Operation method and device of financial institution and storage medium

Abstract

The invention discloses an operation method and device of a financial institution and a storage medium. The method comprises the following steps: determining N branch network sites operated by a target financial institution; based on the multi-dimensional feature data of each branch network, adjusting model parameters of the cash flow prediction model of each branch network to obtain a first model of each branch network; generating a global model based on the parameters of the first model of the N branch mechanism network points; obtaining a target model of each branch network based on the global model and the first model of each branch network; according to the multi-dimensional feature data of each branch network point and the target model of each branch network point, the cash outflow volume of each branch network point is predicted; and determining an operation strategy of the target financial institution according to the cash outflow volume of each branch website. According to the invention, the problem of low operation efficiency of financial institutions in related technologies is solved.

Description

Technical Field

[0001] This application relates to the field of financial technology, and more specifically, to an operating method, apparatus, and storage medium for a financial institution. Background Technology

[0002] In today's digital and rapidly changing financial environment, financial institutions face increasingly complex management challenges. Accurately forecasting cash inflows and outflows is crucial for ensuring smooth operations and efficient cash allocation. However, traditional cash forecasting methods and models often fall short in addressing the dynamics and complexities of modern banking, leading to operational inefficiencies.

[0003] Traditional forecasting models often rely on fixed time windows, such as a week, a month, or a quarter, to analyze historical cash flow data. However, this one-size-fits-all approach ignores the characteristics of cash flow within different business cycles, leading to significant discrepancies between forecasts and actual cash flows. Furthermore, early models typically used expert-defined weights or static weighting mechanisms when processing features, making it difficult to capture key influencing factors and thus reducing forecast accuracy. Therefore, existing forecasting techniques are insufficient in terms of flexibility and adaptability, failing to meet the high-efficiency and sophisticated cash management requirements of financial institutions.

[0004] There is currently no effective solution to the problem of low accuracy in cash flow forecasting in related technologies, which leads to low operational efficiency of financial institutions. Summary of the Invention

[0005] The main objective of this application is to provide an operating method, apparatus, and storage medium for financial institutions to address the problem of low accuracy in cash flow forecasting in related technologies, which leads to low operating efficiency of financial institutions.

[0006] To achieve the above objectives, according to one aspect of this application, an operational method for a financial institution is provided. The method includes: identifying N branch outlets operated by the target financial institution, where N is an integer greater than 1; adjusting the model parameters of a cash flow forecasting model for each branch outlet based on multi-dimensional feature data of each branch outlet to obtain a first model for each branch outlet; generating a global model based on the parameters of the first models of the N branch outlets; obtaining a target model for each branch outlet based on the global model and the first models of each branch outlet; predicting the cash outflow of each branch outlet based on the multi-dimensional feature data of each branch outlet and the target model of each branch outlet; and determining the operational strategy of the target financial institution based on the cash outflow of each branch outlet.

[0007] Optionally, before adjusting the model parameters of the cash flow forecasting model for each branch office based on the multi-dimensional feature data of each branch office to obtain the first model for each branch office, the method includes: determining M sets of feature vectors for different time periods based on the multi-dimensional feature data of each branch office, where M is an integer greater than 1; and performing weighted processing on the M sets of feature vectors based on a self-attention mechanism to obtain M weighted feature vector sets.

[0008] Optionally, the weighting of the M sets of feature vectors based on the self-attention mechanism to obtain M weighted sets of feature vectors includes: generating a query matrix, key matrix, and value matrix corresponding to each set of feature vectors; calculating the similarity between the query matrix and key matrix corresponding to each set of feature vectors; generating self-attention weights for each set of feature vectors based on the similarity between the query matrix and key matrix corresponding to each set of feature vectors using a normalized exponential function; and multiplying the self-attention weights of each set of feature vectors by the value matrix of each set of feature vectors to obtain M weighted sets of feature vectors.

[0009] Optionally, based on the multi-dimensional feature data of each branch office, the model parameters of the cash flow prediction model for each branch office are adjusted to obtain the first model for each branch office, which includes: taking the feature vector set from the M weighted feature vector sets as preset feature vector sets, and inputting the preset feature vector sets into the corresponding cash flow prediction models to predict the cash outflow corresponding to the preset feature vector sets; calculating the prediction score corresponding to the preset feature vector sets based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector sets; and adjusting the model parameters of the corresponding cash flow prediction models based on the prediction scores of the preset feature vector sets to obtain the adjusted cash flow. The prediction model is repeatedly executed, taking the feature vector set from the M weighted feature vector sets as the preset feature vector set and inputting the preset feature vector set into the corresponding cash flow prediction model. The cash flow prediction model predicts the cash outflow corresponding to the preset feature vector set. Based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set, the prediction score corresponding to the preset feature vector set is calculated. The model parameters of the corresponding cash flow prediction model are adjusted according to the prediction score of the preset feature vector set to obtain the adjusted cash flow prediction model. The process is repeated until the preset conditions are met. Based on the adjusted cash flow prediction model obtained when the preset conditions are met, the first model for each branch office is determined.

[0010] Optionally, generating a global model based on the parameters of the first model of N branch network points includes: determining an initial model; serializing the gradient of the parameters of each first model and encrypting the gradient of the parameters of each first model to obtain the encrypted gradients of N first models; and adjusting the parameters of the initial model based on the encrypted gradients of the N first models to obtain the global model.

[0011] Optionally, the target model for each branch office is obtained based on the global model and the first model for each branch office, including: constructing a global base tree based on the global model; determining the restricted subtree corresponding to each branch office based on the target feature vector set corresponding to the first model of each branch office; and fusing the restricted subtree corresponding to each branch office and the global base tree to obtain the target model for each branch office.

[0012] Optionally, after adjusting the first model of each branch office to obtain the target model of each branch office, the method includes: updating the parameters of the global model according to the target model of each branch office; adjusting the target model of each branch office according to the updated parameters of the global model; and obtaining the updated target model according to the multi-dimensional feature data of each branch office and the target model of each branch office.

