Banking outlet intelligent site selection method and device and computer readable storage medium

Abstract

The invention provides a bank outlet intelligent site selection method and device and a computer readable storage medium. In the scheme, a training sample set related to a financial life scene is acquired, the training sample set comprises a plurality of training samples, each training sample comprises feature information and label information, the feature information comprises map information, personnel information, enterprise information and traffic information, and the label information is a result of site selection and scoring by adopting a preset standard; training the large language model by using the training sample set to obtain an initial site selection model; optimizing the initial site selection model by adopting a preset fine tuning strategy to obtain a target site selection model; and inputting the test sample set around the alternative address into the target site selection model for prediction, and generating a reasoning step by adopting a thinking chain to obtain a site selection decision result. According to the scheme, the problem of low efficiency of a bank outlet site selection method in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a method, device, computer-readable storage medium and electronic device for intelligent site selection of bank outlets. Background Art

[0002] As a window for business handling and customer service, bank branches play a decisive role in business development and are characterized by large investments and long time. With the development of cities and the popularization of online financial services, the distribution of professional financial resources among branches is uneven, and the busyness of counters at various branches varies greatly. Therefore, the location of bank branches is crucial.

[0003] The existing site selection methods have a single analysis method, strong subjectivity in decision-making, and lack of scientificity. The GIS (Geographic Information System) technology does not provide enough information for site selection, and the data selection requires high requirements on designers, and the method is not intelligent enough. The site selection evaluation system of big data modeling technology is not three-dimensional enough, the data utilization is not sufficient, the scalability is not strong, the degree of automation is not high enough, and the results still need to be screened. Therefore, a new method is needed to improve the efficiency of bank branch site selection. Summary of the invention

[0004] The main purpose of the present application is to provide a method, device, computer-readable storage medium and electronic device for intelligent bank branch site selection, so as to at least solve the problem of low efficiency of bank branch site selection methods in the prior art.

[0005] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for intelligent site selection of bank branches is provided, comprising: obtaining a training sample set related to financial life scenarios, the training sample set comprising multiple training samples, each of the training samples comprising feature information and label information, the feature information comprising map information, personnel information, enterprise information and traffic information, and the label information being the result of site selection and scoring using preset standards; using the training sample set to train a large language model to obtain an initial site selection model; using a preset fine-tuning strategy to optimize the initial site selection model to obtain a target site selection model; inputting a test sample set around the alternative address into the target site selection model for prediction, and using a thinking chain to generate reasoning steps to obtain a site selection decision result.

[0006] Optionally, the initial site selection model is optimized using a preset fine-tuning strategy, including at least one of the following: a supervised fine-tuning step, performing supervised training on the initial site selection model according to the fine-tuning task, and adjusting the parameters of the initial site selection model; a rejection sampling step, using the rejection sampling technique to exclude samples that are not within a preset standard range from the training sample set; a proximal strategy optimization step, using a proximal strategy optimization algorithm to adjust the strategy selection of the initial site selection model in the decision-making process; a direct strategy optimization step, using the direct strategy optimization technique to optimize the output result of the initial site selection model, and adjusting the parameter configuration of the initial site selection model in the output strategy.

[0007] Optionally, the initial site selection model is supervised trained according to the fine-tuning task, and the parameters of the initial site selection model are adjusted, including: setting the fine-tuning task according to the training sample set, the fine-tuning task including site selection predictions under different preset scenarios, the preset scenarios including financial life scenarios around residential areas, financial life scenarios around office buildings, financial life scenarios around commercial entities, and financial life scenarios around transportation hubs; inputting the fine-tuning task into the initial site selection model for prediction to obtain a prediction result, calculating the difference between the prediction result and the true label, and adjusting the parameters of the initial site selection model according to the difference.

[0008] Optionally, a test sample set around the alternative address is input into the target site selection model for prediction, and a chain of thought technique is used to generate reasoning steps to obtain a site selection decision result, including: setting a prediction task based on the test sample set, the prediction task including site selection based on cost and benefit; using the prediction task as the input of the target site selection model, performing site selection prediction through the target site selection model, and using chain of thought technique to generate reasoning steps related to the prediction task; analyzing and processing the reasoning steps to obtain the site selection decision result, the site selection decision result including the predicted site selection result of the target site selection model and an analysis report on the predicted site selection result.

[0009] Optionally, a training sample set related to financial life scenarios is obtained, the training sample set includes multiple training samples, each of the training samples includes feature information and label information, the feature information includes map information, personnel information, enterprise information and traffic information, and the label information is the result of site selection and scoring using preset standards, including: capturing the map information through an electronic map data interface, wherein the map information includes street layout, building distribution and location of transportation facilities; collecting the personnel information by combining automated survey plans and data collection, wherein the personnel information includes population size, age distribution, occupation type and working hours; extracting the enterprise information using data analysis tools, wherein the enterprise information includes enterprise business type, enterprise scale and enterprise working hours; obtaining the traffic information based on map data and traffic analysis algorithms, wherein the traffic information includes travel modes and commuting times.

