating the direct correlation caused by intermediate lags.

[0035] Residual: refers to the error between the model's predicted values ​​and the actual observed values. Ideally, the residual should be white noise, i.e., without autocorrelation and randomly distributed.

[0036] Seasonality refers to periodic fluctuations in time-series data, such as similar trends in certain months or quarters of a year. In bank customer management, seasonality may manifest as larger fluctuations in customer AUM (Assets Under Management) in certain months.

[0037] The following description, in conjunction with the accompanying drawings, details the prediction model acquisition method, apparatus, device, medium, and program product provided in this application through specific embodiments and application scenarios.

[0038] In commercial bank operations, branch customer management is a critical task. Banks typically manage customers in tiers based on their assets under management (AUM) to ensure that appropriate services and products are provided to customers with different net worths. During this process, a customer's management level is dynamically adjusted according to fluctuations in their AUM. High-net-worth clients usually enjoy better services and products, but if a client's AUM falls short of the required level for several consecutive months, it will trigger a downgrade in management level, commonly known as "relegation."

[0039] With increasingly fierce competition, branches must refine their customer management practices, effectively predict customer migration trends, and make informed decisions in advance. Time-series data forecasting methods can help branches predict the number of customers who may migrate within a future period based on historical data, and make corresponding resource allocation and strategy adjustments.

[0040] In current branch operation practices, there are already various solutions for addressing changes in customer status, which can be mainly categorized as follows: 1. Migration Management Based on Simple Rules Migration management based on simple rules relies primarily on manually set business rules, such as triggering migration when a customer's AUM falls below a certain threshold for three consecutive months. While this method is simple and direct, it lacks the ability to predict future changes and cannot account for fluctuations caused by seasonal factors.

[0041] 2. Time series forecasting based on the ARIMA model ARIMA models are a common and widely used time series forecasting method. In some banking practices, ARIMA is used to predict changes in customer assets under management (AUM) and, by analyzing historical data, to predict the probability of customers migrating down.

[0042] However, relocation management based on simple rules lacks predictive capabilities. This method relies entirely on historical data, fails to consider future trend changes, cannot provide early warnings for branches, and lacks flexibility. Furthermore, this method cannot adequately account for the impact of market conditions, seasonal fluctuations, and other factors, easily leading to erroneous management decisions.

[0043] While the ARIMA model, a commonly used time series forecasting tool, can capture trends and short-term fluctuations in customer AUM data, it cannot effectively capture seasonal fluctuations. However, seasonal fluctuations are common in bank customer AUM data. For example, increased customer liquidity at the end of quarters or years, or fluctuations in funds due to market rhythms and economic policies, all affect changes in AUM. The ARIMA model is insufficient in handling such cyclical changes and cannot effectively address seasonal fluctuations, especially when customer AUM exhibits significant seasonal variations; its predictive performance is poor, failing to accurately reflect the seasonal fluctuations in customer AUM.

[0044] This results in poor accuracy in predicting the number of customers migrating down.

[0045] To address the aforementioned issues, embodiments of this application provide a method, apparatus, device, medium, and program product for obtaining a prediction model. The prediction model acquisition method provided in this application can be applied to scenarios involving the prediction of customer migration numbers.

[0046] For example, a time-series dataset corresponding to the number of customers migrating down within the previous 36 months can be obtained at the current moment of the system. This time-series dataset includes customer migration parameters for each month within that time period, including the number of customers migrating down within a month and the corresponding asset management scale information. Then, based on the customer migration parameters in the time-series dataset, seasonal parameters and model parameters are determined. The seasonal parameters include P, D, Q, and s, and the model parameters include p, d, and q. P is the seasonal autoregression order, D is the seasonal differencing frequency, Q is the seasonal moving average order, s is the number of months within a predetermined unit of time, p is the non-seasonal autoregression order, d is the non-seasonal differencing frequency, and q is the non-seasonal moving average order. Next, the seasonal parameters and model parameters are used to initialize the model and construct a SARIMA model. Finally, the time-series dataset, seasonal parameters, and model parameters are used to fit the SARIMA model to obtain a prediction model for predicting the number of customers migrating down with seasonal variations.

[0047] Thus, since the prediction model is obtained by fitting the SARIMA prediction model constructed using the aforementioned model parameters and seasonal parameters, this prediction model can incorporate seasonal components when processing time series data, thereby better identifying and capturing periodic fluctuations in customer AUM. By effectively modeling these seasonal characteristics, the prediction of customer migration data volume becomes more accurate.

[0048] The prediction model acquisition method provided in this application's embodiments addresses the shortcomings of related technologies in handling seasonal fluctuations and time series trends by introducing seasonal components. Specific advantages are as follows: 1. Effectively capture seasonal fluctuations The aforementioned forecasting model, by incorporating seasonal autoregression, seasonal differencing, and seasonal moving average terms, effectively captures seasonal variations in customer AUM data. Compared to the ARIMA model, this model can handle cyclical fluctuations in customer AUM over certain months or quarters, making it particularly suitable for time series data with significant seasonal variations.

[0049] 2. Handling trends and short-term fluctuations The aforementioned predictive model combines autoregressive and moving average components, enabling it to capture not only the long-term trend of a client's AUM but also to identify short-term fluctuations. This allows the model to effectively model changes across different time scales, improving the comprehensiveness and accuracy of the predictions.

[0050] 3. Improve forecast accuracy and enhance predictability. The aforementioned predictive model can forecast the number of customers migrating to branch offices in the coming months, thus providing early warnings of which customers may migrate due to insufficient AUM (Amount Per Unit). Based on historical data trends and seasonal variations, it can accurately predict future migration patterns, offering crucial decision support to branch managers and reducing reactive management.