[0013] To achieve the above objectives, according to another aspect of this application, an operating apparatus for a financial institution is provided. The apparatus includes: a first determining unit for determining N branch outlets operated by the target financial institution, where N is an integer greater than 1; an adjusting unit for adjusting the model parameters of the cash flow forecasting model for each branch outlet based on multi-dimensional feature data of each branch outlet, to obtain a first model for each branch outlet; a generating unit for generating a global model based on the parameters of the first models of the N branch outlets; a second determining unit for obtaining a target model for each branch outlet based on the global model and the first model of each branch outlet; a predicting unit for predicting the cash outflow of each branch outlet based on the multi-dimensional feature data of each branch outlet and the target model of each branch outlet; and a third determining unit for determining the operating strategy of the target financial institution based on the cash outflow of each branch outlet.

[0014] Optionally, the device further includes: a fourth determining unit, used to determine M sets of feature vectors for different time periods based on the multi-dimensional feature data of each branch office before adjusting the model parameters of the cash flow forecasting model of each branch office based on the multi-dimensional feature data of each branch office to obtain the first model of each branch office, wherein M is an integer greater than 1; and a processing unit, used to perform weighted processing on the M sets of feature vectors based on a self-attention mechanism to obtain M weighted feature vector sets.

[0015] Optionally, the processing unit includes: a first generation module for generating a query matrix, a key matrix, and a value matrix corresponding to each set of feature vectors; a first calculation module for calculating the similarity between the query matrix and the key matrix corresponding to each set of feature vectors; a second generation module for generating self-attention weights for each set of feature vectors based on the similarity between the query matrix and the key matrix corresponding to each set of feature vectors using a normalized exponential function; and a second calculation module for multiplying the self-attention weights of each set of feature vectors by the value matrix of each set of feature vectors to obtain M weighted feature vector sets.

[0016] Optionally, the second determining unit includes: a prediction module, used to take the feature vector sets from the M weighted feature vector sets as preset feature vector sets, and input the preset feature vector sets into the corresponding cash flow prediction models, and predict the cash outflow corresponding to the preset feature vector sets through the corresponding cash flow prediction models; a third calculation module, used to calculate the prediction score corresponding to the preset feature vector set based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set; a first adjustment module, used to adjust the model parameters of the corresponding cash flow prediction model based on the prediction score of the preset feature vector set, to obtain the adjusted cash flow prediction model; and a repetition module, used to repeatedly execute the calculation. The process involves taking the set of features from the M weighted feature vector sets as the preset feature vector set, inputting each preset feature vector set into the corresponding cash flow forecasting model, and using the corresponding cash flow forecasting model to predict the cash outflow corresponding to the preset feature vector set. Based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set, the prediction score corresponding to the preset feature vector set is calculated. The model parameters of the corresponding cash flow forecasting model are adjusted based on the prediction score of the preset feature vector set to obtain the adjusted cash flow forecasting model. This process continues until a preset condition is met. Based on the adjusted cash flow forecasting model obtained when the preset condition is met, the first model for each branch office is determined.

[0017] Optionally, the generation unit includes: a first determining module for determining an initial model; a processing module for serializing the gradients of the parameters of each first model and encrypting the gradients of the parameters of each first model to obtain N encrypted gradients of the first model; and a second adjusting module for adjusting the parameters of the initial model based on the N encrypted gradients of the first model to obtain a global model.

[0018] Optionally, the second determining unit includes: a construction module for constructing a global base tree based on a global model; a second determining module for determining a restricted subtree corresponding to each branch office based on the target feature vector set corresponding to the first model of each branch office; and a fusion module for fusing the restricted subtree and the global base tree corresponding to each branch office to obtain the target model of each branch office.

[0019] Optionally, the device further includes: a first updating unit, used to update the parameters of the global model according to the target model of each branch office after adjusting the first model of each branch office in the global model to obtain the target model of each branch office; an adjusting unit, used to adjust the target model of each branch office according to the updated parameters of the global model; and a second updating unit, used to obtain the updated target model according to the multi-dimensional feature data of each branch office and the target model of each branch office.

[0020] According to another aspect of this application, a computer-readable storage medium is provided, which includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform any kind of financial institution operation method.

[0021] According to another aspect of this application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include methods for performing operations of any financial institution.

[0022] According to another aspect of this application, a computer program product is provided, including computer instructions that, when executed by a processor, implement the steps of an operating method of a financial institution as described above.

[0023] In this embodiment, by identifying N branch outlets operated by the target financial institution, where N is an integer greater than 1; adjusting the model parameters of the cash flow prediction model for each branch outlet based on multi-dimensional feature data, a first model for each branch outlet is obtained; a global model is generated based on the parameters of the first models of the N branch outlets; a target model for each branch outlet is obtained based on the global model and the first model of each branch outlet; the cash outflow of each branch outlet is predicted based on the multi-dimensional feature data and the target model of each branch outlet; and the operating strategy of the target financial institution is determined based on the cash outflow of each branch outlet, thus solving the technical problem of low accuracy in cash flow prediction leading to low operating efficiency of financial institutions.

[0024] In this application, based on the multi-dimensional feature data of N branches of a target financial institution, the model parameters of the cash flow prediction model for each branch are adjusted to obtain a first model for each branch. A global model is generated based on each first model. The target model for each branch is further adjusted based on the global model and each first model. By combining the multi-dimensional feature data of each branch and the target model, the cash outflow of each branch can be predicted more accurately. Thus, the operating strategy of the target financial institution can be determined based on the cash outflow of each branch, achieving the technical effect of improving the operating efficiency of the financial institution. Attached Figure Description

[0025] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0026] Figure 1 A hardware structure block diagram of a computer terminal for implementing the operational methods of a financial institution is shown.

[0027] Figure 2 This is a flowchart of an operational method for a financial institution provided according to an embodiment of this application;

[0028] Figure 3 This is a schematic diagram of an operational method of a financial institution provided according to an embodiment of this application;

[0029] Figure 4 This is a schematic diagram of an operating device for a financial institution provided according to an embodiment of this application;

[0030] Figure 5 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] It should be noted that all relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties. For example, if there is an interface between this system and the relevant user or organization, before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain the relevant information after receiving consent information from the aforementioned user or organization.

[0034] Example 1

[0035] According to an embodiment of this application, a method embodiment for operating a financial institution is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0036] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing operational methods of financial institutions is shown. Figure 1As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0037] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0038] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the financial institution operation method in this embodiment of the application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned financial institution operation method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0039] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0040] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0041] Under the aforementioned operating environment, this application provides the following: Figure 2 The operating methods of financial institutions are shown. Figure 2 This is a flowchart of an operational method of a financial institution according to Embodiment 1 of this application.

