Business data configuration method, device, storage medium and electronic device

Abstract

This application discloses a method, apparatus, storage medium, and electronic device for configuring business data. Relating to the field of artificial intelligence, the method includes: acquiring data to be configured for a target institution, wherein the data to be configured includes at least one of the following: current business data, static business data, liquidity coverage ratio, and the rate of return of static business data; inputting the data to be configured into a target model to obtain a target adjustment amount for the data to be configured, wherein the target model simulates the environment of the target institution, and controls the interaction between the intelligent agent of the target model and the environment based on the adjustment amount, outputting the target adjustment amount that maximizes the return; adjusting the data to be configured according to the target adjustment amount to obtain the adjusted target data of the target institution. This application solves the problem in related technologies where the configuration strategy for business data is inaccurate due to rapid changes in the market environment.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically, to a method, apparatus, storage medium, and electronic device for configuring business data. Background Technology

[0002] In the daily operations of financial institutions, the allocation ratio of cash to short-term assets is a crucial component of asset and liability management. Financial institutions need to ensure they have sufficient liquidity to meet daily customer withdrawals, payment obligations, and the maturity of short-term liabilities, while also allocating additional funds for short-term investments to maximize returns. The allocation ratio of cash to short-term investment assets must meet relevant regulatory requirements; for example, the liquidity coverage ratio and net stable funding ratio of financial institutions must meet relevant regulations. These indicators are designed to ensure that financial institutions maintain sufficient high-quality liquid assets to address potential liquidity risks.

[0003] Financial institutions face challenges in managing short-term liquidity due to the uncertainty of funding needs. Factors such as customer withdrawal requests, market interest rate fluctuations, and loan repayment status all affect a financial institution's liquid assets, necessitating accurate forecasts of future funding demands and market conditions. Furthermore, financial institutions need to strike a balance between the safety of liquidity and investment returns; holding too many liquid assets can lead to losses, while excessive investment in short-term assets may expose them to liquidity risks.

[0004] Short-term asset liquidity management methods in related technologies are based on cash flow forecasting and maturity matching strategies using historical data. However, this approach cannot respond in real time to rapidly changing market and funding needs, especially during sudden liquidity demands or sharp market fluctuations. Financial institutions' allocation strategies between liquid and short-term assets are often inaccurate, making it difficult to cope with the uncertainty of liquidity needs and failing to maximize capital utilization efficiency.

[0005] Currently, no effective solution has been proposed to address the issue of inaccurate business data configuration strategies caused by rapid changes in the market environment in related technologies. Summary of the Invention

[0006] The main objective of this application is to provide a method, apparatus, storage medium, and electronic device for configuring business data, in order to solve the problem of inaccurate business data configuration strategies caused by rapid changes in the market environment in related technologies.

[0007] To achieve the above objectives, according to one aspect of this application, a method for configuring business data is provided. The method includes: acquiring data to be configured for a target institution, wherein the data to be configured includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the rate of return of static business data; inputting the data to be configured into a target model to obtain a target adjustment amount for the data to be configured, wherein the target model simulates the environment of the target institution, and controls the interaction between the agent of the target model and the environment based on the adjustment amount, outputting the target adjustment amount that maximizes the return; adjusting the data to be configured according to the target adjustment amount to obtain the adjusted target data for the target institution.

[0008] Optionally, the target model is obtained by: acquiring historical configuration data within a preset period; constructing an environment simulator based on the historical configuration data; and simulating the environment after each interaction between the agent and the environment using the environment simulator. The historical configuration data includes at least one of the following: historical market conditions, historical transaction records, and historical static business data. The state space, action space, reward function, and agent are determined. The state space includes state data after each interaction between the agent and the environment, and the state data includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, rate of return of static business data, and predicted expenditure business data. The action space includes a first adjustment amount for liquidity business data and a second adjustment amount for static business data. The agent includes a policy network and a value network. The agent interacts with the environment multiple times to obtain interaction data after each interaction. The interaction data includes data in the state space and action space after each interaction. The agent is iteratively trained multiple times based on the interaction data to obtain the target agent. Each iteration of training adjusts the parameters of the policy network and value network in the agent, and each iteration corresponds to one interaction between the agent and the environment. The target model is constructed using the environment simulator, state space, action space, reward function, and target agent.

[0009] Optionally, the environment simulator includes an input layer, a hidden layer, and an output layer, wherein the input layer is used to receive historical configuration data, the hidden layer is used to process the historical configuration data through a preset deep learning algorithm, and the output layer is used to output liquidity coverage ratio, yield of static business data, and predicted expenditure business data.

[0010] Optionally, determining the reward function includes: determining a liquidity adequacy reward based on the liquidity coverage ratio, determining an investment return reward based on the rate of return of static business data, and determining an excessive liquidity penalty based on liquidity business data; calculating the sum of the liquidity adequacy reward, investment return reward, and excessive liquidity penalty to obtain the reward function.

[0011] Optionally, the interaction data obtained by the agent interacting with the environment multiple times includes: for each interaction, acquiring the state data in the current state space of the environment, and inputting the state data into the policy network in the agent to obtain the first adjustment and the second adjustment output by the policy network; updating the state data in the state space based on the first adjustment and the second adjustment to obtain the updated state data; calculating the reward value of the reward function based on the updated state data, and determining the first adjustment, the second adjustment, the state data, the updated state data, and the reward value as the interaction data after the interaction.

[0012] Optionally, the agent is trained iteratively multiple times based on interaction data to obtain the target agent, including: for each iteration, determining a state-value function using the agent's value network and interaction data, wherein the state-value function is used to evaluate the estimated reward corresponding to the state data after each interaction; determining the agent's advantage function using the state-value function, wherein the advantage function is used to evaluate the advantage of the first adjustment and the second adjustment in each interaction; updating the network parameters of the agent's policy network using the advantage function, and updating the network parameters of the agent's value network using the estimated reward and reward value; stopping iterative training when the number of iterations is greater than or equal to a threshold, and determining the policy network and value network after the last iteration as the target agent.