[0010] Optionally, a training sample set related to financial life scenarios is obtained, the training sample set includes multiple training samples, each of the training samples includes feature information and label information, the feature information includes map information, personnel information, enterprise information and traffic information, and the label information is the result of site selection and scoring using preset standards, including: based on the acquired training sample set, a multi-dimensional evaluation is performed using the preset standards to obtain multiple pre-site selection results and scoring scores corresponding to the pre-site selection results; and the label information is generated based on the pre-site selection results and the scoring scores.

[0011] Optionally, a proximal strategy optimization algorithm is used to adjust the strategy selection of the initial site selection model in the decision-making process, including: a setting step, setting a strategy function, and using the output of the initial site selection model as a strategy parameter of the strategy function; an evaluation step, using the strategy parameters to generate a series of site selection decision samples, and evaluating the site selection decision samples, and generating a reward signal through a preset reward mechanism; an updating step, based on the site selection decision samples and the reward signal, adjusting the strategy parameters according to preset update rules to optimize the strategy selection of the initial site selection model in the decision-making process; repeating the evaluation step and the update step until the preset optimization standard is reached.

[0012] According to another aspect of the present application, there is provided an intelligent site selection device for bank branches, comprising: an acquisition unit, for acquiring a training sample set related to a financial life scenario, the training sample set comprising a plurality of training samples, each of the training samples comprising feature information and label information, the feature information comprising map information, personnel information, enterprise information and traffic information, and the label information being a result of site selection and scoring using preset standards; a training unit, for training a large language model using the training sample set to obtain an initial site selection model; an optimization unit, for optimizing the initial site selection model using a preset fine-tuning strategy to obtain a target site selection model; a prediction unit, for inputting a test sample set around an alternative address into the target site selection model for prediction, and using a thinking chain to generate reasoning steps to obtain a site selection decision result.

[0013] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods for intelligent site selection of bank outlets.

[0014] According to another aspect of the present 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 are configured to be executed by the one or more processors, and the one or more programs include means for executing any one of the methods for intelligent site selection for bank branches.

[0015] To apply the technical solution of the present application, first, a training sample set related to financial life scenarios is obtained, the training sample set includes multiple training samples, each training sample includes feature information and label information, the feature information includes map information, personnel information, enterprise information and traffic information, and the label information is the result of site selection and scoring using preset standards; then, the large language model is trained using the training sample set to obtain an initial site selection model; next, the initial site selection model is optimized using a preset fine-tuning strategy to obtain a target site selection model; finally, the test sample set around the alternative address is input into the target site selection model for prediction, and a thinking chain is used to generate reasoning steps to obtain a site selection decision result, thereby solving the problem of low efficiency of the bank branch site selection method in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings constituting part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1A hardware structure block diagram of a mobile terminal for executing a bank branch intelligent site selection method provided in an embodiment of the present application is shown;

[0018] Figure 2 A schematic diagram of a process of a bank branch intelligent site selection method provided according to an embodiment of the present application is shown;

[0019] Figure 3 A schematic diagram of a specific bank branch intelligent site selection method provided according to an embodiment of the present application is shown;

[0020] Figure 4 A schematic diagram of the construction process of a financial life simulation model of a specific bank branch intelligent site selection method provided in an embodiment of the present application is shown;

[0021] Figure 5 A processing flow chart of a financial life simulation model of a specific bank branch intelligent site selection method provided according to an embodiment of the present application is shown;

[0022] Figure 6 A data processing flow chart of a specific bank branch intelligent site selection method provided according to an embodiment of the present application is shown;

[0023] Figure 7 A schematic diagram of a thinking chain of a specific bank branch intelligent site selection method provided according to an embodiment of the present application is shown;

[0024] Figure 8 A flowchart of generating an analysis report of a specific bank branch intelligent site selection method provided according to an embodiment of the present application is shown;

[0025] Fig. 9 A structural block diagram of a bank branch intelligent site selection device provided according to an embodiment of the present application is shown.

[0026] The above drawings include the following reference numerals:

[0027] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0028] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0029] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] As introduced in the background technology, the site selection evaluation system of the big data modeling technology in the prior art does not make sufficient use of data and the degree of automation is not high enough. In order to solve the problem of low efficiency of the bank branch site selection method in the prior art, the embodiments of the present application provide a bank branch intelligent site selection method, device, computer-readable storage medium and electronic device.

[0032] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0033] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 FIG. 1 is a hardware structure block diagram of a mobile terminal of a bank branch intelligent site selection method according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0034] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0035] In this embodiment, a method for intelligent site selection of bank branches running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of 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 can be executed in an order different from that shown here.