[0051] 4. Automatically adapts to data characteristics of different periods The above forecasts are highly flexible and can adapt to the characteristics of data over different time periods. In practical applications, whether using monthly or quarterly data, the forecasting model can adjust the seasonality parameters according to specific business needs, thereby automatically adapting to different time series characteristics.

[0052] It should be noted that the entire process of the prediction model acquisition method provided in this application embodiment can be implemented using Python code to meet the current deployment conditions of the bank and facilitate deployment.

[0053] It should be noted that the predictive model acquisition method provided in this application can be executed by a predictive model acquisition device, an electronic device, or a functional module within an electronic device. Some embodiments of this application use an electronic device executing the predictive model acquisition method as an example to illustrate the predictive model acquisition method provided in this application.

[0054] Figure 1 A flowchart illustrating a method for obtaining a prediction model provided in some embodiments of this application. For example... Figure 1 As shown, the method for obtaining the prediction model includes the following steps 101 to 104.

[0055] Step 101: The electronic device acquires the time series dataset corresponding to the number of customers relocating within the first time period.

[0056] The aforementioned time series dataset includes customer migration parameters for each first unit of time within the first time period. These customer migration parameters include the number of customers migrating within a first unit of time and the asset management scale information corresponding to that number of customers migrating.

[0057] In some embodiments of this application, the electronic device may have pre-installed relevant Python libraries, including: pip library, install library, numpy library, pandas library, matplotlib library, and statsmodels library. The numpy and pandas libraries are primarily used for data processing and analysis; the matplotlib library can be used for data visualization; and the statsmodels library provides time series analysis tools, including the SARIMA model.

[0058] In some embodiments of this application, the first time period can be any time period before the current system time.

[0059] For example, the first time period could be 36 months, 24 months, or 12 months prior to the current system time; or it could be 12 months, 18 months, or 26 months prior to the current system time and starting from a certain historical moment.

[0060] In some embodiments of this application, the aforementioned time series dataset is a dataset in which data is arranged in chronological order, that is, the data in the time series dataset is arranged in chronological order.

[0061] In some embodiments of this application, the first unit duration can be any unit duration such as 1 day, 1 month, or 1 quarter.

[0062] In some embodiments of this application, the aforementioned asset management scale information can be AUM, and the unit of AUM can be ten thousand.

[0063] For example, assuming the first time period is 3 months prior to the current system time, and the first unit duration is 1 month, then the above time series dataset can include the customer migration parameters corresponding to each month within those 3 months; that is, the number of customers migrating in each month within those 3 months and the AUM corresponding to each number of customers migrating; that is, the number of customers migrating in the first month and the AUM corresponding to the number of customers migrating in the first month, the number of customers migrating in the second month and the AUM corresponding to the number of customers migrating in the second month, and the number of customers migrating in the third month and the AUM corresponding to the number of customers migrating in the third month.

[0064] In some embodiments of this application, the electronic device can obtain a time-series dataset corresponding to the number of customers migrating down within a first time period from a database or historical reports.

[0065] In some embodiments of this application, after obtaining the time series dataset corresponding to the number of customers migrating down within a first time period, the electronic device can create the data structure corresponding to the time series dataset using Python.

[0066] In some embodiments of this application, after the electronic device obtains the time series dataset corresponding to the number of customers migrating down within the first time period, it can use Python to draw a time series graph of the number of customers migrating down within the first time period to visualize the number of customers migrating down within the first time period, thereby facilitating the observation and analysis of the overall trend and seasonal characteristics of the number of customers migrating down within the first time period.

[0067] In some embodiments of this application, combined with Figure 1 ,like Figure 2 As shown, step 101 can be implemented through steps 101a to 101c as described below.

[0068] Step 101a: The electronic device acquires the initial time series dataset corresponding to the number of customers relocating within the first time period.

[0069] The aforementioned initial time series dataset includes the initial customer migration parameters corresponding to each first unit of time within the first time period.

[0070] In some embodiments of this application, the initial customer migration parameters may include the number of initial customers migrated within a first unit of time and the asset management scale information corresponding to the number of initial customers migrated.

[0071] For further description of step 101a, please refer to the above description of step 101. To avoid repetition, it will not be repeated here.

[0072] Step 101b: The electronic device performs a stationarity test on the initial customer migration parameters in the initial time series dataset.

[0073] It should be noted that stationarity tests are primarily used to determine whether time series data meet the requirements of statistical stability, thus ensuring the effectiveness of modeling, eliminating spurious correlations between non-stationary series, and improving the reliability of the analysis. The stability of statistics such as mean and variance is used to determine whether the data is suitable for trend analysis and prediction.

[0074] In some embodiments of this application, the above-mentioned stationarity test may also be referred to as the stationarity test of the above-mentioned initial time series dataset.

[0075] In some embodiments of this application, the electronic device can perform a stationarity test on the initial customer migration parameters in the aforementioned initial time series dataset using the Augmented Dickey-Fuller Test (ADF).

[0076] It should be noted that the ADF test is a unit root test method that introduces lag difference terms. It is used to determine whether a unit root exists in time series data, thereby verifying its stationarity and resolving the problem of higher-order autocorrelation interference. The ADF test determines the existence of a unit root by analyzing the coefficients of the first-order lag terms of the time series. When the coefficient equals 1, the model has a unit root, the series exhibits random walk characteristics, leading to non-stationarity. The test results are judged by comparing the DF statistic with the critical value table. The specific operation steps are as follows: 1. Model specification: Establish an autoregressive model that includes constant terms, trend terms, or no-trend terms; 2. Parameter estimation: The least squares method is used to estimate the lag coefficients and standard errors. By introducing lagged variables and coupled autocorrelation considerations, the accuracy of testing for autocorrelation and heteroscedasticity in time series and panel data is improved. 3. Calculation of the DF statistic: The DF statistic is derived according to the formula; 4. Critical value comparison: Determine the confidence level of the statistic by referring to the Dickie-Fowler critical value table; 5. Result interpretation: If the statistic is less than the critical value of the selected significance level, the null hypothesis is rejected and the series is determined to be stationary.