[0013] Optionally, updating the network parameters of the agent's policy network through the advantage function and updating the network parameters of the agent's value network through the estimated reward and bonus values ​​includes: determining the pruning objective function of the policy network, inputting the advantage function into the pruning objective function, updating the network parameters of the policy network until the function value of the pruning objective function is at its maximum; inputting the estimated reward and bonus values ​​into the mean squared error formula, updating the network parameters of the value network until the output value of the mean squared error formula is at its minimum.

[0014] To achieve the above objectives, according to another aspect of this application, a business data configuration apparatus is provided. The apparatus includes: an acquisition unit for acquiring data to be configured for a target institution, wherein the data to be configured includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the rate of return of static business data; an input unit for inputting the data to be configured into a target model to obtain a target adjustment amount for the data to be configured, wherein the target model simulates the environment of the target institution, controls the interaction between the agent of the target model and the environment based on the adjustment amount, and outputs the target adjustment amount that maximizes the return; and an adjustment unit for adjusting the data to be configured according to the target adjustment amount to obtain the adjusted target data for the target institution.

[0015] In this embodiment, the method involves acquiring the target institution's configuration data, which includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the rate of return of static business data. The configuration data is input into a target model to obtain the target adjustment amount. The target model simulates the target institution's environment, and the agent of the target model interacts with the environment based on the adjustment amount, outputting the target adjustment amount that maximizes returns. The configuration data is adjusted according to the target adjustment amount to obtain the adjusted target data for the target institution. By training the target model using a deep reinforcement learning algorithm, simulating the market environment faced by the target institution, and learning the target adjustment amount that maximizes returns through the interaction between the agent and the environment, the configuration data is adjusted based on the target adjustment amount. This achieves a balance between the security and profitability of business data, optimizes the configuration strategy, and thus improves the accuracy of the business data configuration strategy. This solves the technical problem of inaccurate business data configuration strategies caused by rapid changes in the market environment. Attached Figure Description

[0016] 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:

[0017] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a configuration method for business data is shown.

[0018] Figure 2 This is a flowchart of a method for configuring business data according to an embodiment of this application;

[0019] Figure 3 This is a schematic diagram of an updated intelligent agent provided according to an embodiment of this application;

[0020] Figure 4 This is a schematic diagram illustrating an optional service data configuration method provided according to an embodiment of this application;

[0021] Figure 5 This is a schematic diagram of a configuration device for business data according to an embodiment of this application;

[0022] Figure 6 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0023] 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.

[0024] 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.

[0025] It should be noted that the information collected in this application (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.) are 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 this data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding access points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding access points to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0026] Example 1

[0027] According to an embodiment of this application, a method embodiment for configuring business data 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.

[0028] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1A hardware structure block diagram of a computer terminal (or mobile device) for implementing a configuration method for business data is shown. Figure 1 As 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, processing devices such as MCU (Microcontroller Unit) or FPGA (Field-Programmable Gate Array), memory 104 for storing data, and transmission device 106 for communication functions. In addition, it may also include: a display, input / output interfaces (I / O interfaces), a USB (Universal Serial Bus) port (which may be included as one of the ports of a BUS (Business 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.

[0029] 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).

[0030] 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 business data configuration method in this embodiment. 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 above-mentioned business data configuration 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.

[0031] 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.

[0032] The display may be, for example, a touchscreen LCD display that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0033] Under the above operating environment, Figure 2 This is a flowchart of a business data configuration method provided according to an embodiment of this application, such as... Figure 2 As shown, the method includes:

[0034] Step S201: Obtain the data to be configured for the target institution, wherein the data to be configured includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and yield of static business data.

[0035] In step S201, the target institution can be a financial institution. Liquidity data can be data that is freely available for spending or use without restriction, while static data can be data that is restricted from free spending or use within a certain period. In some optional examples, liquidity data can be the target institution's liquidity, such as cash positions, including customer deposits, withdrawals, loan repayments, and investment redemptions. Static data can be short-term invested assets, such as bonds, short-term held stocks, and funds.

[0036] The liquidity coverage ratio (LCR) is the ratio of a financial institution's high-quality liquid assets to its expected net cash outflows over the next 30 days, representing the institution's ability to meet short-term liquidity needs. The yield on static business data can be the average or a specific yield on a financial institution's short-term assets.

[0037] Step S202: Input the data to be configured into the target model to obtain the target adjustment amount of the data to be configured. The target model simulates the environment of the target mechanism. Based on the adjustment amount, the agent of the target model is controlled to interact with the environment and output the target adjustment amount that maximizes the benefit.

[0038] In step S202, the target model can be a machine learning model employing reinforcement learning algorithms. An environment simulator within the target model simulates the real business environment and market conditions of the target institution (such as a financial institution). The environment simulator can include: a state space, an action space, a reward function, and a transition model. The state space is defined by the data to be configured and can include liquid business data, static business data, liquidity coverage ratio, and the rate of return on static business data, etc. The action space represents the adjustment actions that the agent of the target model can take, such as increasing or decreasing liquid business data and static business data. The reward function is defined based on the target institution's profit maximization and liquidity management objectives, for example, using rewards for sufficient liquidity, investment returns, and penalties for excessive liquidity. The transition model describes the rules for state changes, i.e., how the state of the environment changes after the agent takes an action.

[0039] An agent is a learning entity in a reinforcement learning model, used to observe the state of the environment simulated by the model, take actions, and optimize its policy based on rewards from environmental feedback. Acquired configuration data is input as the initial state into the target model. The agent in the target model outputs a target adjustment based on its current state and its policy network, representing its action at the current time step. The environment simulates state changes based on the agent's actions and a built-in transition model, generating new states and immediate rewards. The agent receives feedback (rewards) from the environment and updates its policy (adjustment) according to the advantage function to achieve higher returns for future decisions in similar states.

[0040] The interaction between the agent and the environment is a multi-round iterative process. Each round includes the agent taking an action, the environment responding with a state change, and reward feedback. The agent's goal is to maximize the cumulative reward (i.e., maximize returns) over a long time frame, that is, to maximize investment returns and minimize the cost of holding liquidity while meeting liquidity coverage requirements. After a series of iterative training, the agent in the target model will learn how to adjust its policy network output to maximize the long-term cumulative reward within a given state space.

