Business data prediction methods, devices, equipment, media, and program products based on multi-model fusion

Abstract

This application provides a business data prediction method based on multi-model fusion, applicable to the fields of artificial intelligence, big data, and fintech. The method includes: predicting incremental data for a target year using a Bayesian statistical model based on multi-source business data; obtaining incremental data for a target quarter using a seasonal time-series prediction model based on the multi-source business data; performing long-term decomposition prediction using the seasonal time-series prediction model and a multinomial model based on the incremental data of the target quarter to obtain initial prediction data for the target year and fluctuation data for the remaining quarters; wherein the target quarter and the remaining quarters constitute the target year; and using the incremental data of the target year as trend reference data, and based on the fluctuation data of the remaining quarters, correcting the initial prediction data for the target year to obtain the business data prediction result. This application also provides a business data prediction apparatus, device, storage medium, and program product based on multi-model fusion.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically to the fields of big data and fintech, and more specifically to a business data prediction method, apparatus, device, medium and program product based on multi-model fusion. Background Technology

[0002] In the financial industry, forecasting business data such as outstanding balances, delinquency rates, and prepayment rates is crucial. It helps financial institutions manage credit risk, identify potential risks in advance (e.g., a rising outstanding balance may indicate increased default risk), and provides a basis for business decisions, helping to rationally plan the scale and structure of lending and improve capital utilization efficiency. In large-scale information systems and distributed computing environments, performing high-precision, high-frequency forecasting of massive amounts of multi-source time-series data is a core and computationally intensive task.

[0003] However, existing forecasting models have drawbacks. Traditional models, such as time series analysis, rely heavily on historical data. Traditional methods depend on a single model and cannot simultaneously capture long-term trends, seasonal fluctuations, and long-term anchoring. They are unable to accurately reflect the complex and ever-changing reality, and the forecast results may be out of touch with the actual business situation. The large deviation between the forecast results and the actual situation results in low accuracy of business data forecast results. The forecast deviation will be amplified in automated decision-making systems, which may lead to the incorrect reservation or over-allocation of computing resources (such as memory and bandwidth), or even trigger incorrect automatic scaling strategies, thereby directly affecting the stability, service availability, and operating costs of the entire system. Summary of the Invention

[0004] In view of the above problems, this application provides a business data prediction method, apparatus, device, medium and program product based on multi-model fusion to improve computational efficiency.

[0005] According to a first aspect of this application, a business data prediction method based on multi-model fusion is provided. The method includes: predicting incremental data for a target year using a Bayesian statistical model based on multi-source business data; wherein the Bayesian statistical model is trained and updated online based on a historical business dataset; obtaining incremental data for a target quarter using a seasonal time-series prediction model based on the multi-source business data; wherein the seasonal time-series prediction model is trained and updated online based on the historical business dataset; obtaining initial prediction data for the target year and fluctuation data for the remaining quarters by performing long-term decomposition prediction using the seasonal time-series prediction model and a multinomial model based on the incremental data of the target quarter; wherein the target quarter and the remaining quarters constitute the target year; and correcting the initial prediction data for the target year based on the fluctuation data of the remaining quarters using the incremental data of the target year as trend reference data to obtain a business data prediction result.

[0006] According to an embodiment of this application, the step of obtaining initial forecast data for the target year and fluctuation data for the remaining quarters by performing long-term decomposition forecasting using the seasonal time-series forecasting model and a multinomial model based on the incremental data of the target quarter includes: adjusting the incremental data of the target quarter by quarterly proportion to determine the quarterly incremental percentage; performing trend fitting using the multinomial model based on the quarterly incremental percentage to generate an adjusted fitting value for the target year, and determining the initial forecast data for the target year based on the adjusted fitting value for the target year; performing rolling forecasting using the seasonal time-series forecasting model based on the multi-source business data to output the original forecast value for the target year; wherein the original forecast value for the target year includes the original forecast values ​​for the remaining quarters; and performing fluctuation migration on the original forecast values ​​for the remaining quarters using the multinomial model to obtain the fluctuation data for the remaining quarters.

[0007] According to an embodiment of this application, adjusting the incremental data of the target quarter to determine the quarterly incremental percentage includes: determining the business increment of the remaining quarters based on the incremental data of the target quarter; and allocating the business increment of the remaining quarters based on a preset quarterly incremental percentage to obtain the quarterly incremental percentage.

[0008] According to an embodiment of this application, the step of using the polynomial model to perform fluctuation migration on the original forecast values ​​of the remaining quarters to obtain the fluctuation data of the remaining quarters includes: fitting the original forecast values ​​of the remaining quarters to the polynomial model to obtain the predicted fitted values ​​of the remaining quarters; and calculating the fluctuation data of the remaining quarters based on the original forecast values ​​of the remaining quarters and the predicted fitted values ​​of the remaining quarters.

[0009] According to an embodiment of this application, the method further includes: updating the historical business dataset online based on the multi-source business data; and training and updating the Bayesian statistical model and the seasonal time series prediction model based on the updated historical business dataset.

[0010] According to an embodiment of this application, the method further includes: obtaining first business data through a secure interface based on m business data systems; where m is a positive integer; obtaining industry data through a public interface based on publicly available industry data sources; and obtaining the multi-source business data based on the industry data and the first business data.

[0011] According to an embodiment of this application, obtaining the first business data through a secure interface based on m business data systems includes: reading the second business data of the m business data systems through the secure interface based on a dynamic token authentication method; wherein the second business data corresponding to sensitive fields in the m business data systems is pre-encrypted using an encryption algorithm; transmitting the second business data back through the secure interface according to a standard encryption protocol; and obtaining the first business data based on the second business data, using the decryption algorithm corresponding to the encryption algorithm and the decryption rules corresponding to the standard encryption protocol.

[0012] According to an embodiment of this application, the method further includes: issuing an early warning for abnormal data; wherein the abnormal data is business data in the multi-source business data that exceeds the confidence interval boundary; and performing multi-dimensional drill-down analysis on the multi-source business data to locate the abnormal business unit corresponding to the abnormal data.

[0013] A second aspect of this application provides a business data prediction device based on multi-model fusion, comprising: a statistical prediction module for predicting incremental data for a target year using a Bayesian statistical model based on multi-source business data; wherein the Bayesian statistical model is obtained through online updating and training based on a historical business dataset; a quarterly prediction module for obtaining incremental data for a target quarter using a seasonal time-series prediction model based on the multi-source business data; wherein the seasonal time-series prediction model is obtained through online updating and training based on the historical business dataset; a period decomposition module for performing long-period decomposition prediction based on the incremental data of the target quarter using the seasonal time-series prediction model and a multinomial model to obtain initial prediction data for the target year and fluctuation data for the remaining quarters; wherein the target quarter and the remaining quarters constitute the target year; and a result correction module for correcting the initial prediction data for the target year based on the fluctuation data of the remaining quarters, using the incremental data of the target year as trend reference data, to obtain a business data prediction result.