[0042] Step S201: Determine the N branch outlets operated by the target financial institution, where N is an integer greater than 1.

[0043] Optionally, identifying and determining the branch outlets operated by the target financial institution can be done by filtering information on all operating branch outlets of the target financial institution from a large database or management system, or by filtering out multiple branch outlets for which an operational strategy needs to be determined.

[0044] Step S202: Based on the multi-dimensional feature data of each branch office, adjust the model parameters of the cash flow prediction model for each branch office to obtain the first model for each branch office.

[0045] Optionally, when predicting the cash outflow of each branch office, the degree of influence of different feature data on different branch offices varies. For example, the multi-dimensional feature data mentioned above is shown in Table 1 below, including feature data from multiple dimensions that affect the cash outflow of branch offices, such as historical cash flow data, time characteristics, business-related characteristics, and external environment characteristics. By adjusting the model parameters of the cash flow prediction model through multi-dimensional feature data, the resulting first model can better fit the actual situation of each branch office.

[0046] Optionally, a self-attention mechanism can be used to weight the multi-dimensional feature data, and the weighted multi-dimensional feature data can be used to train the cash flow prediction model for each branch, thereby obtaining a first model that is more in line with the actual situation of each branch network.

[0047] Table 1

[0048]

[0049] Step S203: Generate a global model based on the parameters of the first model of N branch network points.

[0050] Optionally, federated learning techniques can be employed to fuse the parameters of the first-level models from all branch offices, thereby constructing a global model that reflects the cash outflows of the entire target financial institution. The encrypted gradients of all first-level models can be collected by a central server or coordinator, then decrypted and their parameters updated to generate the global model.

[0051] Step S204: Based on the global model and the first model of each branch office / site, obtain the target model of each branch office / site.

[0052] Optionally, the generation of the target model for each branch office is a further personalized adjustment process to the first model. Each branch office can construct a global base tree based on the parameters of the global model, and generate a local restrictive subtree by combining the first model of each branch office, thereby obtaining the target model of each branch office, making the target model of each branch office more suitable for the specific business environment of each branch office.

[0053] Step S205: Based on the multi-dimensional feature data of each branch office and the target model of each branch office, predict the cash outflow of each branch office.

[0054] Optionally, by utilizing the target model for each branch office and combining it with multi-dimensional feature data of each branch office, the cash outflow of each branch office can be predicted. Compared to traditional forecasting methods, this target model-based forecasting will be more accurate and reliable, helping each branch office to adjust its cash reserves more precisely and improve cash utilization efficiency.

[0055] Step S206: Determine the target financial institution's operating strategy based on the cash outflow of each branch office.

[0056] Optionally, based on the above steps to predict the cash outflow of each branch, the target financial institution can formulate more reasonable operating strategies, such as cash reserve strategies, allocation of cash resources, and optimization of teller and self-service equipment operation arrangements.

[0057] For example, if a branch office is predicted to experience a cash shortage, the headquarters can arrange cash delivery in advance; if another branch office is predicted to have excessive cash reserves, the headquarters can collect the excess cash in a timely manner to reduce the cost of cash occupation.

[0058] The operational method for financial institutions provided in this application involves identifying N branch outlets operated by the target financial institution, where N is an integer greater than 1; adjusting the model parameters of the cash flow prediction model for each branch outlet based on multi-dimensional feature data, resulting in a first model for each branch outlet; generating a global model based on the parameters of the first models for the N branch outlets; obtaining a target model for each branch outlet based on the global model and the first models for each branch outlet; predicting the cash outflow of each branch outlet based on its multi-dimensional feature data and target model; and determining the operational strategy of the target financial institution based on the cash outflow of each branch outlet. This method solves the technical problem of low accuracy in cash flow prediction, leading to low operational efficiency of financial institutions. In this application, based on the multi-dimensional feature data of N branches of a target financial institution, the model parameters of the cash flow prediction model for each branch are adjusted to obtain a first model for each branch. A global model is generated based on each first model. The target model for each branch is further adjusted based on the global model and each first model. By combining the multi-dimensional feature data of each branch and the target model, the cash outflow of each branch can be predicted more accurately. Based on the cash outflow of each branch, the operating strategy of the target financial institution can be determined, thereby achieving the technical effect of improving the operating efficiency of the financial institution.

[0059] To focus on more important multi-dimensional feature data, optionally, in the financial institution operation method provided in this application embodiment, before adjusting the model parameters of the cash flow forecasting model for each branch based on the multi-dimensional feature data of each branch to obtain the first model for each branch, the method includes:

[0060] The first step is to determine M sets of feature vectors for different time periods based on the multi-dimensional feature data of each branch office, where M is an integer greater than 1.

[0061] Optionally, the aforementioned multi-dimensional feature data may include at least historical cash flow data (cash inflows and outflows over the past n days, historical cash outflows during the same period, volatility, etc.), time characteristics (such as holidays and working days), business-related indicators (such as payroll disbursements and pension payments), and external environmental characteristics (such as weather and the number of ATMs in the vicinity). By forming M sets of feature vectors for different time periods (e.g., the past 2 days, 7 days, 30 days, etc.), the impact of multi-dimensional feature data on cash outflows within different time periods can be analyzed, thereby improving the accuracy of predictions.

[0062] The second step involves weighting the M sets of feature vectors based on a self-attention mechanism to obtain M weighted feature vector sets.

[0063] Optionally, self-attention is a deep learning technique that focuses on certain parts of multi-dimensional feature data when processing multi-dimensional feature data, rather than processing all multi-dimensional feature data equally. In predicting cash outflows, self-attention can adjust the weights of different features, thereby improving prediction performance. Self-attention can assign a weight to each feature by calculating the correlation between features in an M-set feature vector set, thus focusing on features that have a greater impact on cash outflow prediction while reducing the influence of less important features.

[0064] In summary, by following the steps outlined above, branch offices can utilize the self-attention mechanism to capture key features that are more important for prediction from multi-dimensional feature data, thereby improving the accuracy and timeliness of predictions and enhancing the operational efficiency of financial institutions.

[0065] To focus on more important multi-dimensional feature data, optionally, in the financial institution operation method provided in this application embodiment, the M sets of feature vectors are weighted based on a self-attention mechanism to obtain M weighted feature vector sets, including:

[0066] The first step is to generate the query matrix, key matrix, and value matrix corresponding to each set of feature vectors.