[0041] Step S203: Adjust the data to be configured according to the target adjustment amount to obtain the target data after the target mechanism is adjusted.

[0042] In step S203, after obtaining the target adjustment amount, the data to be configured is adjusted according to the target adjustment amount to obtain the target data. For example, the data to be configured includes mobile business data and static business data. The target adjustment amount includes a first adjustment amount for the mobile business data and a second adjustment amount for the static business data. The mobile business data is adjusted by the first adjustment amount, and the static business data is adjusted by the second adjustment amount, so that the target institution can maximize investment returns and reduce the retention cost of mobile business data while meeting the liquidity coverage ratio requirements.

[0043] In some examples, the target adjustment amount can be a vector or structured data representing recommendations for adjusting financial parameters such as a financial institution's cash reserves and short-term asset investments. Based on the cash reserve recommendations in the target adjustment amount, the financial institution needs to adjust its cash holdings. For example, if the agent recommends increasing cash reserves, the financial institution needs to redeem funds from short-term assets or raise cash through other channels to ensure sufficient liquidity to meet customer needs and address potential liquidity risks. Regarding short-term asset investments, the agent might recommend adjusting the portfolio to pursue higher returns or adjusting risk levels according to changes in market conditions. The financial institution needs to reallocate its portfolio based on the recommendations in the target adjustment amount, such as buying or selling specific short-term assets, to achieve a better asset allocation.

[0044] After adjusting cash and short-term asset allocations, the liquidity coverage ratio and the yield on short-term assets need to be recalculated to reflect changes in the actual business environment. The liquidity coverage ratio needs to ensure that financial institutions comply with the regulatory requirements of relevant authorities, while the updated yield reflects the expected return level of the target data.

[0045] The business data configuration method provided in this application involves acquiring the data to be configured for a target institution, which includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the rate of return of static business data. The data to be configured is input into a target model to obtain a target adjustment amount for the data to be configured. The target model simulates the environment of the target institution, and the agent of the target model interacts with the environment based on the adjustment amount, outputting the target adjustment amount that maximizes the return. The data to be configured is adjusted according to the target adjustment amount to obtain the adjusted target data of the target institution. By training the target model using a deep reinforcement learning algorithm to simulate the market environment faced by the target institution, and by learning the target adjustment amount that maximizes the return through the interaction between the agent and the environment, the method adjusts the data to be configured based on the target adjustment amount. This achieves a balance between the security and profitability of business data, optimizes the configuration strategy, and thus improves the accuracy of the business data configuration strategy. This solves the technical problem of inaccurate business data configuration strategies caused by rapid changes in the market environment.

[0046] To accurately predict the target adjustment amount of the data to be configured, a target model needs to be trained. Optionally, in the business data configuration method provided in this application embodiment, the target model is obtained in the following way: acquiring historical configuration data within a preset period, constructing an environment simulator based on the historical configuration data, and simulating the environment after each interaction between the agent and the environment through the environment simulator, wherein the historical configuration data includes at least one of the following: historical market conditions, historical transaction records, and historical static business data; determining the state space, action space, reward function, and agent, wherein the state space includes the state data after each interaction between the agent and the environment, and the state data includes at least one of the following: flowing business data, static business data, and flowing... The system comprises dynamic coverage, the rate of return on static business data, and predicted expenditure business data. The action space includes a first adjustment for dynamic business data and a second adjustment for static business data. The agent includes a policy network and a value network. The agent interacts with the environment multiple times to obtain interaction data after each interaction. This interaction data includes data in the state space and action space after each interaction. Based on the interaction data, the agent is iteratively trained multiple times to obtain the target agent. Each iteration adjusts the parameters of the policy network and value network in the agent, and each iteration corresponds to one interaction between the agent and the environment. The target model is constructed using an environment simulator, state space, action space, reward function, and the target agent.

[0047] In some examples, historical configuration data for a preset period is collected from the target institution's historical records. This historical configuration data may include historical market conditions (such as interest rates, stock market performance, etc.), historical transaction records (such as cash inflows and outflows, short-term asset trading records, etc.), and historical static business data (such as long-term investment portfolios, fixed costs, etc.). Based on the historical configuration data, an environment simulator is built. The environment simulator can reflect market changes, business operations, and capital flows. The environment simulator provides a virtual scenario for each interaction between the agent and the environment, enabling the agent to learn and optimize its strategies without interfering with actual business operations.

[0048] The state space defines the observable states of the agent at each decision point (adjustment), including but not limited to liquidity data, static data, liquidity coverage ratio, yield on static data, and projected expenditure data. These states describe the current financial condition and market environment of the financial institution. The action space defines the range of decisions the agent can make, including a first adjustment to liquidity data (such as an increase or decrease in cash reserves) and a second adjustment to static data (such as an adjustment to short-term asset investments). For example, the action space records actions using A... t Let A represent this. t ={ΔC t ∈[-ΔCmax ΔC max ], ΔI t ∈[-ΔI max ΔI max ]}, where ΔC max Expressed as the maximum adjustment to cash reserves, ΔI max This represents the maximum adjustment amount for short-term assets, with 0 indicating no adjustment. The agent will act in the environment based on these adjustments. The reward function is used to quantify the quality of the agent's decisions. The reward function can include rewards for sufficient liquidity (i.e., liquidity coverage ratio requirements), investment returns (i.e., returns on short-term assets), and penalties for excessive liquidity (i.e., the opportunity cost of excessive cash reserves), to guide the agent to balance liquidity risk and maximize returns.

[0049] An agent can include a policy network and a value network. The policy network is responsible for selecting the optimal action based on the current state, while the value network evaluates the long-term value of the state. In an environment simulator, the agent interacts with the environment multiple times, collecting data on the state, action, reward, and the next state after each interaction. This data constitutes the learning samples used to train and optimize the agent's policy. Using the collected interaction data, the agent undergoes multiple iterative training iterations through its policy and value networks, aiming to maximize the cumulative reward while limiting the magnitude of policy updates to ensure training stability and policy consistency.