[0014] According to an embodiment of this application, the periodic decomposition module includes: a proportion adjustment unit, used to adjust the incremental data of the target quarter according to a quarterly proportion to determine the quarterly incremental percentage; a trend fitting unit, used to perform trend fitting based on the quarterly incremental percentage using the polynomial model to generate an adjusted fitting value for the target year, and to determine the initial forecast data for the target year based on the adjusted fitting value for the target year; a rolling forecast unit, used to perform rolling forecasting based on the multi-source business data using the seasonal time series forecasting model to output the original forecast value for the target year; wherein the original forecast value for the target year includes the original forecast values ​​for the remaining quarters; and a fluctuation migration unit, used to perform fluctuation migration on the original forecast values ​​for the remaining quarters using the polynomial model to obtain the fluctuation data for the remaining quarters.

[0015] According to an embodiment of this application, the ratio adjustment unit includes: a first ratio adjustment subunit, used to determine the business increment of the remaining quarters based on the incremental data of the target quarter; and a second ratio adjustment subunit, used to allocate the business increment of the remaining quarters based on a preset quarterly increment ratio to obtain the quarterly increment percentage.

[0016] According to an embodiment of this application, the fluctuation migration unit includes: a first fluctuation migration subunit, configured to fit the original predicted values ​​of the remaining quarters to the polynomial model to obtain the predicted fitted values ​​of the remaining quarters; and a second fluctuation migration subunit, configured to calculate the fluctuation data of the remaining quarters based on the original predicted values ​​of the remaining quarters and the predicted fitted values ​​of the remaining quarters.

[0017] According to an embodiment of this application, the apparatus further includes: a model update module, used to update the historical business dataset online based on the multi-source business data; and to train and update the Bayesian statistical model and the seasonal time series prediction model based on the updated historical business dataset.

[0018] According to an embodiment of this application, the apparatus further includes: a business data acquisition module, configured to acquire first business data through a secure interface based on m business data systems; wherein m is a positive integer; acquire industry data through a public interface based on publicly available industry data sources; and obtain the multi-source business data based on the industry data and the first business data.

[0019] According to an embodiment of this application, the business data acquisition module includes: a first data acquisition unit, configured to read second business data from the m business data systems through the security interface based on a dynamic token authentication method; wherein the second business data corresponding to sensitive fields in the m business data systems is pre-encrypted using an encryption algorithm; a second data acquisition unit, configured to transmit the second business data back through the security interface according to a standard encryption protocol; and a third data acquisition unit, configured to obtain the first business data based on the second business data, using a decryption algorithm corresponding to the encryption algorithm and a decryption rule corresponding to the standard encryption protocol.

[0020] According to an embodiment of this application, the apparatus further includes: an anomaly processing module, used to issue an early warning for abnormal data; wherein the abnormal data is business data in the multi-source business data that exceeds the confidence interval boundary; and to perform multi-dimensional drill-down analysis on the multi-source business data to locate the abnormal business unit corresponding to the abnormal data.

[0021] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0022] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0023] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0024] In the embodiments of this application, multi-source business data is used as the basis for prediction. This fully considers the business factors of each data source, providing massive data support for model prediction. It supplements the limitations of a single data source, reduces information blind spots, and enables the model to more accurately capture potential risks or patterns. Integrating multi-source business data allows the model to adapt to more complex real-world business scenarios. Furthermore, by leveraging Bayesian statistical models, seasonal time-series forecasting models, and multinomial models for business data prediction and correction, the advantages of different models are combined, providing a more accurate, flexible, and adaptable prediction solution that effectively combines short-term quarterly and long-term annual forecasts. By employing online-updated Bayesian and seasonal time-series models, the huge computational overhead of traditional batch repetitive training is avoided. The model learns incrementally, requiring only a small amount of computing resources to update parameters, significantly reducing the peak utilization of the central processing unit and memory, and improving computational efficiency. By employing a multi-model collaboration and correction mechanism, incremental data from the target year is used as trend reference data to accurately anchor year-end data, correcting the initial forecast based on quarterly decomposition, and constructing a more fault-tolerant forecasting method to improve forecast accuracy. Even if one model deviates due to short-term data noise, it can be corrected by the trend constraints of another model, reducing forecast errors, mitigating the systemic risk caused by the failure of a single model, and enhancing the robustness of the overall output, thereby making the system more stable and the decision-making more accurate and efficient. Attached Figure Description

[0025] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0026] Figure 1 The illustrations depict application scenarios of a business data prediction method, apparatus, device, medium, and program product based on multi-model fusion according to embodiments of this application.

[0027] Figure 2 A flowchart illustrating a business data prediction method based on multi-model fusion according to an embodiment of this application is shown schematically.

[0028] Figure 3 The illustration shows a rolling prediction result diagram of the business data prediction method based on multi-model fusion according to an embodiment of this application;

[0029] Figure 4 This illustration schematically shows a data acquisition diagram of a business data prediction method based on multi-model fusion according to an embodiment of this application;

[0030] Figure 5 This illustration schematically shows an abnormal data processing flowchart of a business data prediction method based on multi-model fusion according to an embodiment of this application;

[0031] Figure 6 This illustration schematically shows a long-cycle decomposition flowchart of a business data prediction method based on multi-model fusion according to an embodiment of this application;

[0032] Figure 7 The diagram illustrates the outstanding balance prediction results of the business data prediction method based on multi-model fusion according to an embodiment of this application.

[0033] Figure 8 This schematically illustrates a structural block diagram of a business data prediction apparatus based on multi-model fusion according to an embodiment of this application; and

[0034] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing a business data prediction method based on multi-model fusion, according to an embodiment of this application. Detailed Implementation

[0035] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0036] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0037] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0038] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0039] In the financial industry, business data forecasting is crucial, directly impacting the risk management and stability of financial institutions. However, traditional business forecasting models have certain limitations, mainly manifested in the following ways:

[0040] (1) Limitations of model singularity: Traditional methods rely on a single model, making it difficult to capture long-term trends, seasonal fluctuations and long-term anchoring at the same time.

[0041] (2) Difficult to meet the requirements of online systems: In pursuit of prediction accuracy, traditional single prediction models often have complex model structures and huge parameter spaces. When training and inferring on large-scale historical datasets with high dimensions and multiple sources, they generate extremely high computational complexity and memory consumption, resulting in long training cycles and high real-time prediction latency, making it difficult to meet the requirements of online business systems for low latency response and high throughput.

[0042] (3) Insufficient model generalization and stability: Static models are difficult to adapt to the dynamic changes in data flow. Once the trend and pattern of data drift, the performance of static models will deteriorate rapidly, and frequent batch full retraining is not feasible in terms of computational cost, which leads to a decrease in the reliability of prediction results and thus causes errors in resource allocation decisions based on this.