[0067] For example, if the feature vector set has n days of historical data, then the feature matrix input to the self-attention mechanism is:

[0068] X = [X1, X2, X3…X] n ]∈R n*m

[0069] Where X1 represents all the feature vectors from the first day...X n X represents all eigenvectors on day n. n =[X n1 ,X n2 ,X n3 …Xnm ], where m is the number of feature data.

[0070] The feature matrix X will be transformed into a query matrix Q, a key matrix K, and a value matrix V, as follows:

[0071] Q = XW Q K = XW K V = XW V

[0072] Among them, W Q W k W v ∈R n*d This is the trained weight matrix. The attention dimension d can be set to be the same as m; therefore, the calculated Q, K, and V vectors are all R. n*m .

[0073] The second step is to calculate the similarity between the query matrix and the key matrix corresponding to each set of feature vectors.

[0074] Alternatively, the similarity between the query matrix and the key matrix corresponding to each set of feature vectors can be calculated using the following formula:

[0075]

[0076] The third step is to generate self-attention weights for each set of feature vectors based on the similarity between the query matrix and the key matrix corresponding to each set of feature vectors using a normalized exponential function.

[0077] Alternatively, the self-attention weights for each set of feature vectors can be generated using a normalized exponential function:

[0078] α = softmax (Attention Score)

[0079] The fourth step is to multiply the self-attention weights of each feature vector set by the value matrix of each feature vector set to obtain M weighted feature vector sets.

[0080] Optionally, the attention weights α are multiplied by the value matrix V to obtain the weighted set of feature vectors X′:

[0081] X ′ =αV

[0082] The weighted feature vector set X′ has its feature importance redistributed. By setting the attention dimension d to be the same as the number of features m, the weighted feature vector set X′ does not need to undergo subsequent feature dimensionality reduction.

[0083] In summary, the weighted feature vector set obtained through the above steps can be adaptively weighted based on feature data, thereby improving the predictive performance and adaptability of the cash flow forecasting model and optimizing the operational decisions of financial institutions.

[0084] To improve the prediction accuracy of the cash flow forecasting model for each branch office, optionally, in the financial institution operation method provided in this application embodiment, the model parameters of the cash flow forecasting model for each branch office are adjusted based on the multi-dimensional feature data of each branch office to obtain a first model for each branch office, including:

[0085] The first step is to take the feature vector set from the M weighted feature vector sets as the preset feature vector sets, and input the preset feature vector sets into the corresponding cash flow prediction models. The cash outflow corresponding to the preset feature vector sets is then predicted by the corresponding cash flow prediction models.

[0086] Optionally, a reinforcement learning algorithm can be used to select the target feature vector that minimizes the difference between the predicted future cash outflow and the actual cash outflow for each branch office, i.e., the prediction error is minimized. The parameters of the cash flow prediction model for each branch office can then be adjusted to obtain a first model with higher prediction accuracy. The weighted feature vector set X′ for each branch office contains multiple feature vector sets. The weighted feature vector set X′... ′ As a preset set of feature vectors, any one of the feature vectors is input into the cash flow forecasting model, and the cash flow forecasting model outputs the predicted cash flow based on the current feature vector.

[0087] The second step is to calculate the prediction score corresponding to the preset feature vector set based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set.

[0088] Optionally, the predictive ability of the model can be evaluated by calculating a prediction score using mean squared error (MSE), mean absolute error (MAE), or other performance metrics. For example, the prediction score can be calculated using the following formula:

[0089]

[0090] Among them, y t It is the actual cash outflow. This is the cash outflow predicted by the cash flow forecasting model. Since the ultimate goal of this embodiment is to reduce cash inventory, a penalty for excessive cash reserves is included. This represents the waste caused by overestimating the prediction, where λ is the weighting coefficient set for this purpose.

[0091] The third step is to adjust the model parameters of the corresponding cash flow prediction model based on the prediction scores of the preset feature vector set, so as to obtain the adjusted cash flow prediction model.

[0092] Optionally, based on the predicted scores obtained above, optimization algorithms such as gradient descent can be used to adjust the model parameters to reduce the prediction error.

[0093] The fourth step involves repeatedly taking the feature vector set from the M weighted feature vector sets as preset feature vector sets and inputting each preset feature vector set into the corresponding cash flow prediction model. The cash flow prediction model then predicts the cash outflow corresponding to the preset feature vector set. Based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set, a prediction score is calculated for the preset feature vector set. The model parameters of the corresponding cash flow prediction model are adjusted based on the prediction score of the preset feature vector set to obtain the adjusted cash flow prediction model. This process is repeated until a preset condition is met. Based on the adjusted cash flow prediction model obtained when the preset condition is met, the first model for each branch office is determined.

[0094] Optionally, the above steps can be automatically repeated until preset conditions are met, such as when the model parameters are adjusted and the prediction error reaches preset optimization criteria or convergence conditions.

[0095] In summary, through the above steps, the first model for each branch office can accurately predict future cash outflows, thereby providing financial institutions with a more reliable basis for operational decision-making.

[0096] To obtain the target model based on information from all branch offices, optionally, in the financial institution operation method provided in this application embodiment, generating a global model based on the parameters of a first model with N branch offices includes:

[0097] The first step is to determine the initial model.

[0098] Optionally, the initial model described above is the basic model used by each branch site and the central server before they begin joint training during the federated learning process.

[0099] The second step is to serialize the gradient of the parameters of each first model and encrypt the gradient of the parameters of each first model to obtain N encrypted gradients of the first model.

[0100] Optionally, serialization can convert complex objects (such as gradients of model parameters) into a format that can be directly stored or transmitted, such as text or a byte stream. Encryption can use cryptographic techniques to encode the data, ensuring that only those holding a specific decoding key can access the true content of the data, thus protecting data privacy and security. Each branch office serializes and encrypts the gradients of the parameters of the first model before sending them to the central server. This ensures secure data transmission while allowing the gradients to be used by the central server for updating the global model without exposing the original data.

[0101] The third step is to adjust the parameters of the initial model based on the encrypted gradients of the N first models to obtain the global model.

[0102] Optionally, the encrypted gradients of all branch network points are collected, the gradients are decrypted and fused, and then the parameters of the initial model can be updated based on these gradients to generate a global model.