[0050] After thorough training, the agent learns a policy network and a value network that can predict optimal actions and state values ​​based on current state data. Combining the environment simulator, state space, action space, reward function, and the trained target agent forms the final target model, which can output optimal cash and short-term asset allocation recommendations based on market conditions and financial institution business data.

[0051] This embodiment trains a target model that can simulate the capital allocation decisions of financial institutions under market fluctuations. The policy network learned by the agent in the model can guide financial institutions to adjust their cash reserves and short-term asset investments in real time or periodically to achieve the goals of sufficient liquidity, maximized returns, and reduced opportunity costs. The target model enables real-time and dynamic adjustment of cash and short-term asset allocation strategies, allowing financial institutions to make more accurate capital allocation decisions based on future market fluctuations and funding needs. This significantly improves the flexibility and adaptability of capital allocation.

[0052] In the target model, an environment simulator is needed to simulate the environment of the target organization. Optionally, in the business data configuration method provided in this application embodiment, the environment simulator includes an input layer, a hidden layer and an output layer. The input layer is used to receive historical configuration data, the hidden layer is used to process the historical configuration data through a preset deep learning algorithm, and the output layer is used to output liquidity coverage ratio, the rate of return of static business data and predicted expenditure business data.

[0053] In some examples, the input layer of the environment simulator receives historical configuration data and preprocesses it, such as normalization, missing value imputation, and outlier detection. The hidden layer can process the input data using pre-defined deep learning algorithms, including but not limited to convolutional neural networks, long short-term memory networks, gated recurrent units, or self-attention mechanisms. Features are extracted from the preprocessed data in the hidden layer, and a complex relationship is established between these features and future predictive metrics. The output layer generates three key predictive metrics: the liquidity coverage ratio, the yield on static business data, and predicted expenditure business data. The predicted liquidity coverage ratio helps financial institutions estimate future liquidity conditions and their ability to meet regulatory requirements. The predicted yield provides expectations about asset investment returns, guiding capital allocation decisions to maximize returns. The predicted expenditure business data helps financial institutions estimate future funding needs, such as customer withdrawals and loan repayments.

[0054] This embodiment simulates possible future changes in the environment using an environment simulator, providing the agent with a virtual environment to learn and optimize. This allows the agent to learn how to make optimal capital allocation decisions under different market conditions without actual market fluctuations, thereby achieving more efficient and robust liquidity management in practical applications. It significantly improves the efficiency and responsiveness of short-term liquidity management.

[0055] The procedure for evaluating the merits of each action of an agent using a reward function may optionally include, in the business data configuration method provided in this application embodiment, determining the reward function includes: determining a sufficient liquidity reward item based on the liquidity coverage ratio, determining an investment return reward item based on the rate of return of static business data, and determining an excessive liquidity penalty item based on the liquidity business data; calculating the sum of the sufficient liquidity reward item, the investment return reward item, and the excessive liquidity penalty item to obtain the reward function.

[0056] In some examples, the reward function can be calculated using the following formula:

[0057]

[0058] in, For rewards related to sufficient liquidity, As an investment return reward item, This is a penalty for excessive liquidity. The reward for sufficient liquidity can be expressed by the following formula:

[0059]

[0060] If the daily liquidity coverage ratio (LCR) t Meets the regulatory requirements of relevant agencies (set as LCR) target If liquidity requirements are met, a positive reward is given; otherwise, a penalty is imposed. Here, α1 is the positive reward for meeting liquidity requirements, and α2 is the penalty coefficient, which increases with the degree of liquidity insufficiency. The investment return reward can be expressed by the following formula:

[0061]

[0062] Where β is the reward coefficient for investment returns. Financial institutions generate daily returns from short-term asset investments, r. t Let I be the yield of a short-term asset on day t. t This represents the amount of short-term assets held on that day. Investment return rewards are proportional to daily returns. The excessive liquidity penalty can be expressed by the following formula:

[0063]

[0064] Where γ is the penalty coefficient for excess working capital; if C t ≤C max ,but While excessive cash reserves can improve liquidity, they also incur opportunity costs because they cannot generate investment returns. t Let C be the cash reserve amount on day t. If the cash reserve amount exceeds the upper limit C preset by the financial institution... max If so, then punishment needs to be imposed.

[0065] This embodiment evaluates the merits of each action selected by the agent by determining a reward function, thereby helping financial institutions to select more accurate allocation strategies in capital allocation.

[0066] To optimize the agent, it is necessary to acquire interaction data between the agent and the environment. Optionally, in the business data configuration method provided in this application embodiment, the interaction data obtained after each interaction through multiple interactions between the agent and the environment includes: for each interaction, acquiring the state data in the current state space of the environment, and inputting the state data into the policy network in the agent to obtain the first adjustment amount and the second adjustment amount output by the policy network; updating the state data in the state space based on the first adjustment amount and the second adjustment amount to obtain the updated state data; calculating the reward value of the reward function based on the updated state data, and determining the first adjustment amount, the second adjustment amount, the state data, the updated state data, and the reward value as the interaction data after the interaction.

[0067] In some examples, at the start of each interaction, the environment provides the agent with current state data. This state data reflects the target institution's financial condition and market conditions, and may include liquidity data, static data, liquidity coverage ratio, yield on static data, and projected expenditure data. The agent takes the received state data as input and passes it to its internal policy network. Based on the current state and the learned policy, the policy network outputs the action the agent should take: a first adjustment to the liquidity data and a second adjustment to the static data. Examples include specific operations such as increasing or decreasing cash reserves or adjusting short-term asset investments.

[0068] The actions taken by the agent affect the state of the environment; that is, the environment simulator updates its internal state space based on a first adjustment and a second adjustment, generating updated state data. The environment calculates the reward value of the reward function based on the agent's actions and the updated state data. The reward value reflects the impact of the action on the target institution's capital management efficiency and risk control, including rewards for sufficient liquidity, rewards for investment returns, and penalties for excessive cash holdings.

[0069] The interaction data after an interaction consists of the agent's first adjustment, second adjustment, the original state data of the environment, the updated state data, and the calculated reward value. This data is collected and stored by the agent for subsequent model training and policy optimization. The interaction between the agent and the environment is repeated over multiple time steps. The data from each interaction is accumulated to build an experience replay pool, which is the foundation for training deep reinforcement learning algorithms.