[0043] (4) Low data security. From the perspective of data security, a single model relies on a single data perspective and lacks a multi-model collaborative verification mechanism, making it more sensitive to abnormal inputs or potential data pollution.

[0044] (2) Lack of interpretability of models in the field of machine learning: Although machine learning models have certain advantages in prediction accuracy, their internal operating mechanisms are complex. The complex internal calculations make it difficult to clearly explain how the model arrives at specific prediction results. In the financial field, when risk assessment and regulation are involved, it is difficult to explain the decision-making basis to business personnel and regulatory agencies.

[0045] (3) Lack of guidance or restrictions based on business understanding: Existing models often only start from the data itself for training and prediction, without fully combining the actual characteristics and rules of financial business, failing to combine other information within financial institutions that may affect business, failing to consider business factors such as the impact of macroeconomic policies on business, and failing to incorporate professional judgments or restrictions, which may lead to the prediction results being out of touch with the actual business situation.

[0046] (4) The dynamic update mechanism has security and coordination defects: the existing dynamic feedback technology lacks end-to-end encryption protection. When the predicted value is corrected in real time, the sensitive data is transmitted in plaintext, which is vulnerable to man-in-the-middle attacks or data tampering, resulting in a disconnect between security strategy and real-time update requirements.

[0047] This application provides a business data prediction method based on multi-model fusion, comprising: predicting incremental data for a target year using a Bayesian statistical model based on multi-source business data; wherein the Bayesian statistical model is trained and updated online based on historical business datasets; obtaining incremental data for a target quarter using a seasonal time-series prediction model based on multi-source business data; wherein the seasonal time-series prediction model is trained and updated online based on historical business datasets; obtaining initial prediction data for the target year and fluctuation data for the remaining quarters by performing long-term decomposition prediction using the seasonal time-series prediction model and a multinomial model based on the incremental data of the target quarter; wherein the target quarter and the remaining quarters constitute the target year; and correcting the initial prediction data for the target year based on the fluctuation data of the remaining quarters using the incremental data of the target year as trend reference data to obtain the business data prediction result.

[0048] In the embodiments of this application, multi-source business data is used as the basis for prediction. This fully considers the business factors of each data source, providing massive data support for model prediction. It supplements the limitations of a single data source, reduces information blind spots, and enables the model to more accurately capture potential risks or patterns. Integrating multi-source business data allows the model to adapt to more complex real-world business scenarios. Furthermore, by leveraging Bayesian statistical models, seasonal time-series forecasting models, and multinomial models for business data prediction and correction, the advantages of different models are combined, providing a more accurate, flexible, and adaptable prediction solution that effectively combines short-term quarterly and long-term annual forecasts. By employing online-updated Bayesian and seasonal time-series models, the huge computational overhead of traditional batch repetitive training is avoided. The model learns incrementally, requiring only a small amount of computing resources to update parameters, significantly reducing the peak utilization of the central processing unit and memory, and improving computational efficiency. By employing a multi-model collaboration and correction mechanism, incremental data from the target year is used as trend reference data to accurately anchor year-end data, correcting the initial forecast based on quarterly decomposition, and constructing a more fault-tolerant forecasting method to improve forecast accuracy. Even if one model deviates due to short-term data noise, it can be corrected by the trend constraints of another model, reducing forecast errors, mitigating the systemic risk caused by the failure of a single model, and enhancing the robustness of the overall output, thereby making the system more stable and the decision-making more accurate and efficient.

[0049] Figure 1 The illustration shows an application scenario of a business data prediction method, apparatus, device, medium, and program product based on multi-model fusion according to embodiments of this application.

[0050] like Figure 1As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0051] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0052] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0053] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0054] It should be noted that the business data prediction method based on multi-model fusion provided in this application embodiment can generally be executed by server 105. Correspondingly, the business data prediction device based on multi-model fusion provided in this application embodiment can generally be located in server 105. The business data prediction method based on multi-model fusion provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the business data prediction device based on multi-model fusion provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0055] It should be understood that Figure 1The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0056] It should be noted that the business data prediction method, apparatus, equipment, media, and program products based on multi-model fusion of this application can be used in the fields of artificial intelligence, big data, and fintech, as well as in any field other than fintech. The application areas of the business data prediction method, apparatus, equipment, media, and program products based on multi-model fusion of this application are not limited. This application combines regression statistical inference, time series analysis, and multinomial estimation to construct a hybrid prediction method for mixed business data, which can be applied to scenarios such as business management of commercial banks, management of outstanding business of financial institutions, liquidity risk early warning, and regulatory capital measurement.

[0057] The following will be based on Figure 1 The described scene, through Figures 2-7 A detailed description is provided of the business data prediction method based on multi-model fusion according to embodiments of this application.

[0058] Figure 2 A flowchart illustrating a business data prediction method based on multi-model fusion according to an embodiment of this application is shown.

[0059] like Figure 2 As shown, the business data prediction method based on multi-model fusion in this embodiment includes operations S210 to S240. This method does not limit the specific executing entity. The executing entity can be any electronic device, such as a terminal device or a server device. The executing entity can also be any software application or client.

[0060] In operation S210, based on multi-source business data, a Bayesian statistical model is used to predict incremental data for the target year; the Bayesian statistical model is trained and updated online based on historical business datasets.

[0061] Multi-source business data includes business data from internal business data systems and industry data from publicly available external industry data sources. Among them, business includes, but is not limited to, credit business (such as the company's outstanding business, outstanding business volume, credit business revenue, non-performing outstanding amount, etc.), marketing and sales (such as customer growth, product sales, system transaction volume, marketing activity growth, etc.), and risk management business (such as credit default rate, liquidity outstanding risk, etc.).

[0062] Bayesian statistical models include, but are not limited to, Bayesian linear regression models, Bayesian lasso regression models, and Bayesian ridge regression models. A trend baseline is constructed using the annual incremental data predicted by Bayesian statistical models to provide data-supported predictions of the incremental data for the following year, thus anchoring the incremental data for that year.

[0063] Preferably, a Bayesian ridge regression model can be used to predict the incremental data for the target year. For example, a Bayesian ridge regression model can be used to predict the total annual increase of the company's outstanding balance in the coming year. The outstanding balance represents the amount that the borrower needs to repay at a certain point in time. The input parameters are mainly business data from the internal business data system and supplemented by publicly available market industry data from external sources. The internal business data is shown in Table 1, and the external industry data is shown in Table 2 below.

[0064] Table 1 Internal Business Data

[0065]

[0066] Table 2 External Industry Data

[0067]

[0068] Using the aforementioned internal and external variables as parameters, and the annual increase in the company's outstanding balance as the output, these parameters are incorporated into the training set of the Bayesian Ridge regression model. A 10-year period of business data can be used as the training set to train the Bayesian Ridge regression model. By inputting annual parameters (based on the business data of each department's regulation, plan, or reasonable expectations) into the trained model, the total annual increase can be obtained. Business departments can also adjust certain business data (such as the deposit reserve ratio) based on the actual situation to provide a trend anchor for the overall forecast.