[0077] In practical applications, the statsmodels library in Python can be used to automate the tests, and the analysis can be completed by extracting the ADF statistic, p-value, and critical value.

[0078] For specific methods of stationarity testing, please refer to the relevant technical descriptions. To avoid repetition, they will not be elaborated here.

[0079] Step 101c: The electronic device determines the time series dataset based on the stationarity test results.

[0080] In some embodiments of this application, the test result of stationarity can be a p-value; the larger the p-value, the less stationary the initial time series dataset is; the smaller the p-value, the more stationary the initial time series dataset is.

[0081] In some embodiments of this application, the electronic device can determine the time series dataset based on the magnitude of the p-value.

[0082] In some embodiments of this application, since the initial customer migration parameters in the initial time series dataset can be tested for stationarity first, and then the time series dataset can be determined based on the test results, it can be ensured that the time series dataset is relatively stable, which is beneficial to improving the accuracy of constructing the SARIMA prediction model.

[0083] The following section provides a detailed explanation of the specific method used by electronic devices to determine the aforementioned time series dataset based on the results of the stationarity test.

[0084] In some embodiments of this application, combined with Figure 2 ,like Figure 3 As shown, step 101c can be implemented through step 101c1 or step 101c2 as described below.

[0085] Step 101c1: If the test result is greater than or equal to the first threshold, the electronic device performs differential processing on the initial customer migration parameters in the initial time series dataset to obtain the time series dataset.

[0086] In some embodiments of this application, the above test result can be the above p-value, and the first threshold can be any threshold such as 0.05 or 0.06. The specific threshold can be determined according to the actual implementation, and this application does not limit it.

[0087] In some embodiments of this application, when the above test result is greater than or equal to the first threshold, it can be considered that the initial time series dataset has poor stationarity; at this time, the initial customer migration parameters in the initial time series dataset can be differentially processed to eliminate non-stationary noise data, so that the obtained time series dataset has better stationarity.

[0088] It should be noted that differential processing is a technique widely used in quantitative finance, signal analysis, and data comparison. Its core is to extract useful information or eliminate noise by calculating the differences between adjacent data points. The differential methods in quantitative finance mainly include: First-order differencing: calculates daily fluctuations, i.e., today's value minus yesterday's value, suitable for short-term volatility analysis; Second-order differencing: reflects the rate of change and can identify short-term reversal signals; Distributed differencing, i.e., periodic differencing: simultaneously calculates the differences at different time intervals to capture trends across multiple time scales; Seasonal differencing: for periodic data, it eliminates the influence of cycles through fixed-interval differencing; Logarithmic differencing: first takes the logarithm and then subtracts, transforming multiplicative growth into additive growth, which is closer to a normal distribution and suitable for calculating the returns of financial assets.

[0089] In some embodiments of this application, the electronic device performs differential processing on the initial customer migration parameters in the aforementioned initial time series dataset to obtain a time series dataset, including: the electronic device deletes the initial customer migration parameters with poor stationarity in the aforementioned initial time series dataset, and arranges the remaining initial customer migration parameters in chronological order to obtain the aforementioned time series dataset.

[0090] For specific methods of differential processing, please refer to the detailed descriptions in relevant technologies. To avoid repetition, they will not be elaborated here.

[0091] Step 101c2: If the test result is less than the first threshold, the electronic device determines the initial time series dataset as the time series dataset.

[0092] In some embodiments of this application, when the above test result is less than the first threshold, it can be considered that the initial time series dataset has good stationarity; at this time, the initial time series dataset can be determined as the above time series dataset, that is, the SARIMA model can be constructed directly based on the initial time series dataset.

[0093] In some embodiments of this application, since the initial time series dataset can be differentially processed based on the relationship between the test result of the stationarity test and the first threshold to obtain the time series dataset; or the initial time series dataset can be determined as the time series dataset, the time series dataset with better stationarity can be flexibly determined based on whether the initial time series dataset is stationary.

[0094] Step 102: Based on the customer migration parameters in the time series dataset, the electronic device determines the seasonal parameters and model parameters.

[0095] Among them, the seasonal parameters mentioned above include P, D, Q and s, and the model parameters mentioned above include p, d and q. P is the seasonal autoregression order, D is the seasonal difference number, Q is the seasonal moving average order, s is the number of the first unit of time within a predetermined unit of time, p is the non-seasonal autoregression order, d is the non-seasonal difference number, and q is the non-seasonal moving average order.

[0096] In some embodiments of this application, the above-mentioned model parameters may also be referred to as non-seasonal parameters.

[0097] In some embodiments of this application, the aforementioned seasonal parameters and model parameters can be parameters of the SARIMA model.

[0098] In some embodiments of this application, P is used to characterize the degree to which the current observation is affected by the previous P seasonally lagged observations.

[0099] In some embodiments of this application, D represents the number of times seasonal differencing is performed on the aforementioned time series dataset to eliminate seasonal non-stationarity in the data, such as periodic fluctuations.

[0100] In some embodiments of this application, Q is used to measure the degree to which the current value is affected by past seasonal error terms.

[0101] For example, when Q=1, the model will consider the impact of the prediction error from the previous seasonal period (e.g., 12 months ago) on the current value.

[0102] In some embodiments of this application, s is the number of first unit durations within a predetermined unit duration; wherein, the predetermined unit duration can be any possible unit duration such as 1 year, 1 quarter or 1 month; the predetermined unit duration is longer than the first unit duration.