[0070] This embodiment uses the interaction data between the agent and the environment as training data to continuously learn and optimize the agent's capital allocation strategy. This enables the agent to make better decisions in a complex and ever-changing market environment, thereby optimizing the target organization's business data management.

[0071] After acquiring the interaction data, the agent is iteratively trained using the interaction data. Optionally, in the business data configuration method provided in this application embodiment, the agent is iteratively trained multiple times based on the interaction data to obtain the target agent, including: for each iteration of training, determining a state value function using the agent's value network and interaction data, wherein the state value function is used to evaluate the estimated reward corresponding to the state data after each interaction; determining the agent's advantage function using the state value function, wherein the advantage function is used to evaluate the advantage of the first adjustment and the second adjustment in each interaction; updating the network parameters of the agent's policy network using the advantage function, and updating the network parameters of the agent's value network using the estimated reward and reward value; stopping the iterative training if the number of iterations is greater than or equal to a threshold, and determining the policy network and value network after the last iteration of training as the target agent.

[0072] In some examples, the agent's value network is used to evaluate the state data in the collected interaction data, calculating the state value function, denoted as V(s). The state value function V(s) reflects the average future reward that the agent can expect to obtain in state s according to the current policy. The advantage function measures the difference between the reward of taking action a in state s and the average reward of taking any action in state s. The advantage function is calculated based on the state value function V(s) and the action value function Q(s,a). Q(s,a) is the future reward that the agent can expect to obtain according to the current policy after taking action a in state s.

[0073] The formula for calculating the dominance function can be:

[0074] A(s,a)=Q(s,a)-V(s);

[0075] Where A(s,a) represents the advantage of taking action a in state s.

[0076] The policy network is updated based on the advantage function A(s,a), using a gradient ascent method to optimize the probability distribution of actions taken by the agent. By adjusting the parameters of the policy network, the probability of taking a high-advantage action in a beneficial state is increased. This embodiment can use the pruning objective function from the PPO algorithm (Proximal Policy Optimization, an algorithm for optimizing reinforcement learning policies) to update the policy network. This function considers the ratio of new to old policies, limiting the magnitude of policy updates and ensuring the stability of training and the continuity of the policy.

[0077] The value network is updated based on estimated rewards and immediate reward values ​​from interaction data. Estimated rewards are derived from a dynamic evaluation of the agent's policy and the environment, reflecting the long-term gains the agent can expect after taking a series of actions. Mean squared error is used as the loss function to minimize the difference between the predicted and actual rewards, thus updating the value network's parameters. This helps the value network more accurately assess the long-term value of each state.

[0078] During iterative training, a threshold N can be set. When the number of iterations reaches or exceeds N, the training process stops. This threshold can be set experimentally to balance training efficiency and model performance. After the final iteration, the agent's policy network and value network parameters are updated, resulting in the target agent.

[0079] For example, Figure 3 This is a schematic diagram of an updated intelligent agent provided according to an embodiment of this application, such as... Figure 3 As shown, the system collects raw state data at each interaction between the bank and the market environment, as well as post-interaction state data and rewards. An actor network (i.e., a policy network) executes adjustments to cash and short-term asset reserves, while a critic network (i.e., a value network) evaluates the long-term value of each interaction. The PPO algorithm is used to update both the actor and critic networks. For the actor network, a policy pruning mechanism is introduced to limit the magnitude of change between the old and new policies, and the adjustment strategy for each interaction is updated by maximizing the advantage function. For the critic network, the parameters are optimized by minimizing the mean squared error between the estimated return and the reward value.

[0080] The PPO algorithm is used to automatically learn the optimal policy. This algorithm adjusts the policy update magnitude through a "pruning" method, controlling the size of each policy change. Specifically, PPO limits the magnitude of policy updates, ensuring that each update does not deviate too far from the original policy. This constraint avoids the problem of training instability caused by excessive policy changes. During each update, PPO adjusts the step size by comparing the policy before and after the update, ensuring that the model can robustly converge to the optimal solution.

[0081] This embodiment uses interactive data to iteratively train the agent. The agent's value network learns the ability to evaluate the value of a state, while the policy network learns the strategy to take the optimal action in different states. This process enables the agent to dynamically adjust cash reserves and short-term asset investments in a simulated financial institution environment based on state information such as current business data, static business data, liquidity coverage ratio, the rate of return on static business data, and predicted expenditure business data, to achieve the optimal balance between sufficient liquidity and maximizing investment returns. The resulting target agent can output optimal capital allocation decisions based on actual conditions, thereby improving the bank's capital management efficiency and risk control capabilities.

[0082] Optionally, in the business data configuration method provided in this application embodiment, updating the network parameters of the agent's policy network through the advantage function and updating the network parameters of the agent's value network through the estimated reward and bonus values ​​includes: determining the pruning objective function of the policy network, inputting the advantage function into the pruning objective function, updating the network parameters of the policy network until the function value of the pruning objective function is at its maximum value; inputting the estimated reward and bonus values ​​into the mean squared error formula, updating the network parameters of the value network until the output value of the mean squared error formula is at its minimum value.

[0083] In some examples, the pruning objective function can be expressed as follows:

[0084]

[0085] in, ε is the ratio of the old to the new policy, and ε is the clipping magnitude. `clip` represents the clipping function; the clipping objective function ensures that the policy network updates do not "jump" too much. A t Represents the dominance function. It expresses expectation.

[0086] By using gradient descent or other optimization algorithms, the value of the pruning objective function is maximized, and the parameters of the policy network are updated. This process can be accomplished using the backpropagation algorithm, adjusting the weights and biases in the policy network to maximize the value of the pruning objective function, thereby optimizing the decision-making ability of the policy network.

[0087] The value network is updated based on the estimated reward (i.e., an estimate of the state-value function) and the actual immediate reward received. The estimated reward can be obtained by weighting the immediate reward and the future discounted reward. The value network can use mean squared error as its loss function, which can be expressed by the following formula:

[0088] L VF (θ)=E t [(V θ (s t)-R t )] 2 ;

[0089] Among them, V θ (s t R is the value of the predicted state-value function, i.e., the estimated return. t This represents the actual reward for the current state. The parameters of the value network are updated by minimizing the mean squared error loss function using the gradient descent algorithm.