[0069] In operating S220, based on multi-source business data, a seasonal time-series forecasting model is used to obtain incremental data for the target quarter; the seasonal time-series forecasting model is trained and updated online based on historical business datasets.

[0070] Seasonal time-series forecasting models include, but are not limited to, the Seasonal Exponential Smoothing (SES) model, the Seasonal Autoregressive (SAR) model, and the Seasonal Autoregressive Integrated Moving Average (SARIMA) model, which are used to forecast short-term seasonal fluctuations in business data.

[0071] Preferably, the rolling forecasting mechanism of the SARIMA model can be utilized. The parameters of its non-seasonal component can be set to have no autoregressive term, first-order differencing, and no moving average term. First-order differencing means the model primarily focuses on short-term trends in the data (i.e., the difference between adjacent time points), such as the relationship between February and January in year Y. The parameters of the seasonal component can be set to have a first-order seasonal autoregressive term, first-order seasonal differencing, no seasonal moving average term, and a period length of 12. This parameter setting effectively captures the cyclical fluctuations in the data. The parameters combine the trends of short-term adjacent months with the patterns of months in the same cycle from the previous year. For example, the forecast for February in year Y will be mainly influenced by January of year Y, January of year Y-1, and February of year Y-1. This parameter effectively captures the cyclical fluctuations in the data, effectively capturing the seasonal fluctuations with a 12-month cycle, making the forecast results more closely match the annual recurring pattern.

[0072] The process of obtaining incremental data for the target quarter includes: using a seasonal time-series forecasting model based on multi-source business data to predict the original value of the target quarter; and subtracting the actual value at the end of the previous quarter from the original value of the target quarter to obtain the incremental data for the target quarter.

[0073] For example, a 10-year historical outstanding balance can be used as input to train a SARIMA model. The trained SARIMA model can then predict the outstanding balance for the next three months (the target quarter) and output a confidence interval. Based on the predicted outstanding balance for the target quarter, the predicted outstanding balance at the end of the target quarter is determined. Subtracting the balance at the end of the previous quarter gives the increment for that target quarter. Alternatively, any value within the confidence interval can be selected as the short-term predicted outstanding balance. For instance, when the predicted short-term outstanding balance is likely to be higher due to policy influences, the business department can choose values ​​from the original confidence interval to its upper limit.

[0074] In operation S230, based on the incremental data of the target quarter, long-term decomposition forecasts are performed using a seasonal time-series forecasting model and a multinomial model to obtain the initial forecast data for the target year and the fluctuation data for the remaining quarters; whereby the target quarter and the remaining quarters constitute the target year.

[0075] Polynomial models include, but are not limited to, Ordinary Least Squares (OLS) polynomial models, ridge regression polynomial models, and lasso regression polynomial models.

[0076] The target year includes the target quarter and the remaining quarters. In the long-cycle decomposition process, the characteristics of seasonal time series forecasting models and multinomial models are combined to avoid the errors that accumulate in the later stages of the remaining quarters in the seasonal time series forecasting model. First, the quarterly increment ratio of the remaining quarters is adjusted based on the incremental data of the target quarter, and the original forecast value of the target year is predicted by the seasonal time series forecasting model. Second, based on the adjusted quarterly increment ratio and the original forecast value, the adjusted fitted value of the target year and the predicted fitted value of the remaining quarters are obtained by fitting the model with a multinomial model. Finally, the initial forecast data of the target year is determined based on the adjusted fitted value of the target year, and the fluctuation data of the remaining quarters are determined based on the original forecast value and the predicted fitted value of the remaining quarters.

[0077] In operation S240, the incremental data of the target year is used as the trend reference data, and the initial forecast data of the target year is corrected based on the fluctuation data of the other quarters to obtain the business data forecast results.

[0078] Using the incremental data of the target year as trend reference data, the prediction of the annual increment can be combined with the Bayesian ridge regression model to constrain the overall trend. The fluctuation data of the remaining quarters are then superimposed onto the initial forecast data of those quarters to obtain the final forecast data for the remaining quarters. Based on the initial forecast data of the target quarter and the final forecast data of the remaining quarters, a business data forecast result is generated, which is the final forecast result for the target year. This approach superimposes fluctuation information onto the quarterly proportional fitting curve corresponding to the adjusted fitted value. In the final forecast data of the remaining quarters, seasonal or random fluctuation biases in trend forecasting are corrected, avoiding the accumulation of errors over time in medium- and long-term forecasts.

[0079] The SARIMA model tends to have a large prediction error over a 12-month period. In multi-step predictions, each step depends on the prediction value of the previous step, and the error gradually accumulates during the iteration process. Therefore, by using the fluctuation data of other quarters to correct the initial prediction data for the target year, the prediction error can be reduced. Combining this with the prediction of the annual increment by a Bayesian statistical model, the original fluctuations can be transferred to the initial prediction, thus avoiding the error that accumulates continuously in the later stages of the SARIMA model.

[0080] In the embodiments of this application, multi-source business data is used as the basis for prediction. This fully considers the business factors of each data source, providing massive data support for model prediction. It supplements the limitations of a single data source, reduces information blind spots, and enables the model to more accurately capture potential risks or patterns. Integrating multi-source business data allows the model to adapt to more complex real-world business scenarios. Furthermore, by leveraging Bayesian statistical models, seasonal time-series forecasting models, and multinomial models for business data prediction and correction, the advantages of different models are combined, providing a more accurate, flexible, and adaptable prediction solution that effectively combines short-term quarterly and long-term annual forecasts. By employing online-updated Bayesian and seasonal time-series models, the huge computational overhead of traditional batch repetitive training is avoided. The model learns incrementally, requiring only a small amount of computing resources to update parameters, significantly reducing the peak utilization of the central processing unit and memory, and improving computational efficiency. By employing a multi-model collaboration and correction mechanism, incremental data from the target year is used as trend reference data to accurately anchor year-end data, correcting the initial forecast based on quarterly decomposition, and constructing a more fault-tolerant forecasting method to improve forecast accuracy. Even if one model deviates due to short-term data noise, it can be corrected by the trend constraints of another model, reducing forecast errors, mitigating the systemic risk caused by the failure of a single model, and enhancing the robustness of the overall output, thereby making the system more stable and the decision-making more accurate and efficient.

[0081] Furthermore, the forecast results output by this application's business data prediction not only represent predicted values ​​but also constitute structured data objects containing trend and fluctuation components. These forecast results provide richer and more interpretable decision-making support for downstream monitoring, alarm, and resource scheduling systems. For example, the system can perform long-term capacity planning based on robust trend components while triggering short-term, fine-grained elastic scaling strategies based on fluctuation components, thereby improving the automated operational efficiency and stability of the entire data center.