[0103] For example, the central server verifies the integrity of the encrypted gradient parameters uploaded by each client (each branch office / site). Then, it applies a dynamic weighted aggregation algorithm, adjusting the weight coefficients based on the data quality assessment results of each branch office / site. During aggregation, the server uses homomorphic encryption to perform pre-decryption aggregation of the gradients, ensuring that the parameter information of individual clients is not exposed. Updating the parameters of the initial model can employ an improved federated averaging algorithm with a momentum term, while injecting Gaussian noise that conforms to differential privacy. The entire aggregation process can be completed in a trusted execution environment, ultimately generating a global model.

[0104] In summary, the global model obtained through the above steps reflects the comprehensive information of all branch offices and strictly protects the data privacy of each office.

[0105] To update the first model for each branch office based on information from all branch offices and obtain a target model with higher prediction accuracy, optionally, in the financial institution operation method provided in this application embodiment, obtaining the target model for each branch office based on the global model and the first model for each branch office includes:

[0106] The first step is to construct a global foundation tree based on the global model.

[0107] Optionally, the aforementioned global foundation tree is a basic decision tree structure built by the central server using the parameters of the global model within the federated learning framework. It serves as the basis for local model optimization by all branch offices, ensuring that the model of each office maintains consistency and standardization from a global perspective.

[0108] The second step is to determine the restricted subtree corresponding to each branch office based on the target feature vector set corresponding to the first model of each branch office.

[0109] Optionally, the aforementioned restrictive subtrees typically focus more on the details of each branch's data, such as specific customer withdrawal patterns or major events in the region where the branch is located. The target feature vector set for each branch contains the key features that best reflect that branch, possessing high representativeness and predictive value. Each branch determines its corresponding restrictive subtree based on the first model and the target feature vector set, reflecting the data characteristics and predictive needs of each branch.

[0110] The third step is to merge the restricted subtrees and the global base tree corresponding to each branch office / outlet to obtain the target model for each branch office / outlet.

[0111] Optionally, the restricted subtrees of each branch office can be merged into the global base tree, so that each branch office obtains a target model that includes both global trends and reflects local characteristics.

[0112] In summary, by combining the data characteristics and forecasting needs of each branch office with the comprehensive information of all branches office included in the global model through the above steps, the resulting target model not only possesses the generalization ability of the global model but also includes personalized adjustments for each branch office, further improving the forecasting accuracy of cash outflows for each branch office.

[0113] To continuously update the target model for each branch office and improve prediction accuracy, optionally, in the financial institution operation method provided in this application embodiment, after adjusting the first model for each branch office in the global model to obtain the target model for each branch office, the method further includes:

[0114] The first step is to update the parameters of the global model based on the target model of each branch office.

[0115] Optionally, after each branch office trains a localized target model based on its target feature vector set and the first model, the parameters of these models will continue to be fed back to the central server to further update the global model and generate new global model parameters.

[0116] The second step is to adjust the target model for each branch office based on the parameters of the updated global model.

[0117] Optionally, the updated global model parameters can be distributed to all branch offices via a central server. Each branch office will combine the updated global model parameters with its local target model to further update the target model parameters and optimize the prediction capabilities.

[0118] The third step is to obtain the updated target model based on the multi-dimensional feature data of each branch office and the target model of each branch office.

[0119] Optionally, each branch office can use the latest global model parameters and multi-dimensional feature data to continue training and adjusting the target model to ensure that the model can simultaneously reflect local characteristics and global trends.

[0120] For example, such as Figure 3 This is a schematic diagram illustrating the operational method of a financial institution according to an embodiment of this application. Branch outlet A and branch outlet B upload their local parameters (the latest model parameters) to a central server. The central server updates the parameters of the global model based on the local parameters of all branch outlets (branch outlet A and branch outlet B). The central server then distributes the updated global model parameters to branch outlet A and branch outlet B, which further update their model parameters. Specifically, each branch outlet can train the model based on feature combinations within different local time periods, and can also incorporate expert rule corrections to update the model parameters based on the prediction results. This update process is repeated to periodically check and improve the performance of the models at each branch outlet, ensuring that each branch outlet can predict more accurate cash outflows.

[0121] In summary, through continuous interaction between each branch office and the central server, and bidirectional updates of model parameters, the target model for each branch office is continuously optimized and evolved. In this way, the target model for each branch office not only captures local business characteristics and customer needs but also benefits from the generalization capabilities of global data, ensuring the accuracy, timeliness, and personalization of predictions.

[0122] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0123] Example 2

[0124] This application also provides an operating device for a financial institution. It should be noted that the operating device for a financial institution provided in this application can be used to execute the operating method for a financial institution provided in this application. The operating device for a financial institution provided in this application is described below.

[0125] According to embodiments of this application, an apparatus for implementing the aforementioned operational method of a financial institution is also provided, such as... Figure 4 As shown, the device includes: a first determining unit 401, an adjusting unit 402, a generating unit 403, a second determining unit 404, a predicting unit 405, and a third determining unit 406.

[0126] Specifically, the first determining unit 401 is used to determine N branch outlets operated by the target financial institution, where N is an integer greater than 1;

[0127] The adjustment unit 402 is used to adjust the model parameters of the cash flow forecasting model of each branch office based on the multi-dimensional feature data of each branch office, so as to obtain the first model of each branch office.

[0128] Generation unit 403 is used to generate a global model based on the parameters of the first model of N branch network points;

[0129] The second determining unit 404 is used to obtain the target model of each branch office based on the global model and the first model of each branch office.

[0130] Prediction unit 405 is used to predict the cash outflow of each branch based on the multi-dimensional feature data of each branch and the target model of each branch.

[0131] The third determining unit 406 is used to determine the operating strategy of the target financial institution based on the cash outflow of each branch outlet.

[0132] The financial institution operation device provided in this application embodiment determines N branch outlets operated by the target financial institution through a first determining unit 401, where N is an integer greater than 1; an adjusting unit 402 adjusts the model parameters of the cash flow prediction model of each branch outlet based on the multi-dimensional feature data of each branch outlet to obtain a first model for each branch outlet; a generating unit 403 generates a global model based on the parameters of the first models of the N branch outlets; a second determining unit 404 obtains a target model for each branch outlet based on the global model and the first model of each branch outlet; a prediction unit 405 predicts the cash outflow of each branch outlet based on the multi-dimensional feature data of each branch outlet and the target model of each branch outlet; and a third determining unit 406 determines the operating strategy of the target financial institution based on the cash outflow of each branch outlet. This solves the technical problem of low accuracy in cash flow prediction, which leads to low operating efficiency of financial institutions, and achieves the technical effect of improving the operating efficiency of financial institutions.