[0090] This embodiment, through updating the policy network and value network, enables the agent to gradually optimize its strategy in the environment through multiple iterations, achieving more efficient capital management and risk control. Minimizing the pruning objective function and mean squared error loss function ensures that the learning process of the policy network and value network converges quickly while remaining robust and continuous. This allows the agent to make more rational and optimized decisions in complex and ever-changing financial market environments.

[0091] According to another embodiment of this application, an optional method for configuring business data is also provided. Figure 4 This is a schematic diagram illustrating an optional service data configuration method provided in an embodiment of this application, such as... Figure 4 As shown, historical market conditions and bank liquidity management data are collected and preprocessed. A dynamic optimization model for bank cash and short-term asset allocation is established, and a reward function is constructed. The model is trained using the dataset, environment simulator, and reward function. The trained model is then applied to obtain the optimal short-term liquidity allocation scheme for the bank.

[0092] This embodiment, through an optional business data configuration method and combined with deep learning to predict market conditions, cash outflows, and short-term asset returns, can dynamically adjust cash and short-term asset allocation strategies in real time. This enables banks to make more accurate fund allocation decisions based on future market fluctuations and funding needs, significantly improving the flexibility and adaptability of fund allocation.

[0093] 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.

[0094] Example 2

[0095] This application also provides a business data configuration device. It should be noted that the business data configuration device of this application can be used to execute the business data configuration method provided in this application. The business data configuration device provided in this application will be described below.

[0096] According to an embodiment of this application, an apparatus for implementing the above-described method for configuring business data is also provided. Figure 5 This is a schematic diagram of a configuration device for business data provided in the embodiments of this application, such as... Figure 5 As shown, the device includes:

[0097] The acquisition unit 501 is used to acquire the data to be configured of the target institution, wherein the data to be configured includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and yield of static business data;

[0098] The input unit 502 is used to input the data to be configured into the target model to obtain the target adjustment amount of the data to be configured. The target model simulates the environment of the target mechanism, and controls the interaction between the intelligent agent of the target model and the environment based on the adjustment amount to output the target adjustment amount that maximizes the benefit.

[0099] The adjustment unit 503 is used to adjust the data to be configured according to the target adjustment amount to obtain the target data after the target mechanism is adjusted.

[0100] The business data configuration apparatus provided in this application embodiment acquires the data to be configured for a target institution through an acquisition unit 501. The data to be configured includes at least one of the following: current business data, static business data, liquidity coverage ratio, and the rate of return of static business data. An input unit 502 inputs the data to be configured into a target model to obtain the target adjustment amount of the data to be configured. The target model simulates the environment of the target institution and controls the interaction between the intelligent agent of the target model and the environment based on the adjustment amount to output the target adjustment amount that maximizes the return. An adjustment unit 503 adjusts the data to be configured according to the target adjustment amount to obtain the target data of the target institution after adjustment. By training the target model using a deep reinforcement learning algorithm, the target model simulates the market environment faced by the target institution. Through the interaction between the intelligent agent and the environment, the target adjustment amount that maximizes the return is learned. The data to be configured is adjusted based on the target adjustment amount, thereby achieving a balance between the security and profitability of business data, optimizing the configuration strategy, and thus realizing the technical effect of improving the accuracy of the business data configuration strategy. This solves the technical problem of inaccurate business data configuration strategies caused by rapid changes in the market environment.

[0101] Optionally, in the business data configuration device provided in this application embodiment, the device further includes: a first construction unit, configured to acquire historical configuration data within a preset period, construct an environment simulator based on the historical configuration data, and simulate the environment after each interaction between the agent and the environment through the environment simulator, wherein the historical configuration data includes at least one of the following: historical market conditions, historical transaction records, and historical static business data; and a first determination unit, configured to determine the state space, action space, reward function, and agent, wherein the state space includes state data after each interaction between the agent and the environment, and the state data includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, rate of return of static business data, and predicted expenditure. The system comprises: a business data unit, an action space including a first adjustment for flowing business data and a second adjustment for static business data; an agent including a policy network and a value network; an interaction unit, used to interact with the environment multiple times through the agent to obtain interaction data after each interaction, wherein the interaction data includes data in the state space and action space after each interaction; a training unit, used to perform multiple iterative training on the agent based on the interaction data to obtain the target agent, wherein each iteration of training adjusts the parameters of the policy network and value network in the agent, and each iteration of training corresponds to one interaction between the agent and the environment; and a second construction unit, used to construct the target model through an environment simulator, state space, action space, reward function, and target agent.

[0102] Optionally, in the business data configuration device provided in the embodiments of this application, the environment simulator includes an input layer, a hidden layer, and an output layer, wherein the input layer is used to receive historical configuration data, the hidden layer is used to process the historical configuration data through a preset deep learning algorithm, and the output layer is used to output liquidity coverage ratio, the rate of return of static business data, and predicted expenditure business data.

[0103] Optionally, in the business data configuration device provided in this application embodiment, the first determining unit includes: a first determining module, used to determine a liquidity adequacy reward item based on the liquidity coverage ratio, an investment return reward item based on the rate of return of static business data, and an excessive liquidity penalty item based on the liquidity business data; and a first calculation module, used to calculate the sum of the liquidity adequacy reward item, the investment return reward item, and the excessive liquidity penalty item to obtain a reward function.

[0104] Optionally, in the business data configuration device provided in this application embodiment, the interaction unit includes: an acquisition module, used to acquire state data in the current state space of the environment for each interaction, and input the state data into the policy network in the agent to obtain a first adjustment amount and a second adjustment amount output by the policy network; a first update module, used to update the state data in the state space based on the first adjustment amount and the second adjustment amount to obtain updated state data; and a second calculation module, used to calculate the reward value of the reward function based on the updated state data, and determine the first adjustment amount, the second adjustment amount, the state data, the updated state data, and the reward value as the interaction data after the interaction.