[0103] For example, assuming the above-mentioned predetermined unit of time is 1 year and the first unit of time is 1 month, then s is the number of months in 1 year, which is 12; assuming the above-mentioned predetermined unit of time is 1 quarter and the first unit of time is 1 month, then s is the number of months in 1 quarter, which is 3.

[0104] In some embodiments of this application, p is used to capture the autocorrelation of non-seasonal components in the aforementioned time series dataset.

[0105] In some embodiments of this application, d represents the number of times the above-mentioned time series dataset is subjected to non-seasonal differencing, in order to eliminate trend non-stationarity in the data, such as long-term growth or decline trends.

[0106] In some embodiments of this application, q is used to measure the degree to which the current value is affected by past non-seasonal error terms.

[0107] For example, when q=1, the model will consider the impact of the prediction error in the previous period (such as time t-1) on the current value.

[0108] In some embodiments of this application, electronic devices can determine the aforementioned seasonal parameters and model parameters by analyzing the customer migration parameters in the aforementioned time series dataset.

[0109] The specific methods for determining the aforementioned seasonal parameters and model parameters for electronic devices are explained in detail below.

[0110] In some embodiments of this application, combined with Figure 1 ,like Figure 4 As shown, step 102 can be implemented through steps 102a to 102d below.

[0111] Step 102a: The electronic device substitutes the customer migration parameters from the time series dataset into the autocorrelation function to calculate Q and q.

[0112] In some embodiments of this application, the value of Q is equal to the value of q.

[0113] In some embodiments of this application, the electronic device substitutes the customer migration parameters from the aforementioned time series dataset into the autocorrelation function for calculation, and can output the values ​​of Q and q, thereby obtaining Q and q.

[0114] In some embodiments of this application, the electronic device may first plot a seasonal ACF graph and determine the value of Q by observing the truncation of the ACF at seasonal lag points (such as 12, 24, 36, etc.). If the ACF is truncated after lag points 12, 24, 36, etc., then Q is the order corresponding to the truncation point.

[0115] In some embodiments of this application, the electronic device can also compare the fitting effect of different Q-value models using model evaluation metrics such as the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC) to select the optimal value of Q.

[0116] In some embodiments of this application, the value of Q ranges from 1 to 3, and an excessively high order may lead to overfitting.

[0117] For example, for monthly data (s=12), if the ACF is significant at lags of 12 and 24, then Q=2.

[0118] In some embodiments of this application, in Python, the value of Q can be fitted using the Seasonal Autoregressive Integrated Moving Average with exogenous regressors (SARIMAX) model from the statsmodels library.

[0119] In some embodiments of this application, the electronic device may first draw an ACF diagram. If the ACF is truncated after a lag of q, i.e., the subsequent coefficients are not significant, then q is the truncation point. For example, if the ACF is truncated after a lag of 2, then q = 2.

[0120] In some embodiments of this application, the electronic device may also select the q value that minimizes AIC or BIC through grid search.

[0121] In some embodiments of this application, the value of q ranges from 1 to 3, and an excessively high order may lead to overfitting.

[0122] For example, if the ACF is truncated after a lag of 1, then q = 1.

[0123] In some embodiments of this application, in Python, the value of q can be set using the ARIMA model from the statsmodels library for fitting.

[0124] Step 102b: The electronic device substitutes the customer migration parameters from the time series dataset into the partial autocorrelation function to calculate P and p.

[0125] In some embodiments of this application, the value of P is equal to the value of p.

[0126] In some embodiments of this application, the electronic device substitutes the customer migration parameters in the aforementioned time series dataset into the partial autocorrelation function for calculation, and can output the values ​​of P and p, thereby obtaining P and p.

[0127] In some embodiments of this application, the electronic device may first draw a seasonal PACF diagram. If the SPACF is truncated after a lag of P×s (e.g., when s=12, it is truncated after lags of 12 or 24), then P is the truncation order.

[0128] In some embodiments of this application, the electronic device may also select the P value that minimizes AIC or BIC through grid search. In some embodiments of this application, the value of P ranges from 1 to 2; an excessively high order may lead to overfitting.

[0129] For example, if the monthly data is truncated by SPACF after a 12-order lag, then P=1.

[0130] In some embodiments of this application, the electronic device may first draw a PACF plot, and if the PACF is truncated after a lag of p (subsequent coefficients are not significant), then p is the truncation point. For example, if the PACF is truncated after a lag of 2, then p=2.

[0131] In some embodiments of this application, the electronic device may also select the p-value that minimizes AIC or BIC through grid search. In some embodiments of this application, the value of p ranges from 1 to 3; excessively high orders may lead to overfitting.

[0132] Step 102c: The electronic device determines the number of differential processing steps corresponding to the customer migration parameters in the time series dataset as D and d.

[0133] In some embodiments of this application, the value of D is equal to the value of d.

[0134] In some embodiments of this application, D and d can both be 0 or 1, and at most 2.

[0135] Step 102d: The electronic device determines the number of the first unit of time in a year as s.

[0136] For example, if the first unit of time is 1 month, then s is the number of months in 1 year, which is 12; if the first unit of time is 1 quarter, then s is the number of quarters in 1 year, which is 4.

[0137] In some embodiments of this application, since Q and q can be determined by the autocorrelation function, P and p can be determined by the partial autocorrelation function, D and d can be determined by the difference times corresponding to the customer migration parameters, and s can be determined as the number of the first unit duration in one year; therefore, the model parameters and seasonal parameters can be determined in a corresponding manner according to the characteristics of each parameter, so as to ensure that the model parameters and seasonal parameters can be determined quickly and accurately, thereby improving the accuracy of constructing the SARIMA prediction model.

[0138] Step 103: The electronic device uses seasonal parameters and model parameters to initialize the model and build a SARIMA model.