[0082] According to embodiments of this application, the business data prediction method based on multi-model fusion further includes: updating the historical business dataset online based on multi-source business data; and training and updating the Bayesian statistical model and the seasonal time series prediction model based on the updated historical business dataset.

[0083] During model updates, distributed computing can be used to accelerate the process. For computationally intensive tasks such as posterior updates of Bayesian statistical models and rolling forecasts of seasonal time series prediction models, distributed graphics processing unit (GPU) clusters can be deployed to achieve parallel computation and iterative optimization of model parameters.

[0084] The security interface pushes data, triggering model data updates. The model update process can be scheduled to start daily at midnight to ensure the latest data is included in the forecasts. The forecast frequency can be accurate to daily, and business changes (such as sudden policy changes) can be reflected in the forecast results the following day. The Bayesian statistical model and seasonal time-series forecast model are updated online. Upon receiving new data, the actual values ​​are updated at a customizable frequency. The seasonal time-series forecast model continuously updates the forecast range for the next three months. For the annual increment, the Bayesian regression result at the beginning of the year can be maintained, or the estimate of the annual increment can be adjusted upwards or downwards based on changes in external indicators. If the supply growth rate is lower than expected, the annual average supply forecast can be lowered, thus affecting the annual increment result of the Bayesian ridge regression. For example, if the credit rating migration rate increases on a certain day (B→A level customers increase by 0.1%), the future outstanding balance forecast will be automatically adjusted upwards after the model is updated the next day. The business department can view the latest forecast curve daily and dynamically adjust the outstanding payment strategy. The audit department can trace the daily data update records to ensure operational compliance.

[0085] Figure 3 The illustration shows a rolling prediction result diagram of the business data prediction method based on multi-model fusion according to an embodiment of this application, such as... Figure 3 As shown, the original predicted value represents the original value predicted by the seasonal time series forecasting model, the rolling predicted value represents the business data forecast result after the rolling update of the model, the actual test value represents the business data statistically collected by the actual business data system, and the strip area represents the confidence interval predicted by the seasonal time series forecasting model. When the rolling forecast is rolled every three months, the accuracy of the data is significantly improved compared with the annual frequency forecast. If the rolling update frequency is increased, the data forecast accuracy will also be improved.

[0086] In the embodiments of this application, the model is updated online so that the predicted values ​​can be corrected in real time with the actual data, and multi-source business data is incorporated into annual and quarterly incremental information in a timely manner to ensure that the predictions fit the actual dynamic changes in the business.

[0087] Figure 4 The diagram illustrates a data acquisition method for business data prediction based on multi-model fusion according to an embodiment of this application.

[0088] like Figure 4 As shown, before operation 210 predicts the incremental data for the target year based on multi-source business data using a Bayesian statistical model, operations S410 to S430 are included.

[0089] When operating S410, based on m business data systems, the first business data is obtained through a security interface; where m is a positive integer.

[0090] By connecting to internal data sources such as risk management systems, credit systems, and business processing systems through secure interfaces, the system pushes the latest business data (such as updated credit rating migration rates, pre-approved credit utilization rates, product approval rates, existing pre-approved credit utilization rates, and outstanding mortgage / pledge discount rates) through these secure interfaces. Data pushes trigger model data updates, and the model update process can be scheduled to start daily at midnight to ensure the latest data is included in predictions. The secure interface refers to an application programming interface (API) that uses technical means to ensure data transmission, authentication, and access control, primarily preventing unauthorized access, unauthorized calls, and malicious attacks.

[0091] In the technical solution of this application, when business data involves user information, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0092] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0093] According to the embodiments of this application, in operation S410, which obtains the first business data through a security interface based on m business data systems, operations S4101 to S4103 are included.

[0094] In operation S4101, based on dynamic token authentication, the second business data of m business data systems is read through a secure interface; wherein, the second business data corresponding to the sensitive fields in the m business data systems is pre-encrypted using an encryption algorithm.

[0095] With end-to-end encryption, sensitive fields are encrypted using encryption algorithms, such as the Advanced Encryption Standard (AES), before being transmitted through the secure interface. The keys are managed centrally by the financial institution's internal system.

[0096] Dynamic token authentication refers to a method where secure interfaces accessing business data systems require callers to authenticate using tokens that change dynamically over time or under certain conditions. Tokens are issued using a dynamic token tool. The caller uses this dynamic token to send a request to the secure interface. The business data system generates a currently valid token using the same key and algorithm, compares it with the dynamic token submitted by the caller, and checks its validity within a time window (e.g., 1 minute). If the comparison and check pass, the secure interface can read business data from the business data system. Dynamic token authentication ensures the caller's identity is genuine (not stolen). Since the token is not a fixed key or password, this mechanism prevents stolen tokens from remaining valid indefinitely, significantly improving the security of interface access.

[0097] When operating the S4102, second business data is transmitted back through a secure interface according to the standard encryption protocol.

[0098] Establishing a secure transmission channel can utilize standard encryption protocols in the financial industry to ensure data transmission is protected against theft and tampering. Encrypted transmission safeguards data security, ensuring that model predictions are synchronized with business changes, while also meeting financial data compliance requirements. This ensures that encrypted transmission and storage comply with financial data security requirements, keeping sensitive data confidential and achieving data security compliance.

[0099] In operation S4103, based on the second business data, the first business data is obtained according to the decryption algorithm corresponding to the encryption algorithm and the decryption rules corresponding to the standard encryption protocol.

[0100] The second business data is the business data in the business data system read through the security interface, and the first business data is the decrypted business data corresponding to the second business data.

[0101] The secure interface can also authorize only the model service to receive and decrypt data, achieving access control isolation. Input data, intermediate parameters, and prediction results for each model are all encrypted using Advanced Encryption Standard – Galois Counter Mode (AES-GCM) before storage, supporting periodic key rotation, recording daily data updates and model adjustment logs, and enabling auditing and verification.

[0102] In the embodiments of this application, dynamic feedback of business data is established, and a dynamic feedback mechanism is implemented to achieve encrypted real-time transmission and key management, ensuring data confidentiality throughout the entire transmission, computation, and storage chain, and improving data security.

[0103] When operating S420, industry data is obtained through public interfaces based on publicly available industry data sources.

[0104] By using publicly available industry data sources (such as public institution platforms, industry association platforms, etc.) and their provided public interfaces, requests are sent according to the interface specifications, and the returned data is received and parsed to complete the acquisition of industry data.

[0105] By operating the S430, multi-source business data is obtained based on industry data and primary business data.

[0106] The acquired industry data and primary business data are cleaned, merged, and standardized in format to obtain multi-source business data.

[0107] In the embodiments of this application, first business data is obtained through a secure interface, and industry data is obtained through a public interface. This secure acquisition of business data and industry data ensures data compliance and improves the diversity and reliability of data sources.