[0133] Optionally, in the financial institution operation device provided in the embodiments of this application, the device further includes: a fourth determining unit, used to determine M sets of feature vectors in different time periods based on the multi-dimensional feature data of each branch outlet before adjusting the model parameters of the cash flow prediction model of each branch outlet based on the multi-dimensional feature data of each branch outlet to obtain the first model of each branch outlet, wherein M is an integer greater than 1; and a processing unit, used to perform weighted processing on the M sets of feature vectors based on a self-attention mechanism to obtain M weighted feature vector sets.

[0134] Optionally, in the financial institution operating device provided in this application embodiment, the processing unit includes: a first generation module, used to generate a query matrix, a key matrix, and a value matrix corresponding to each set of feature vectors; a first calculation module, used to calculate the similarity between the query matrix and the key matrix corresponding to each set of feature vectors; a second generation module, used to generate self-attention weights for each set of feature vectors based on the similarity between the query matrix and the key matrix corresponding to each set of feature vectors using a normalized exponential function; and a second calculation module, used to multiply the self-attention weights of each set of feature vectors by the value matrix of each set of feature vectors to obtain M weighted feature vector sets.

[0135] Optionally, in the financial institution operating device provided in this application embodiment, the second determining unit 404 includes: a prediction module, used to take the feature vector sets in the M weighted feature vector sets as preset feature vector sets, and input the preset feature vector sets into the corresponding cash flow prediction models respectively, and predict the cash outflow corresponding to the preset feature vector sets through the corresponding cash flow prediction models; a third calculation module, used to calculate the prediction score corresponding to the preset feature vector sets based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector sets; and a first adjustment module, used to adjust the model parameters of the corresponding cash flow prediction model based on the prediction score of the preset feature vector sets, to obtain the adjusted cash flow prediction model; The repetition module is used to repeatedly execute the following steps: taking the feature vector set from the M weighted feature vector sets as a preset feature vector set, inputting the preset feature vector set into the corresponding cash flow prediction model, predicting the cash outflow corresponding to the preset feature vector set through the corresponding cash flow prediction model; calculating the prediction score corresponding to the preset feature vector set based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set; adjusting the model parameters of the corresponding cash flow prediction model based on the prediction score of the preset feature vector set to obtain the adjusted cash flow prediction model, until a preset condition is met; and determining the first model for each branch office based on the adjusted cash flow prediction model obtained when the preset condition is met.

[0136] Optionally, in the financial institution operation device provided in the embodiments of this application, the generation unit 403 includes: a first determining module, used to determine an initial model; a processing module, used to serialize the gradient of the parameters of each first model and encrypt the gradient of the parameters of each first model to obtain N encrypted gradients of the first model; and a second adjusting module, used to adjust the parameters of the initial model based on the N encrypted gradients of the first model to obtain a global model.

[0137] Optionally, in the financial institution operation device provided in this application embodiment, the second determining unit 404 includes: a construction module, used to construct a global base tree based on a global model; a second determining module, used to determine a restricted subtree corresponding to each branch office based on the target feature vector set corresponding to the first model of each branch office; and a fusion module, used to fuse the restricted subtree and the global base tree corresponding to each branch office to obtain a target model for each branch office.

[0138] Optionally, in the financial institution operation device provided in the embodiments of this application, the device further includes: a first updating unit, used to update the parameters of the global model according to the target model of each branch after adjusting the first model of each branch in the global model to obtain the target model of each branch; an adjusting unit, used to adjust the target model of each branch according to the updated parameters of the global model; and a second updating unit, used to obtain the updated target model according to the multi-dimensional feature data of each branch and the target model of each branch.

[0139] It should be noted that the first determining unit 401, the adjustment unit 402, the generation unit 403, the second determining unit 404, the prediction unit 405, and the third determining unit 406 mentioned above correspond to steps S201 to S206 in Embodiment 1. The six units and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0140] Example 3

[0141] Embodiments of this application may provide a computer terminal, which may be any computer terminal device in a group of computer terminals. Optionally, in this embodiment, the aforementioned computer terminal may also be replaced with a mobile terminal or an electronic device, etc.

[0142] Optionally, in this embodiment, the computer terminal may be located in at least one of a plurality of network devices in a computer network.

[0143] In this embodiment, the aforementioned computer terminal can execute program code for the following steps in the financial institution's operation method: determining N branch outlets operated by the target financial institution, where N is an integer greater than 1; adjusting the model parameters of the cash flow prediction model for each branch outlet based on multi-dimensional feature data of each branch outlet to obtain a first model for each branch outlet; generating a global model based on the parameters of the first models of the N branch outlets; obtaining a target model for each branch outlet based on the global model and the first model of each branch outlet; predicting the cash outflow of each branch outlet based on the multi-dimensional feature data of each branch outlet and the target model of each branch outlet; and determining the operation strategy of the target financial institution based on the cash outflow of each branch outlet.

[0144] Optionally, the aforementioned computer terminal can execute program code for the following steps in the financial institution's operational method: before adjusting the model parameters of the cash flow forecasting model for each branch based on the multi-dimensional feature data of each branch to obtain the first model for each branch, the method further includes: determining M sets of feature vectors for different time periods based on the multi-dimensional feature data of each branch, where M is an integer greater than 1; and performing weighted processing on the M sets of feature vectors based on a self-attention mechanism to obtain M weighted feature vector sets.

[0145] Optionally, the aforementioned computer terminal can execute program code for the following steps in the financial institution's operational method: weighting M sets of feature vectors based on a self-attention mechanism to obtain M weighted feature vector sets, including: generating a query matrix, key matrix, and value matrix corresponding to each set of feature vectors; calculating the similarity between the query matrix and key matrix corresponding to each set of feature vectors; generating self-attention weights for each set of feature vectors based on the similarity between the query matrix and key matrix corresponding to each set of feature vectors using a normalized exponential function; and multiplying the self-attention weights of each set of feature vectors by the value matrix of each set of feature vectors to obtain M weighted feature vector sets.