[0139] In some embodiments of this application, electronic devices can use the statsmodels library in Python to initialize the model using the aforementioned seasonal parameters and model parameters in order to construct a SARIMA model.

[0140] For example, if P, p, D, d, Q, and q are all 1 and s is 12, then a SARIMA model can be constructed using the parameters (1,1,1) and (1,1,1,12) through the statsmodels library in Python.

[0141] Step 104: The electronic device uses the time series dataset, seasonal parameters, and model parameters to fit the SARIMA model to obtain the prediction model.

[0142] The aforementioned prediction model is used to predict the number of customers migrating down the city due to seasonal variations.

[0143] In some embodiments of this application, after constructing the SARIMA model, the SARIMA model can be fitted using the aforementioned time series dataset, seasonal parameters, and model parameters via Python to obtain the aforementioned prediction model.

[0144] In some embodiments of this application, after the above-mentioned prediction model is obtained by fitting, the model summary can be printed using Python.

[0145] In some embodiments of this application, after obtaining the above-mentioned prediction model, the number of customers migrating down within a certain historical time period can be input into the prediction model, and the prediction model can output the predicted number of customers migrating down within a future time period.

[0146] In some embodiments of this application, the prediction model acquisition method is provided by fitting a SARIMA prediction model constructed using the aforementioned model parameters and seasonal parameters. Therefore, this prediction model allows for the introduction of seasonal components when processing time-series data, thereby better identifying and capturing periodic fluctuations in customer AUM. Effectively modeling these seasonal characteristics makes the prediction of customer migration data volume more accurate.

[0147] In some embodiments of this application, combined with Figure 1 ,like Figure 5 As shown, after step 104 above, some embodiments of this application provide a prediction model acquisition method that may further include step 105 below.

[0148] Step 105: The electronic device processes the number of customers migrating down in the first historical time period through the prediction model to obtain the predicted number of customers migrating down in the second time period.

[0149] The second time period is a preset time period following the current moment.

[0150] In some embodiments of this application, the electronic device can first input the number of customers migrating down within a first historical time period into the above-mentioned prediction model, and then process the number of customers migrating down within the first historical time period through the prediction model to obtain the predicted number of customers migrating down within a second time period.

[0151] In some embodiments of this application, the first historical time period can be any time period prior to the current moment.

[0152] For example, the first historical time period can be 12 months, 24 months, or 36 months prior to the current moment.

[0153] In some embodiments of this application, the aforementioned preset time period can be any preset time period; for example, it can be 6 months or 12 months, etc.

[0154] In some embodiments of this application, when the electronic device inputs the number of customers migrating down within the first historical time period into the prediction model, it can also input the second time period as a prediction parameter into the prediction model, so that the prediction model can output the predicted number of customers migrating down within the second time period.

[0155] In some embodiments of this application, after obtaining the predicted number of customers migrating to the second time period, the electronic device can visualize the prediction results using Python. At this time, the predicted number of customers migrating to the second time period will be plotted in a chart to intuitively display the predicted value of the number of customers migrating to the second time period.

[0156] In some embodiments of this application, since the number of customers migrating to the next period can be obtained by processing the number of customers migrating to the next period through the prediction model after obtaining the prediction model, the number of customers migrating to the next period can be predicted by the prediction model, thereby making the prediction of the amount of customer migration data more accurate and convenient.

[0157] In some embodiments of this application, combined with Figure 5 ,like Figure 6 As shown, after step 105 above, some embodiments of this application provide a prediction model acquisition method that may further include steps 106 and 107 below.

[0158] Step 106: The electronic device performs residual analysis on the predicted number of customers migrating to the second time period and the actual number of customers migrating to the second time period.

[0159] In some embodiments of this application, the electronic device can obtain the residual between the predicted number of customers migrating to the second time period and the actual number of customers migrating to the second time period in Python through the resid attribute of the prediction model described above; then, the obtained residual is analyzed using a residual test method, which may include at least one of the following: ACF test, Ljung-Box test, and normality test.

[0160] Step 107: If the analysis result of the residual analysis is greater than or equal to the second threshold, and the preset model evaluation index is greater than or equal to the third threshold, the electronic device reconstructs the SARIMA model.

[0161] In some embodiments of this application, the electronic device reconstructs the SARIMA model, including: the electronic device adjusting the model parameters of the aforementioned SARIMA model, the electronic device redetermining the seasonal parameters and model parameters, and reconstructing a new SARIMA model.

[0162] In some embodiments of this application, when the analysis result of the residual analysis is greater than or equal to the second threshold and the preset model evaluation index is greater than or equal to the third threshold, the obtained residual can be considered unsatisfactory. At this time, the model parameters can be readjusted or other models can be considered.

[0163] In some embodiments of this application, the analysis results of the above residual analysis may include all lag orders, or p-values.

[0164] In some embodiments of this application, the above-mentioned preset model evaluation index can be AIC or BIC.

[0165] In some embodiments of this application, the second threshold and the third threshold can be system defaults or can be arbitrarily set by the user according to actual usage needs.

[0166] In some embodiments of this application, if the analysis result of the above residual analysis is less than the second threshold and the above preset model evaluation index is less than the third threshold, then the obtained residual can be considered as an ideal residual, that is, it presents a white noise distribution and has no significant autocorrelation; at this time, the above prediction model can continue to be used.

[0167] In some embodiments of this application, after predicting the number of customers migrating to lower-level cities within a future time period, the prediction model can be evaluated by residual analysis of the predicted and actual values, as well as by model evaluation metrics. If the residual analysis result is greater than or equal to a second threshold and the model evaluation metrics are greater than or equal to a third threshold, the SARIMA model can be reconstructed, thereby ensuring that the SARIMA model used always has good performance.

[0168] The following description, in conjunction with the accompanying drawings, exemplarily illustrates some embodiments of the prediction model acquisition method provided in this application.