[0108] Figure 5 The diagram illustrates an anomaly data processing flowchart of a business data prediction method based on multi-model fusion according to an embodiment of this application.

[0109] like Figure 5 As shown, the business data prediction method based on multi-model fusion also includes operations S510 to S520.

[0110] When operating S510, an alert is issued for abnormal data; the abnormal data refers to business data from multiple sources that exceeds the confidence interval boundary.

[0111] A predictive confidence interval can be constructed based on a pre-set confidence level (e.g., 70%) using a seasonal time-series forecasting model. This confidence interval can then be used to issue early warnings for abnormal data. Daily driving of the model with key indicators such as credit rating migration rates improves the timeliness of risk warnings. When the actual observed value in multi-source business data exceeds the boundary of the corresponding confidence interval, it indicates a statistically significant deviation in the business data indicator, automatically triggering an anomaly warning mechanism.

[0112] To enhance business flexibility, a confidence interval selection mechanism can be built, with annual and quarterly increments supporting both manual and default inputs, providing business departments with a wider range of short-term forecast results to meet diverse business needs.

[0113] When operating S520, multi-source business data is subjected to multi-dimensional drill-down analysis to locate the abnormal business units corresponding to abnormal data.

[0114] Taking the monitoring of abnormal outstanding balances of enterprises as an example, when conducting multi-dimensional drill-down analysis of multi-source business data, the process begins with cleaning the data and using data analysis tools to pinpoint the time periods of abnormal indicators. Then, the data is drilled down to specific months and weeks, and further broken down by industry (e.g., manufacturing, service) and enterprise size (e.g., large, medium, small, and micro-enterprises), and geographically to branch offices. Cross-analysis revealed that the combined dimension of "a regional branch + manufacturing + medium-sized enterprise owner" resulted in a 70% decrease in outstanding balances. Reviewing approval logs confirmed that this was due to overcapacity in the regional manufacturing sector leading to tighter credit, and that the risk control model had adjusted the entry threshold for medium-sized enterprises. Simultaneously, the SARIMA model can be used to predict a continued downward trend in outstanding balances, identifying the abnormal business units as the corporate credit approval department and risk control policy research team of a regional branch.

[0115] Multidimensional drill-down analysis can accurately pinpoint the source of anomalies, enabling dual-dimensional anomaly attribution diagnosis across regions and industries. It can locate specific business branches or industries, and specifically identify market segments or business units with structural deviations.

[0116] In the embodiments of this application, anomalies are detected and warned in a timely manner, risks are reduced, business stability is ensured, data reliability and decision-making accuracy are improved, security is enhanced, and the source of anomalies is accurately located.

[0117] Figure 6 The diagram illustrates a long-cycle decomposition flowchart of a business data prediction method based on multi-model fusion according to an embodiment of this application.

[0118] like Figure 6 As shown, in operation 230, based on the incremental data of the target quarter, long-term decomposition prediction is performed using a seasonal time-series forecasting model and a multinomial model to obtain the initial forecast data for the target year and the fluctuation data for the remaining quarters, including operations S610 to S640.

[0119] When operating S610, the incremental data for the target quarter is adjusted to a quarterly percentage to determine the quarterly incremental proportion.

[0120] After the seasonal time series forecasting model completes the forecast for the short-term target quarter (such as Q1), if the predicted quarterly incremental percentage deviates from the initial setting, the percentages for the following three quarters (such as Q2, Q3, and Q4) need to be reallocated and adjusted to determine the quarterly incremental percentage.

[0121] According to the embodiments of this application, in operation S610, the incremental data of the target quarter is adjusted to a quarterly ratio to determine the quarterly incremental percentage, which includes operations S6101 to S6102.

[0122] In operation S6101, the incremental data of the target quarter is used to determine the business increment for the remaining quarters.

[0123] The business growth in other quarters = 100% - the percentage of growth in the target quarter. For example, if the growth in the first quarter (Q1) accounts for 70% of the annual growth, the business growth in the other quarters will be 30%.

[0124] When operating S6102, based on the preset quarterly incremental ratio, the business incremental for the remaining quarters is allocated to obtain the quarterly incremental percentage.

[0125] The preset quarterly incremental percentage can be the historical average percentage of the incremental percentage of each quarter in the historical data, such as 30% for Q1, 30% for Q2, 20% for Q3, and 20% for Q4. Alternatively, the incremental percentage of the four quarters can be set freely according to business rules, and the sum of the four quarters must be 100%, such as 25% for Q1, 25% for Q2, 20% for Q3, and 30% for Q4.

[0126] If the business increment in other quarters is 30%, the remaining increment in other quarters will be allocated according to the historical average ratio or business rules. For example, if the average ratio of Q2:Q3:Q4 is 4:3:3, then Q2 will be allocated: 30%×4 / 10=12%, Q3 will be allocated: 30%×3 / 10=9%, and Q4 will be allocated 9%.

[0127] In the embodiments of this application, the incremental business volume of the remaining quarters can be accurately determined and allocated, thereby improving the rationality of the quarterly proportion and the accuracy of business forecasting.

[0128] When operating S620, based on the quarterly incremental percentage, a trend is fitted using a multinomial model to generate the adjusted fitted value for the target year, and based on the adjusted fitted value for the target year, the initial forecast data for the target year is determined.

[0129] Using the incremental percentage of the target quarter as trend reference data, a multinomial model is used to generate adjusted fitted values ​​for the target quarter. Based on the adjusted fitted values ​​of the remaining quarters and the target quarter, initial forecast data for the target year is obtained, resulting in a quarterly proportion fitting curve. Preferably, an OLS multinomial model, such as a cubic OLS multinomial, is used. For example, based on the adjusted quarterly incremental percentage, a cubic OLS multinomial is used to perform trend fitting on the original forecast values ​​for the remaining quarters (e.g., Q2-Q4) to generate adjusted fitted values ​​for the remaining quarters.

[0130] When operating S630, based on multi-source business data, a seasonal time-series forecasting model is used to perform rolling forecasts and output the original forecast value for the target year; where the original forecast value for the target year includes the original forecast values ​​for the other quarters.

[0131] By using seasonal time-series forecasting models and updating data on a rolling basis, the predicted results of the target's original business data are obtained, i.e., the original forecast values. For example, using the SARIMA model to perform rolling forecasts of business data throughout the year, the output SARIMA original values ​​include seasonal fluctuations and trends.

[0132] To predict the monthly original values ​​for the entire year Y, firstly, use business data up to December of year Y-1 to predict the original value for January of year Y; secondly, once the actual data for January of year Y is generated, incorporate the original value into the seasonal time series forecasting model, update the model parameters, and then predict the original value for February of year Y; and so on. For each new actual data obtained, use historical data plus the latest actual data to re-predict the next point in time until the entire year Y forecast is completed.