[0146] Optionally, the aforementioned computer terminal can execute program code for the following steps in the financial institution's operational method: Based on the multi-dimensional feature data of each branch office, adjusting the model parameters of the cash flow forecasting model for each branch office to obtain a first model for each branch office includes: taking the feature vector set from the M weighted feature vector sets as preset feature vector sets, and inputting the preset feature vector sets into the corresponding cash flow forecasting models; predicting the cash outflow corresponding to the preset feature vector set using the corresponding cash flow forecasting models; calculating the prediction score corresponding to the preset feature vector set based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set; and adjusting the corresponding cash flow forecasting model based on the prediction score of the preset feature vector set. The model parameters are used to obtain the adjusted cash flow forecasting model; the process of repeatedly taking the feature vector set from the M weighted feature vector sets as preset feature vector sets and inputting the preset feature vector sets into the corresponding cash flow forecasting models, and predicting the cash outflow corresponding to the preset feature vector sets through the corresponding cash flow forecasting models; calculating the prediction score corresponding to the preset feature vector sets based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector sets; adjusting the model parameters of the corresponding cash flow forecasting models based on the prediction scores of the preset feature vector sets to obtain the adjusted cash flow forecasting model, and repeating the process until preset conditions are met, and determining the first model for each branch office based on the adjusted cash flow forecasting model obtained when the preset conditions are met.

[0147] Optionally, the aforementioned computer terminal can execute program code for the following steps in the financial institution's operation method: generating a global model based on the parameters of a first model of N branch outlets includes: determining an initial model; serializing the gradient of the parameters of each first model and encrypting the gradient of the parameters of each first model to obtain encrypted gradients of N first models; adjusting the parameters of the initial model based on the encrypted gradients of the N first models to obtain the global model.

[0148] Optionally, the aforementioned computer terminal can execute program code for the following steps in the financial institution's operation method: obtaining the target model for each branch office based on the global model and the first model for each branch office includes: constructing a global base tree based on the global model; determining the restrictive subtree corresponding to each branch office based on the target feature vector set corresponding to the first model of each branch office; and fusing the restrictive subtree corresponding to each branch office and the global base tree to obtain the target model for each branch office.

[0149] Optionally, the aforementioned computer terminal can execute program code for the following steps in the financial institution's operational method: after adjusting the first model of each branch office in the global model to obtain the target model of each branch office, the method includes: updating the parameters of the global model according to the target model of each branch office; adjusting the target model of each branch office according to the updated parameters of the global model; and obtaining the updated target model according to the multi-dimensional feature data of each branch office and the target model of each branch office.

[0150] Optionally, Figure 5 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 5 As shown, the electronic device may include: one or more ( Figure 5 Only one of the components is shown: processor 502, memory 504, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module, and display.

[0151] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the financial institution operation method and apparatus in this application embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned financial institution operation method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0152] The processor can access the information and application programs stored in the memory via the transmission device to execute the aforementioned steps in the operating methods of the financial institutions.

[0153] This application provides an operational scheme for a financial institution. It identifies N branch offices of a target financial institution, where N is an integer greater than 1; adjusts the model parameters of the cash flow forecasting model for each branch office based on multi-dimensional feature data, resulting in a first model for each branch office; generates a global model based on the parameters of the first models for the N branch offices; obtains a target model for each branch office based on the global model and the first models for each branch office; predicts the cash outflow of each branch office based on its multi-dimensional feature data and target model; and determines the operational strategy of the target financial institution based on the cash outflow of each branch office. This solves the technical problem of low accuracy in cash flow forecasting, leading to low operational efficiency of financial institutions, and achieves the technical effect of improving the operational efficiency of financial institutions.

[0154] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 5 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5 The different configurations shown.

[0155] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0156] Example 4

[0157] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the financial institution's operating method provided in Embodiment 1.

[0158] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0159] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: identifying N branch outlets operated by the target financial institution, where N is an integer greater than 1; adjusting the model parameters of the cash flow forecasting model for each branch outlet based on multi-dimensional feature data of each branch outlet to obtain a first model for each branch outlet; generating a global model based on the parameters of the first models of the N branch outlets; obtaining a target model for each branch outlet based on the global model and the first model of each branch outlet; predicting the cash outflow of each branch outlet based on the multi-dimensional feature data of each branch outlet and the target model of each branch outlet; and determining the operating strategy of the target financial institution based on the cash outflow of each branch outlet.

[0160] Optionally, the storage medium is also configured to store program code for performing the following steps: before adjusting the model parameters of the cash flow forecasting model for each branch office based on the multi-dimensional feature data of each branch office to obtain the first model for each branch office, the method further includes: determining M sets of feature vectors for different time periods based on the multi-dimensional feature data of each branch office, where M is an integer greater than 1; and weighting the M sets of feature vectors based on a self-attention mechanism to obtain M weighted feature vector sets.

[0161] Optionally, the storage medium is also configured to store program code for performing the following steps: weighting M sets of feature vectors based on a self-attention mechanism to obtain M weighted sets of feature vectors, including: generating a query matrix, key matrix, and value matrix corresponding to each set of feature vectors; calculating the similarity between the query matrix and key matrix corresponding to each set of feature vectors; generating self-attention weights for each set of feature vectors based on the similarity between the query matrix and key matrix corresponding to each set of feature vectors using a normalized exponential function; and multiplying the self-attention weights of each set of feature vectors by the value matrix of each set of feature vectors to obtain M weighted sets of feature vectors.

[0162] Optionally, the storage medium is also configured to store program code for performing the following steps: adjusting the model parameters of the cash flow forecasting model for each branch office based on the multi-dimensional feature data of each branch office to obtain a first model for each branch office, including: taking the feature vector set from the M weighted feature vector sets as preset feature vector sets, and inputting the preset feature vector sets into the corresponding cash flow forecasting models respectively, and predicting the cash outflow corresponding to the preset feature vector set through the corresponding cash flow forecasting models; calculating the prediction score corresponding to the preset feature vector set based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set; and adjusting the model of the corresponding cash flow forecasting model based on the prediction score of the preset feature vector set. The parameters are used to obtain the adjusted cash flow forecasting model; the process of repeatedly taking the feature vector set from the M weighted feature vector sets as preset feature vector sets and inputting the preset feature vector sets into the corresponding cash flow forecasting models, and predicting the cash outflow corresponding to the preset feature vector sets through the corresponding cash flow forecasting models; calculating the prediction score corresponding to the preset feature vector sets based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector sets; adjusting the model parameters of the corresponding cash flow forecasting models based on the prediction scores of the preset feature vector sets to obtain the adjusted cash flow forecasting model, and repeating the process until preset conditions are met, and determining the first model for each branch office based on the adjusted cash flow forecasting model obtained when the preset conditions are met.