[0169] For example, taking a first time period of 36 months starting from 2020, and a first unit of time of 1 month, after environmental preparation, such as Figure 7As shown, some embodiments of this application provide a predictive model acquisition method that may include the processes shown in steps one through six below. The environment preparation includes: preparing electronic devices and installing relevant Python libraries. These relevant Python libraries include: pip, install, numpy, pandas, matplotlib, and statsmodels. The numpy and pandas libraries are primarily used for data processing and analysis; the matplotlib library can be used for data visualization; and the statsmodels library provides time series analysis tools, including the SARIMA model.

[0170] I. Data Preparation The electronic device retrieves a 36-month time-series dataset of customer relocation numbers from branch A starting in 2020. This dataset includes the number of customers relocating each month and the corresponding Assets Under Management (AUM). The monthly AUM and the corresponding number of customers relocating are listed in the AUM and Downgrade columns. A data structure is then created.

[0171] II. Data Visualization and Exploration Electronic devices can provide simple visualizations of data to help understand overall trends and seasonal characteristics. For example, they can plot time series graphs of customer migration numbers, making it easy to observe the overall trend and seasonal fluctuations in customer migration.

[0172] III. Stationarity Test and Difference Processing The SARIMA model requires that the input data be stationary. To this end, the electronic device uses the ADF test to check the stationarity of the time series dataset. If the time series dataset is not stationary, it is differencing. The differrated time series dataset is then plotted.

[0173] Specifically, if the p-value of the ADF test is greater than 0.05, it indicates that the time series dataset is not stationary, and differencing is required. Differencing the time series dataset will eliminate the trend in the data, making it more stationary.

[0174] IV. SARIMA Model Parameter Selection The SARIMA model parameters include model parameters (p, d, q) and seasonality parameters (P, D, Q, s). Electronic devices can generate ACF and PACF plots, where the ACF plot determines q, the PACF plot determines p, and d can be determined by the difference order. The seasonality parameters (P, D, Q, s) are determined through analysis of the periodic fluctuations in the data; for example, for monthly data, the periodicity s can be set to 12.

[0175] V. Constructing a SARIMA model and making predictions Based on the ACF and PACF plots, and the analysis of seasonal fluctuations, appropriate (p, d, q) and (P, D, Q, s) parameters are selected. For example, assuming the determined parameters are SARIMA(1, 1, 1)(1, 1, 1, 12), a SARIMA model can be constructed and fitted, and the model summary can be printed. After the model fitting is complete, predictions can be made using the obtained prediction model, and the prediction results can be visualized. The prediction results will then be plotted in a chart, showing the predicted number of customers migrating to lower-tier cities within a future time period.

[0176] VI. Model Evaluation To validate the model's effectiveness, residual analysis and several model evaluation metrics can be used for assessment. These include residual diagnostics and printing AIC and BIC. Ideally, the residuals should exhibit a white noise distribution, i.e., without significant autocorrelation.

[0177] If the residual analysis result is greater than or equal to the second threshold, and the preset model evaluation index is greater than or equal to the third threshold, then data can be acquired again, parameters determined, and the SARIMA model reconstructed and fitted. If the residual analysis result is less than the second threshold, and the preset model evaluation index is less than the third threshold, then the current prediction model can continue to be used.

[0178] Thus, based on the SARIMA model, the number of customers migrating to lower-tier banks can be accurately predicted by analyzing trends and seasonal characteristics in historical data. The prediction model acquisition method provided in some embodiments of this application has the following significant advantages: 1. Accurately capture seasonal fluctuations The aforementioned forecasting model incorporates seasonality when processing time-series data, thereby better identifying and capturing cyclical fluctuations in customer AUM. This is particularly important in banking, as customer asset liquidity is often affected by cyclical factors such as quarterly and annual cycles. Effectively modeling these seasonal characteristics makes customer migration forecasts more accurate and flexible.

[0179] 2. Highly adaptable, capable of handling different types of time series. The aforementioned forecasting model is adaptable to different types of time series data and possesses excellent scalability. Whether calculating customer management data monthly, quarterly, or annually, the model can be flexibly applied to different scenarios by adjusting seasonal periodic parameters. Furthermore, the model performs exceptionally well in handling long-term trends and short-term fluctuations, effectively modeling and forecasting different fluctuation patterns in customer AUM.

[0180] 3. Improve prediction accuracy By combining seasonal autoregression, seasonal differencing, and seasonal moving average structures, the aforementioned forecasting model can more accurately reflect future trends when predicting the number of bank branch customers migrating to lower-tier cities. Compared to traditional ARIMA models or management methods based on simple rules, this forecasting model can significantly improve forecast accuracy, thereby reducing errors in migration prediction and enhancing managers' foresight regarding future changes.

[0181] 4. Enhanced early warning function supports management decision-making. Bank branches can anticipate potential customer migration trends and take corresponding management measures in advance. For example, if the aforementioned predictive model foreshadows a decline in the assets under management (AUM) of certain customers in the coming months, branch managers can intervene early and take measures to retain and enhance customer relationships. This early warning mechanism helps bank branches manage customer resources more proactively and meticulously, reducing the risks of reactive management.

[0182] 5. High efficiency and easy to implement The aforementioned predictive model can be implemented using widely used open-source Python components. Its simple structure makes it easy to implement and maintain. The predictive model can be quickly deployed through preprocessing of time series data and model fitting. This efficiency and ease of use enable rapid deployment in banking practices, further optimizing branch operational efficiency.

[0183] The above-described method embodiments, or various possible implementations of the method embodiments, can be executed individually, or, provided there are no contradictions, they can be combined with each other. The specific implementation can be determined according to actual usage requirements, and this application embodiment does not impose any restrictions on this.