[0133] When operating S640, the original forecast values ​​for the remaining quarters are transferred using a polynomial model to obtain the fluctuation data for the remaining quarters.

[0134] In the embodiments of this application, the original predicted values ​​of the seasonal time series forecasting model are subjected to fluctuation migration. By utilizing the accurate fluctuation information of the seasonal time series forecasting model, the error that accumulates continuously in the later stages of the seasonal time series forecasting model is avoided, thus avoiding the disadvantage of the error accumulating continuously in the medium and long term forecasting and improving the forecast accuracy.

[0135] According to the embodiments of this application, in operation S640, the original predicted values ​​of the remaining quarters are transferred to the fluctuation data of the remaining quarters through a polynomial model, including operations S6401 to S6402.

[0136] In operation S6401, the original predicted values ​​for the remaining quarters are fitted using a polynomial model to obtain the fitted predicted values ​​for the remaining quarters.

[0137] For the original forecast values ​​of the remaining quarters, a polynomial model is used to fit the trend term to obtain the predicted fitted values ​​for the remaining quarters. For example, for the original SARIMA values ​​of the last 9 months (Q2~Q4), an OLS cubic polynomial is used to fit the trend term to obtain the predicted fitted values ​​for the last 9 months. If the original forecast value of the target year is used to fit the trend term using a polynomial model to obtain the predicted fitted value for the target year, then the original forecast value fitting curve is obtained.

[0138] In operation S6402, the fluctuation data for the remaining quarters are calculated based on the original forecast values ​​and the fitted forecast values ​​for the remaining quarters.

[0139] The volatility data includes volatility information values. The volatility information values ​​for other quarters are equal to the original forecast values ​​for other quarters minus the fitted forecast values ​​for other quarters. The volatility information values ​​can reflect the part of short-term volatility that deviates from the long-term trend.

[0140] In the embodiments of this application, based on the original predicted value and the fitted value, fluctuations are accurately mapped to achieve residual fluctuation migration. The fitted value of short-term fluctuation characteristics is accurately mapped to the original predicted value curve of medium and long term, and the fluctuations of the original predicted value of the seasonal time series prediction model are migrated to the fitted line to reflect the monthly fluctuation characteristics and improve the comprehensiveness of the prediction.

[0141] Taking the prediction of an enterprise's outstanding balance using the business data prediction method based on multi-model fusion proposed in this application as an example, Figure 7 The diagram illustrates the outstanding balance prediction results of the business data prediction method based on multi-model fusion according to an embodiment of this application, as shown in the figure. Figure 7 In the diagram, the horizontal axis covers the prediction time range, the vertical axis represents the numerical fluctuation range, the SARIMA predicted value represents the original prediction value, the SARIMA predicted value OLS fitting curve represents the original prediction value fitting curve, the adjusted predicted value represents the business data prediction result, the new planned value OLS fitting represents the quarterly proportion fitting curve, the test actual value represents the business data statistically collected by the actual business data system, the training actual value represents the business data used to input into the training model, and the strip area represents the confidence interval of the model's prediction of the target quarter. The original prediction value, the original prediction value fitting curve, the business data prediction result, and the quarterly proportion fitting curve have been explained in the corresponding operation methods, and will not be elaborated on here.

[0142] Based on the aforementioned business data prediction method based on multi-model fusion, this application also provides a business data prediction device based on multi-model fusion. The following will combine... Figure 8 The device is described in detail.

[0143] Figure 8 The diagram illustrates the structure of a business data prediction apparatus based on multi-model fusion according to an embodiment of this application.

[0144] like Figure 8 As shown, the business data prediction device 800 based on multi-model fusion in this embodiment includes a statistical prediction module 810, a quarterly prediction module 820, a periodic decomposition module 830, and a result correction module 840.

[0145] The statistical prediction module 810 is used to predict incremental data for a target year based on multi-source business data using a Bayesian statistical model; wherein the Bayesian statistical model is trained and updated online based on historical business datasets. In one embodiment, the statistical prediction module 810 can be used to perform the operation S210 described above, which will not be repeated here.

[0146] The quarterly forecasting module 820 is used to obtain incremental data for the target quarter based on multi-source business data and a seasonal time-series forecasting model; wherein, the seasonal time-series forecasting model is trained and updated online based on historical business datasets. In one embodiment, the quarterly forecasting module 820 can be used to perform the operation S220 described above, which will not be repeated here.

[0147] The cycle decomposition module 830 is used to perform long-cycle decomposition forecasting based on the incremental data of the target quarter using a seasonal time-series forecasting model and a multinomial model to obtain the initial forecast data for the target year and the fluctuation data for the remaining quarters; wherein, the target quarter and the remaining quarters constitute the target year. In one embodiment, the cycle decomposition module 830 can be used to perform the operation S230 described above, which will not be repeated here.

[0148] The result correction module 840 is used to correct the initial forecast data for the target year based on the incremental data of the target year as trend reference data and the fluctuation data of the remaining quarters, thereby obtaining the business data forecast result. In one embodiment, the result correction module 840 can be used to perform the operation S240 described above, which will not be repeated here.

[0149] According to an embodiment of this application, the periodic decomposition module 830 includes: a ratio adjustment unit, used to adjust the incremental data of the target quarter by quarterly ratio to determine the quarterly incremental percentage; a trend fitting unit, used to perform trend fitting using a polynomial model based on the quarterly incremental percentage to generate an adjusted fitting value for the target year, and to determine the initial forecast data for the target year based on the adjusted fitting value for the target year; a rolling forecast unit, used to perform rolling forecasting using a seasonal time series forecasting model based on multi-source business data to output the original forecast value for the target year; wherein the original forecast value for the target year includes the original forecast values ​​for the remaining quarters; and a fluctuation migration unit, used to perform fluctuation migration on the original forecast values ​​for the remaining quarters using a polynomial model to obtain the fluctuation data for the remaining quarters.

[0150] According to an embodiment of this application, the ratio adjustment unit includes: a first ratio adjustment subunit, used to determine the business increment of the remaining quarters based on the incremental data of the target quarter; and a second ratio adjustment subunit, used to allocate the business increment of the remaining quarters based on a preset quarterly increment ratio to obtain the quarterly increment percentage.

[0151] According to an embodiment of this application, a fluctuation migration unit includes: a first fluctuation migration subunit, configured to fit the original predicted values ​​of the remaining quarters using a polynomial model to obtain the predicted fitted values ​​of the remaining quarters; and a second fluctuation migration subunit, configured to calculate the fluctuation data of the remaining quarters based on the original predicted values ​​of the remaining quarters and the predicted fitted values ​​of the remaining quarters.

[0152] According to an embodiment of this application, the apparatus 800 further includes: a model update module, used to update the historical business dataset online based on multi-source business data; and to train and update the Bayesian statistical model and the seasonal time series prediction model based on the updated historical business dataset.