[0163] Optionally, the storage medium is also configured to store program code for performing the following steps: generating a global model based on the parameters of the first model of N branch network points, including: determining an initial model; serializing the gradient of the parameters of each first model and encrypting the gradient of the parameters of each first model to obtain encrypted gradients of N first models; adjusting the parameters of the initial model based on the encrypted gradients of the N first models to obtain the global model.

[0164] Optionally, the storage medium is also configured to store program code for performing the following steps: obtaining a target model for each branch office based on the global model and the first model of each branch office includes: constructing a global base tree based on the global model; determining a restricted subtree corresponding to each branch office based on the target feature vector set corresponding to the first model of each branch office; and fusing the restricted subtree corresponding to each branch office and the global base tree to obtain a target model for each branch office.

[0165] Optionally, the storage medium is also configured to store program code for performing the following steps: after the global model adjusts the first model of each branch office to obtain a target model for each branch office, the method includes: updating the parameters of the global model according to the target model of each branch office; adjusting the target model of each branch office according to the updated parameters of the global model; and obtaining an updated target model based on the multi-dimensional feature data of each branch office and the target model of each branch office.

[0166] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing the operational methods and steps of a financial institution.

[0167] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0168] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0169] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0170] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0171] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0172] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0173] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An operational method for a financial institution, characterized in that, include: Identify the N branch outlets operated by the target financial institution, where N is an integer greater than 1; Based on the multi-dimensional feature data of each branch office, the model parameters of the cash flow prediction model for each branch office are adjusted to obtain the first model for each branch office. A global model is generated based on the parameters of the first model with N branch offices and outlets. Based on the global model and the first model of each branch office, the target model of each branch office is obtained; Based on the multi-dimensional feature data of each branch and the target model of each branch, the cash outflow of each branch is predicted. The operational strategy of the target financial institution is determined based on the cash outflow of each branch office.

2. The method according to claim 1, characterized in that, Before adjusting the model parameters of the cash flow forecasting model for each branch office based on its multi-dimensional feature data to obtain the first model for each branch office, the method includes: Based on the multi-dimensional feature data of each branch office, determine M sets of feature vectors for different time periods, where M is an integer greater than 1; The M sets of feature vectors are weighted using a self-attention mechanism to obtain M weighted feature vector sets.

3. The method according to claim 2, characterized in that, Based on the self-attention mechanism, the M sets of feature vectors are weighted to obtain M weighted sets of feature vectors, including: Generate the query matrix, key matrix, and value matrix corresponding to each set of feature vectors; Calculate the similarity between the query matrix and the key matrix corresponding to each set of feature vectors; The self-attention weights for each set of feature vectors are generated by using a normalized exponential function based on the similarity between the query matrix and the key matrix corresponding to each set of feature vectors. Multiply the self-attention weights of each feature vector set by the value matrix of each feature vector set to obtain M weighted feature vector sets.

4. The method according to claim 2, characterized in that, Based on the multi-dimensional feature data of each branch office, the model parameters of the cash flow forecasting model for each branch office are adjusted to obtain the first model for each branch office, including: Each of the M weighted feature vector sets is taken as a preset feature vector set, and the preset feature vector set is input into the corresponding cash flow prediction model. The cash outflow corresponding to the preset feature vector set is predicted by the corresponding cash flow prediction model. Based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set, the prediction score corresponding to the preset feature vector set is calculated. The model parameters of the corresponding cash flow prediction model are adjusted according to the prediction scores of the preset feature vector set to obtain the adjusted cash flow prediction model. The process involves repeatedly taking the feature vector set from each of the M weighted feature vector sets as a preset feature vector set, inputting the preset feature vector set into the corresponding cash flow prediction model, and predicting the cash outflow corresponding to the preset feature vector set through the corresponding cash flow prediction model. Based on the actual cash outflow and the predicted cash outflow within the time period corresponding to the preset feature vector set, a prediction score is calculated for the preset feature vector set. The model parameters of the corresponding cash flow prediction model are adjusted based on the prediction score of the preset feature vector set to obtain the adjusted cash flow prediction model. This process is repeated until a preset condition is met. Based on the adjusted cash flow prediction model obtained when the preset condition is met, the first model for each branch office is determined.

5. The method according to claim 1, characterized in that, Based on the parameters of the first model with N branch office locations, a global model is generated, including: Determine the initial model; The gradients of the parameters of each first model are serialized and encrypted to obtain N encrypted gradients of the first model. Based on the gradients of the N encrypted first models, the parameters of the initial model are adjusted to obtain the global model.

6. The method according to claim 4, characterized in that, Based on the global model and the first model for each branch office / location, the target model for each branch office / location includes: Construct a global base tree based on the global model; Based on the set of target feature vectors corresponding to the first model of each branch office, determine the restricted subtree corresponding to each branch office. By merging the restricted subtree corresponding to each branch office and the global base tree, the target model of each branch office is obtained.

7. The method according to claim 1, characterized in that, After adjusting the first model of each branch office in the global model to obtain the target model of each branch office, the method includes: Update the parameters of the global model based on the target model of each branch office; Adjust the target model for each branch office based on the parameters of the updated global model; Based on the multi-dimensional feature data of each branch office and the target model of each branch office, the updated target model is obtained.

8. An operating device for a financial institution, characterized in that, include: The first determining unit is used to determine the N branch outlets operated by the target financial institution, where N is an integer greater than 1; The adjustment unit is used to adjust the model parameters of the cash flow forecasting model for each branch office based on the multi-dimensional feature data of each branch office, so as to obtain the first model for each branch office. The generation unit is used to generate a global model based on the parameters of the first model of N branch network points; The second determining unit is used to obtain the target model of each branch office based on the global model and the first model of each branch office. The prediction unit is used to predict the cash outflow of each branch office based on the multi-dimensional feature data of each branch office and the target model of each branch office. The third determining unit is used to determine the operating strategy of the target financial institution based on the cash outflow of each branch office.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the computer-readable storage medium is located to perform the operating method of a financial institution as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the operating method of the financial institution as described in any one of claims 1 to 7.

11. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the operating method of the financial institution as described in any one of claims 1 to 7.

Images