[0184] Figure 8 A schematic diagram of the structure of a prediction model acquisition device provided for some embodiments of this application. For example... Figure 8 As shown, the prediction model acquisition device 80 includes: an acquisition module 81, a determination module 82, a construction module 83, and a fitting module 84.

[0185] The acquisition module 81 can be used to acquire the time series dataset corresponding to the number of customers migrating down within a first time period. This time series dataset includes customer migration parameters for each first unit of time within the first time period, including the number of customers migrating down within a first unit of time and the corresponding asset management scale information. The determination module 82 can be used to determine seasonal parameters and model parameters based on the customer migration parameters in the time series dataset acquired by the acquisition module 81. The seasonal parameters include P, D, Q, and s, and the model parameters include p, d, and q. P is the seasonal autoregression order, D is the seasonal differencing frequency, Q is the seasonal moving average order, s is the number of first units of time within a predetermined unit of time, p is the non-seasonal autoregression order, d is the non-seasonal differencing frequency, and q is the non-seasonal moving average order. The construction module 83 can be used to initialize the model and construct the SARIMA model using the seasonal parameters and model parameters determined by the determination module 82. The fitting module 84 can be used to fit the SARIMA model constructed by the building module 83 using the time series dataset, the seasonal parameters, and the model parameters to obtain a prediction model, which is used to predict the number of customers migrating down the line with seasonal variations.

[0186] In other embodiments, the acquisition module 81 may be specifically used to acquire an initial time series dataset corresponding to the number of customers migrating down within a first time period. The initial time series dataset includes initial customer migration parameters corresponding to each first unit duration within the first time period. The initial customer migration parameters in the initial time series dataset are then subjected to a stationarity test. Based on the results of the stationarity test, the aforementioned time series dataset is determined.

[0187] In other embodiments, the acquisition module 81 may be specifically used to perform differential processing on the initial customer migration parameters in the initial time series dataset to obtain the time series dataset when the test result is greater than or equal to the first threshold; or, when the test result is less than the first threshold, to determine the initial time series dataset as the time series dataset.

[0188] In other embodiments, the determining module 82 can be specifically used to substitute the customer migration parameters in the above time series dataset into the autocorrelation function to calculate Q and q; and to substitute the customer migration parameters in the above time series dataset into the partial autocorrelation function to calculate P and p; and to determine the number of differential processing times corresponding to the customer migration parameters in the above time series dataset as D and d; and to determine the number of the first unit duration in 1 year as s.

[0189] In other embodiments, the prediction model acquisition device 80 may further include a processing module. The processing module may be used to, after the fitting module 84 uses the aforementioned time series dataset, the aforementioned seasonal parameters, and the aforementioned model parameters to fit the aforementioned SARIMA model to obtain a prediction model, process the number of customers migrating down within a first historical time period using the prediction model to obtain the predicted number of customers migrating down within a second time period, where the second time period is a preset time period following the current moment.

[0190] In other embodiments, the above-mentioned processing module can also be used to process the number of customers migrating to the first historical time period through the above-mentioned prediction model to obtain the number of customers migrating to the second time period, and then perform residual analysis on the predicted number of customers migrating to the second time period and the actual number of customers migrating to the second time period; and reconstruct the SARIMA model if the analysis result of the residual analysis is greater than or equal to the second threshold and the preset model evaluation index is greater than or equal to the third threshold.

[0191] The predictive model acquisition device provided in this application embodiment can execute the method shown in the above method embodiment. Its implementation principle and beneficial effects can be referred to the relevant description in the method embodiment, and will not be repeated here.

[0192] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 9 As shown, the electronic device 50 includes at least one processor 501 and a memory 502.

[0193] Processor 501 can be a general-purpose processor, including a Central Processing Unit (CPU), a network processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Processor 501 can also be other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0194] The memory 502 is used to store computer program code, which includes computer instructions. These computer instructions run in the aforementioned electronic device to implement the method shown in the above-described method embodiments. For example, the memory may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, portable hard drive, read-only memory, disk, or optical disc, etc.

[0195] Processor 501 and memory 502 are communicatively connected. For example, memory 502 can be connected to processor 501 via a system bus and communicate with it. The system bus can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, an industry standard architecture (ISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus.

[0196] Optionally, the memory 502 can be either standalone or integrated with the processor 501. When the memory 502 is set up independently, it is connected to the processor 501 via the system bus.

[0197] This application also provides a chip for executing instructions, which is used to execute the technical solution of the prediction model acquisition method in the above embodiments.

[0198] This application also provides a computer-readable storage medium storing computer instructions. When these computer instructions are executed by a processor, they are used to implement the technical solution of the prediction model acquisition method described in the above embodiments. Specifically, when the computer instructions are executed by a processor, the electronic device can perform the technical solution of the prediction model acquisition method described in the above embodiments.

[0199] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the prediction model acquisition method in the above embodiments. The aforementioned computer-readable storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage media can be any available medium accessible to a general-purpose or special-purpose computer. An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic control unit or main control device; this application does not limit this.

[0200] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0201] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0202] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0203] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0204] It should be understood that the steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

[0205] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0206] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for obtaining a prediction model, characterized in that, The method comprises the following steps: obtaining a time series data set corresponding to the number of customer downgrades in a first time period, wherein the time series data set comprises a customer downgrade parameter corresponding to each first unit time length in the first time period, and the customer downgrade parameter comprises the number of customer downgrades in a first unit time length and asset management scale information corresponding to the number of customer downgrades; determining seasonal parameters and model parameters based on the customer downgrade parameters in the time series data set, wherein the seasonal parameters comprise P, D, Q and s, the model parameters comprise p, d and q, P is the order of seasonal autoregression, D is the order of seasonal difference, Q is the order of seasonal moving average, s is the number of the first unit time length in a predetermined unit time length, p is the order of non-seasonal autoregression, d is the order of non-seasonal difference, and q is the order of non-seasonal moving average; using the seasonal parameters and the model parameters to initialize the model and construct a seasonal autoregressive and moving average (SARIMA) model; fitting the SARIMA model using the time series data set, the seasonal parameters and the model parameters to obtain a prediction model, wherein the prediction model is used to predict the number of customer downgrades with seasonal variation.