[0153] According to an embodiment of this application, the apparatus 800 further includes: a business data acquisition module, configured to acquire first business data through a secure interface based on m business data systems; wherein m is a positive integer; acquire industry data through a public interface based on publicly available industry data sources; and obtain multi-source business data based on the industry data and the first business data.

[0154] According to an embodiment of this application, a business data acquisition module includes: a first data acquisition unit, configured to read second business data from m business data systems via a secure interface based on a dynamic token authentication method; wherein the second business data corresponding to sensitive fields in the m business data systems is pre-encrypted using an encryption algorithm; a second data acquisition unit, configured to transmit the second business data back via a secure interface according to a standard encryption protocol; and a third data acquisition unit, configured to obtain first business data based on the second business data, using a decryption algorithm corresponding to the encryption algorithm and a decryption rule corresponding to the standard encryption protocol.

[0155] According to an embodiment of this application, the device 800 further includes: an anomaly processing module, used to issue an early warning for abnormal data; wherein the abnormal data is business data in multi-source business data that exceeds the confidence interval boundary; and to perform multi-dimensional drill-down analysis on the multi-source business data to locate the abnormal business unit corresponding to the abnormal data.

[0156] According to embodiments of this application, any multiple modules among the statistical prediction module 810, quarterly prediction module 820, periodic decomposition module 830, and result correction module 840 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the statistical prediction module 810, quarterly prediction module 820, periodic decomposition module 830, and result correction module 840 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the statistical forecasting module 810, the quarterly forecasting module 820, the periodic decomposition module 830, and the result correction module 840 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0157] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing a business data prediction method based on multi-model fusion, according to an embodiment of this application.

[0158] like Figure 9 As shown, an electronic device 900 according to an embodiment of this application includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0159] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 902 and / or RAM 903. It should be noted that programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in one or more memories.

[0160] According to embodiments of this application, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0161] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0162] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0163] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the business data prediction method based on multi-model fusion provided in the embodiments of this application.

[0164] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0165] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0166] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0167] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0168] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0169] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A business data prediction method based on multi-model fusion, characterized in that, The method includes: Based on multi-source business data, a Bayesian statistical model is used to predict incremental data for the target year; wherein, the Bayesian statistical model is trained and updated online based on historical business datasets; Based on the multi-source business data, incremental data for the target quarter is obtained using a seasonal time-series prediction model; wherein, the seasonal time-series prediction model is trained and updated online based on the historical business dataset; Based on the incremental data of the target quarter, long-term decomposition forecasts are performed using the seasonal time-series forecasting model and a multinomial model to obtain the initial forecast data for the target year and the fluctuation data for the remaining quarters; wherein, the target quarter and the remaining quarters constitute the target year; and Using the incremental data of the target year as trend reference data, and based on the fluctuation data of the remaining quarters, the initial forecast data of the target year is corrected to obtain the business data forecast result.

2. The method according to claim 1, characterized in that, The step of obtaining initial forecast data for the target year and fluctuation data for the remaining quarters by performing long-term decomposition forecasting using the seasonal time-series forecasting model and the multinomial model based on the incremental data of the target quarter includes: The incremental data for the target quarter is adjusted according to the quarterly ratio to determine the quarterly incremental percentage; Based on the quarterly incremental percentage, the trend is fitted using the multinomial model to generate the adjusted fitted value for the target year, and based on the adjusted fitted value for the target year, the initial forecast data for the target year is determined. Based on the multi-source business data, a rolling forecast is performed using the seasonal time-series forecasting model to output the original forecast value for the target year; wherein, the original forecast value for the target year includes the original forecast values ​​for the remaining quarters; and By using the polynomial model, the original predicted values ​​for the remaining quarters are subjected to fluctuation shifting to obtain the fluctuation data for the remaining quarters.

3. The method according to claim 2, characterized in that, The step of adjusting the incremental data for the target quarter to determine the quarterly incremental percentage includes: Based on the incremental data of the target quarter, determine the business increment for the remaining quarters; and Based on the preset quarterly incremental ratio, the business incremental for the remaining quarters is allocated to obtain the quarterly incremental percentage.

4. The method according to claim 2, characterized in that, The step of using the polynomial model to perform fluctuation shifting on the original forecast values ​​for the remaining quarters to obtain fluctuation data for the remaining quarters includes: By fitting the original predicted values ​​for the remaining quarters to the polynomial model, the predicted fitted values ​​for the remaining quarters are obtained; and The fluctuation data for the remaining quarters are calculated based on the original forecast values ​​and the fitted forecast values ​​for the remaining quarters.

5. The method according to claim 1, characterized in that, The method further includes: Based on the aforementioned multi-source business data, the historical business dataset is updated online; and Based on the updated historical business dataset, the Bayesian statistical model and the seasonal time series prediction model are trained and updated.

6. The method according to claim 1, characterized in that, The method further includes: Based on m business data systems, the first business data is obtained through a secure interface; where m is a positive integer. Industry data is obtained through publicly available industry data sources and public APIs; and The multi-source business data is obtained based on the industry data and the first business data.

7. The method according to claim 6, characterized in that, The system based on m business data obtains the first business data through a secure interface, including: Based on the dynamic token authentication method, the second business data of the m business data systems is read through the security interface; wherein, the second business data corresponding to the sensitive fields in the m business data systems is pre-encrypted using an encryption algorithm; According to the standard encryption protocol, the second service data is transmitted back through the secure interface; and Based on the second business data, the first business data is obtained according to the decryption algorithm corresponding to the encryption algorithm and the decryption rules corresponding to the standard encryption protocol.

8. The method according to claim 1, characterized in that, The method further includes: The system provides early warnings for abnormal data; wherein, the abnormal data refers to business data from the multi-source business data that exceeds the confidence interval boundary; and The multi-source business data is subjected to multi-dimensional drill-down analysis to locate the abnormal business units corresponding to the abnormal data.

9. A business data prediction device based on multi-model fusion, characterized in that, The device includes: The statistical prediction module is used to predict incremental data for a target year based on multi-source business data using a Bayesian statistical model; wherein the Bayesian statistical model is trained and updated online based on historical business datasets. The quarterly forecasting module is used to obtain incremental data for the target quarter based on the multi-source business data and using a seasonal time-series forecasting model; wherein the seasonal time-series forecasting model is trained and updated online based on the historical business dataset; The period decomposition module is used to perform long-term period decomposition forecasting based on the incremental data of the target quarter, using the seasonal time-series forecasting model and the multinomial model, to obtain the initial forecast data for the target year and the fluctuation data for the remaining quarters; wherein, the target quarter and the remaining quarters constitute the target year; and The result correction module is used to use the incremental data of the target year as trend reference data and, based on the fluctuation data of the remaining quarters, correct the initial forecast data of the target year to obtain the business data forecast result.

10. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

Images