[0105] Optionally, in the business data configuration device provided in this application embodiment, the training unit includes: a second determining module, used to determine a state value function for each iteration of training through the agent's value network and interaction data, wherein the state value function is used to evaluate the estimated reward corresponding to the state data after each interaction; a third determining module, used to determine the agent's advantage function through the state value function, wherein the advantage function is used to evaluate the advantage of the first adjustment amount and the second adjustment amount in each interaction; a second updating module, used to update the network parameters of the agent's policy network through the advantage function, and update the network parameters of the agent's value network through the estimated reward and reward value; and a fourth determining module, used to stop iterative training when the number of iterations is greater than or equal to a threshold number, and determine the policy network and value network after the last iteration of training as the target agent.

[0106] Optionally, in the business data configuration device provided in this application embodiment, the second update module includes: a determination submodule, used to determine the pruning objective function of the strategy network, input the advantage function into the pruning objective function, and update the network parameters of the strategy network until the function value of the pruning objective function is the maximum value; and an input submodule, used to input the estimated return and reward value into the mean square error formula, and update the network parameters of the value network until the output value of the mean square error formula is the minimum value.

[0107] It should be noted that the acquisition unit 501, input unit 502, and adjustment unit 503 mentioned above correspond to steps S201 to S203 in Embodiment 1. The two modules 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 the device and run in the computer terminal 10 provided in Embodiment 1.

[0108] Example 3

[0109] Embodiments of this application may provide an electronic device. Figure 6 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (Only one is shown) processor 602, memory 604, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0110] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. 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.

[0111] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: acquiring the data to be configured for the target institution, wherein the data to be configured includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the rate of return of static business data; inputting the data to be configured into the target model to obtain the target adjustment amount of the data to be configured, wherein the target model simulates the environment of the target institution, controls the agent of the target model to interact with the environment based on the adjustment amount, and outputs the target adjustment amount that maximizes the return; adjusting the data to be configured according to the target adjustment amount to obtain the adjusted target data of the target institution.

[0112] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: acquiring historical configuration data within a preset period; constructing an environment simulator based on the historical configuration data; and simulating the environment after each interaction between the agent and the environment using the environment simulator. The historical configuration data includes at least one of the following: historical market conditions, historical transaction records, and historical static business data. The processor also determines the state space, action space, reward function, and agent. The state space includes state data after each interaction between the agent and the environment. The state data includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the returns from static business data. The system uses rate and predicted expenditure business data, with the action space including a first adjustment for flowing business data and a second adjustment for static business data. The agent includes a policy network and a value network. The agent interacts with the environment multiple times to obtain interaction data after each interaction, which includes data in the state space and action space after each interaction. Based on the interaction data, the agent is trained iteratively multiple times to obtain the target agent. Each iteration adjusts the parameters of the policy network and value network in the agent, and each iteration corresponds to one interaction between the agent and the environment. A target model is constructed using an environment simulator, state space, action space, reward function, and the target agent.

[0113] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: The environment simulator includes an input layer, a hidden layer and an output layer, wherein the input layer is used to receive historical configuration data, the hidden layer is used to process the historical configuration data through a preset deep learning algorithm, and the output layer is used to output liquidity coverage ratio, yield of static business data and predicted expenditure business data.

[0114] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: determine the liquidity adequacy reward based on the liquidity coverage ratio, determine the investment return reward based on the rate of return of static business data, and determine the excessive liquidity penalty based on the liquidity business data; calculate the sum of the liquidity adequacy reward, the investment return reward, and the excessive liquidity penalty to obtain the reward function.

[0115] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: for each interaction, obtain the state data in the current state space of the environment, and input the state data into the policy network in the agent to obtain the first adjustment and the second adjustment output by the policy network; update the state data in the state space based on the first adjustment and the second adjustment to obtain the updated state data; calculate the reward value of the reward function based on the updated state data, and determine the first adjustment, the second adjustment, the state data, the updated state data and the reward value as the interaction data after the interaction.

[0116] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: For each training iteration, determine a state value function using the agent's value network and interaction data, wherein the state value function is used to evaluate the estimated reward corresponding to the state data after each interaction; determine the agent's advantage function using the state value function, wherein the advantage function is used to evaluate the advantage of the first adjustment and the second adjustment in each interaction; update the network parameters of the agent's policy network using the advantage function, and update the network parameters of the agent's value network using the estimated reward and reward value; if the number of training iterations is greater than or equal to a threshold, stop the training iteration, and determine the policy network and value network after the last training iteration as the target agent.

[0117] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: determine the pruning objective function of the policy network, input the advantage function into the pruning objective function, update the network parameters of the policy network until the function value of the pruning objective function is at its maximum; input the estimated return and reward value into the mean squared error formula, update the network parameters of the value network until the output value of the mean squared error formula is at its minimum.

[0118] This application provides a method for obtaining configuration data for a target institution, wherein the configuration data includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the rate of return of static business data. The configuration data is input into a target model to obtain a target adjustment amount for the configuration data. The target model simulates the environment of the target institution, and the agent of the target model interacts with the environment based on the adjustment amount, outputting the target adjustment amount that maximizes the return. The configuration data is adjusted according to the target adjustment amount to obtain the adjusted target data of the target institution. By using a deep reinforcement learning algorithm to train the target model, simulating the market environment faced by the target institution, and through the interaction between the agent and the environment, the target adjustment amount that maximizes the return is learned. The configuration data is then adjusted based on the target adjustment amount, achieving a balance between the security and profitability of business data, optimizing the configuration strategy, and thus improving the accuracy of the configuration strategy for business data. This solves the technical problem of inaccurate configuration strategies for business data caused by rapid changes in the market environment.

[0119] Those skilled in the art will understand that Figure 6 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 6This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 6 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 6 The different configurations shown.

[0120] 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.

[0121] Example 4

[0122] 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 configuration method for business data provided in Embodiment 1.

[0123] 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.

[0124] This application also provides a computer program product, which, when executed on a data processing device, is suitable for performing steps of configuring business data.

[0125] 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.

[0126] 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.

[0127] 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.

[0128] 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.

[0129] 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.

[0130] 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.