2. The method of claim 1, wherein, The method comprises the following steps: obtaining an initial time series data set corresponding to the number of customer downgrades in the first time period, wherein the initial time series data set comprises an initial customer downgrade parameter corresponding to each first unit time length in the first time period; performing stationarity test on the initial customer downgrade parameters in the initial time series data set; determining the time series data set according to the test result of the stationarity test.

3. The method of claim 2, wherein, The method comprises the following steps: in the case where the test result is greater than or equal to a first threshold, performing difference processing on the initial customer downgrade parameters in the initial time series data set to obtain the time series data set; or, in the case where the test result is less than the first threshold, determining the initial time series data set as the time series data set.

4. The method of claim 1, wherein, The method comprises the following steps: substituting the customer downgrade parameters in the time series data set into an autocorrelation function to obtain Q and q; substituting the customer downgrade parameters in the time series data set into a partial autocorrelation function to obtain P and p; determining the difference processing times corresponding to the customer downgrade parameters in the time series data set as D and d; determining the number of the first unit time length in a year as s.

5. The method according to any one of claims 1 to 4, characterized in that, After the prediction model is obtained by fitting the SARIMA model using the time series data set, the seasonal parameters and the model parameters, the method further comprises the following steps: processing the number of customer downgrades in a first historical time period by using the prediction model to obtain the predicted number of customer downgrades in a second time period, wherein the second time period is a preset time period after the current time.

6. The method of claim 5, wherein, After the processing of the number of customer downgrades in the first historical time period by the prediction model to obtain the number of customer downgrades in the second time period, the method further comprises: performing residual analysis on the predicted number of customer downgrades in the second time period and the actual number of customer downgrades in the second time period; in the case that the analysis result of the residual analysis is greater than or equal to a second threshold value, and a preset model evaluation index is greater than or equal to a third threshold value, reconstructing the SARIMA model. 7.A prediction model acquisition apparatus characterized by comprising: comprise: an acquisition module, configured to acquire a time series data set corresponding to the number of customer downgrades in a first time period, wherein the time series data set comprises a customer downgrade parameter corresponding to each first unit time length in the first time period, and the customer downgrade parameter comprises the number of customer downgrades in a first unit time length and asset management scale information corresponding to the number of customer downgrades; a determination module, configured to determine a seasonal parameter and a model parameter based on the customer downgrade parameter in the time series data set, wherein the seasonal parameter comprises P, D, Q and s, the model parameter comprises p, d and q, P is a seasonal autoregressive order, D is a seasonal difference number, Q is a seasonal moving average order, s is the number of the first unit time length in a predetermined unit time length, p is a non-seasonal autoregressive order, d is a non-seasonal difference number, and q is a non-seasonal moving average order; a construction module, configured to perform model initialization using the seasonal parameter and the model parameter, and construct a seasonal autoregressive and moving average SARIMA model; a fitting module, configured to fit the SARIMA model using the time series data set, the seasonal parameter and the model parameter to obtain a prediction model, wherein the prediction model is used to predict the number of customer downgrades with seasonal variation.

8. The apparatus of claim 7, wherein, The acquisition module is specifically configured to acquire an initial time series data set corresponding to the number of customer downgrades in the first time period, and the initial time series data set comprises an initial customer downgrade parameter corresponding to each first unit time length in the first time period; and perform stationarity test on the initial customer downgrade parameter in the initial time series data set. The acquisition module is specifically configured to acquire an initial time series data set corresponding to the number of customer downgrades in the first time period, and the initial time series data set comprises an initial customer downgrade parameter corresponding to each first unit time length in the first time period; and perform stationarity test on the initial customer downgrade parameter in the initial time series data set.

9. The apparatus of claim 8, wherein, The determination module is specifically configured to substitute the customer downgrade parameter in the time series data set into an autocorrelation function to obtain Q and q; substitute the customer downgrade parameter in the time series data set into a partial autocorrelation function to obtain P and p; determine D and d as the difference processing number corresponding to the customer downgrade parameter in the time series data set; and determine s as the number of the first unit time length in one year.

10. The apparatus of claim 7, wherein, The prediction model acquisition device further comprises:

11. The apparatus of any one of claims 7 to 10, wherein, ​ The processing module is configured to process, by using the prediction model, the number of customer downgrades in the first historical time period to obtain the predicted number of customer downgrades in the second time period after the fitting module uses the time series data set, the seasonal parameter and the model parameter to fit the SARIMA model to obtain the prediction model.

12. The apparatus of claim 11, wherein, The processing module is further configured to perform residual analysis on the predicted number of customer downgrades in the second time period and the actual number of customer downgrades in the second time period after processing, by using the prediction model, the number of customer downgrades in the first historical time period to obtain the number of customer downgrades in the second time period; and reconstruct the SARIMA model when the analysis result of the residual analysis is greater than or equal to a second threshold value and a preset model evaluation index is greater than or equal to a third threshold value.

13. An electronic device, comprising: The electronic device comprises: a memory and at least one processor; the memory is in communication connection with the processor; the memory is configured to store computer program code, the computer program code comprises computer instructions; when the processor executes the computer instructions, the electronic device executes the method in any one of claims 1-6.

14. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are executed by the processor to implement the method in any one of claims 1-6.

15. A computer program product, characterised in that, When the computer program product is running on the computer / being executed by the processor of the computer, the method in any one of claims 1-6 is implemented.

Images