[0131] 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. A method for configuring business data, characterized in that, include: Obtain the data to be configured for the target institution, wherein the data to be configured includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the yield of the static business data; The data to be configured is input into the target model to obtain the target adjustment amount of the data to be configured. The target model simulates the environment of the target mechanism. Based on the adjustment amount, the agent of the target model is controlled to interact with the environment and output the target adjustment amount that maximizes the benefit. Adjust the data to be configured according to the target adjustment amount to obtain the target data after the target mechanism is adjusted. The target model is obtained as follows: acquiring historical configuration data within a preset period, constructing an environment simulator based on the historical configuration data, and simulating the environment after each interaction between the agent and the environment through the environment simulator. The historical configuration data includes at least one of the following: historical market conditions, historical transaction records, and historical static business data. A state space, action space, reward function, and agent are determined. The state space includes state data after each interaction between the agent and the environment, and the state data includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, the rate of return of the static business data, and predicted expenditure business data. The action space includes the following: The system includes a first adjustment amount for the mobile business data and a second adjustment amount for the static business data. The agent comprises a policy network and a value network. The agent interacts with the environment multiple times to obtain interaction data after each interaction. This interaction data includes data in the state space and action space after each interaction. Based on this interaction data, the agent undergoes multiple iterative training iterations to obtain a target agent. Each iteration adjusts the parameters of the policy network and the value network within the agent, and each iteration corresponds to one interaction between the agent and the environment. The target model is constructed using the environment simulator, the state space, the action space, the reward function, and the target agent. Determining the reward function includes: determining a liquidity adequacy reward based on the liquidity coverage ratio, a return on investment reward based on the rate of return of static business data, and a penalty for excessive liquidity based on liquidity business data; and calculating the sum of the liquidity adequacy reward, the return on investment reward, and the penalty for excessive liquidity to obtain the reward function.

2. The method according to claim 1, characterized in that, The environment simulator includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the historical configuration data, the hidden layer is used to process the historical configuration data through a preset deep learning algorithm, and the output layer is used to output the liquidity coverage ratio, the rate of return of the static business data, and the predicted expenditure business data.

3. The method according to claim 1, characterized in that, The interaction data obtained after each interaction through multiple interactions between the intelligent agent and the environment includes: For each interaction, the state data in the current state space of the environment is obtained, and the state data is input into the policy network in the agent to obtain the first adjustment amount and the second adjustment amount output by the policy network; The state data in the state space is updated based on the first adjustment amount and the second adjustment amount to obtain the updated state data. The reward value of the reward function is calculated based on the updated state data, and the first adjustment amount, the second adjustment amount, the state data, the updated state data, and the reward value are determined as the interaction data after the interaction.

4. The method according to claim 1, characterized in that, Based on the interaction data, the agent is trained iteratively multiple times to obtain the target agent, which includes: For each iteration of training, a state value function is determined through the agent's value network and the interaction data, wherein the state value function is used to evaluate the estimated reward corresponding to the state data after each interaction. The agent's advantage function is determined by the state value function, wherein the advantage function is used to evaluate the advantage of the first adjustment and the second adjustment in each interaction. The network parameters of the agent's policy network are updated using the advantage function, and the network parameters of the agent's value network are updated using the estimated reward and bonus value. If the number of iterations is greater than or equal to the threshold, stop the iteration training and determine the policy network and value network after the last iteration training as the target agent.

5. The method according to claim 4, characterized in that, Updating the network parameters of the agent's policy network using the advantage function, and updating the network parameters of the agent's value network using the estimated reward and bonus value, includes: Determine the pruning objective function of the policy network, input the advantage function into the pruning objective function, and update the network parameters of the policy network until the function value of the pruning objective function reaches its maximum value; The estimated return and the reward value are input into the mean squared error formula to update the network parameters of the value network until the output value of the mean squared error formula is minimized.

6. A device for configuring business data, characterized in that, include: The acquisition unit is used to acquire the data to be configured of the target institution, wherein the data to be configured includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, and the rate of return of the static business data; An input unit is used to input the data to be configured into a target model to obtain a target adjustment amount of the data to be configured, wherein the target model simulates the environment of the target mechanism, controls the agent of the target model to interact with the environment based on the adjustment amount, and outputs a target adjustment amount that maximizes the benefit; An adjustment unit is used to adjust the data to be configured according to the target adjustment amount to obtain the target data after the target mechanism is adjusted. The device further includes: a first construction unit, configured to acquire historical configuration data within a preset period, construct an environment simulator based on the historical configuration data, and simulate the environment after each interaction between the agent and the environment through the environment simulator, wherein the historical configuration data includes at least one of the following: historical market conditions, historical transaction records, and historical static business data; and a first determination unit, configured to determine a state space, an action space, a reward function, and an agent, wherein the state space includes state data after each interaction between the agent and the environment, and the state data includes at least one of the following: liquidity business data, static business data, liquidity coverage ratio, the rate of return of the static business data, and predicted expenditure business data; and the action space includes the liquidity business data. The system includes a first adjustment amount and a second adjustment amount for the static business data. The agent comprises a policy network and a value network. An interaction unit is configured to interact with the environment multiple times through the agent to obtain interaction data after each interaction, wherein the interaction data includes data in the state space and the action space after each interaction. A training unit is configured to perform multiple iterative trainings on the agent based on the interaction data to obtain a target agent, wherein each iterative training adjusts the parameters of the policy network and the value network in the agent, and each iterative training corresponds to one interaction between the agent and the environment. A second construction unit is configured to construct the target model using the environment simulator, the state space, the action space, the reward function, and the target agent. The first determining unit includes: a first determining module, used to determine a liquidity adequacy reward item based on the liquidity coverage ratio, an investment return reward item based on the rate of return of static business data, and an excessive liquidity penalty item based on liquidity business data; and a first calculation module, used to calculate the sum of the liquidity adequacy reward item, the investment return reward item, and the excessive liquidity penalty item to obtain the reward function.

7. 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 configuration method for business data as described in any one of claims 1 to 5.

8. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program executes the configuration method for business data according to any one of claims 1 to 5 when it runs.

9. 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 business data configuration method according to any one of claims 1 to 5.

Images