[0036] Figure 2 FIG. 1 is a flow chart of a method for intelligently selecting a bank branch location according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0037] Step S201, obtaining a training sample set related to a financial life scenario, the training sample set including a plurality of training samples, each of the training samples including feature information and label information, the feature information including map information, personnel information, enterprise information and traffic information, the label information being the result of site selection and scoring using a preset standard;

[0038] Specifically, before starting model training, we first need to obtain data, namely the training sample set, which is closely related to the location decision of bank branches in financial life scenarios. Financial life scenarios can be understood as the environment and conditions around bank branches, including commercial areas, residential areas, etc., as well as the operation of bank branches and customer behavior in these scenarios.

[0039] The training sample set is a collection of multiple training samples, each of which consists of feature information and label information. Each sample represents the potential location of a bank branch in a specific location, as well as the environmental characteristics and evaluation results related to the location.

[0040] The label information is the site selection and scoring result of the training samples based on a set of preset evaluation criteria, including market potential, cost-benefit analysis, customer convenience, competition analysis, etc. The label information is the goal of model training, which helps the model learn how to judge and predict the site selection effect of bank outlets given the feature information.

[0041] In general, the process of obtaining a training sample set involves collecting detailed information about the environment around bank branches, namely feature information, and selecting and scoring these feature information according to the site selection evaluation criteria formulated by experts to obtain label information, thereby obtaining a training sample set containing feature information and label information, which will be used to train the large language model. This process ensures that the input data for model training is both comprehensive and targeted.

[0042] Step S202, using the above training sample set to train the large language model to obtain an initial site selection model;

[0043] Specifically, to train the large language model using the training sample set, the acquired training sample set must first be preprocessed, including data cleaning and standardization. The purpose of preprocessing is to ensure that the model can effectively learn from the data. The large language model is trained using the pre-trained training sample set. The training process aims to adjust the parameters of the large language model so that it can minimize the difference between the predicted results and the true labels. Through a large number of iterations, the large language model gradually learns how to make site selection decisions similar to expert evaluations based on feature information.

[0044] After training, the large language model can predict the location and score based on the input feature information, and obtain the initial location selection model, which can preliminarily understand the financial life scenarios and make location selection decisions.

[0045] Through step S202, using a large amount of data related to financial life scenarios, the initial site selection model obtained through training has learned how to predict the decision of bank branch site selection based on map information, personnel information, enterprise information and traffic information.

[0046] Step S203, optimizing the initial site selection model using a preset fine-tuning strategy to obtain a target site selection model;

[0047] Specifically, the initial site selection model obtained in step S202 already has the basic ability to predict bank branch site selection decisions based on feature information, but still needs to be further optimized to better meet specific business scenarios and goals.

[0048] The purpose of the preset fine-tuning strategy is to optimize the performance of the initial site selection model on a specific task. In this embodiment, a combination of multiple algorithms including multiple algorithms is used to fine-tune the initial site selection model to obtain a target site selection model. The target site selection model can more effectively understand the complexity of financial life scenarios, more accurately predict the potential and risks of different locations as bank outlets, and generate more reasonable site selection suggestions based on the input feature information.

[0049] Step S204, input the test sample set around the candidate address into the above-mentioned target site selection model for prediction, and use the thinking chain to generate reasoning steps to obtain the site selection decision result.

[0050] Specifically, the test sample set refers to a data set of environmental characteristics (such as population, traffic information, etc.) around the candidate address. These data are used to simulate the financial life scenarios of different candidate addresses so that the model can perform predictive analysis. The test sample set is ingested into the trained and optimized target site selection model, which comprehensively analyzes the potential benefits and social impact of the candidate address and outputs the prediction result, which is the result of site selection and scoring.

[0051] Thought chaining is a technique used in the field of artificial intelligence, especially in natural language processing and decision-making tasks, to improve the reasoning ability and explainability of models. Thought chaining originates from the human thinking process when solving complex problems, that is, through a series of intermediate steps or reasoning links, the final decision or answer is gradually derived. In the AI ​​model, the goal of thought chaining is to enable the model to not only give the correct prediction results, but also show the thinking process of how to derive the prediction results step by step from the input information.

[0052] Thinking chain can decompose complex reasoning processes and decompose complex tasks into a series of smaller and simpler subtasks. The model can solve these subtasks one by one, and finally integrate the outputs of all subtasks to form the final decision. While the model outputs the results, the thinking chain also generates a detailed reasoning process, which makes the model's decision-making process more transparent and understandable to the user. This is crucial for applications that require transparency and explainability, such as financial decisions, medical diagnosis, etc. By showing the complete reasoning chain, thinking chain technology can help identify potential errors or deviations in model predictions, allowing targeted adjustments and optimization of the model to improve the accuracy of decision-making. Traditional models go directly from input to output and lack intermediate steps. Thinking chain allows the model to learn and apply information in multi-step reasoning, which is very helpful for solving tasks that require multi-stage logical judgments.

[0053] In summary, thought chaining is a method of decomposing a complex decision-making process into a series of simple reasoning steps, which can help the model consider more dimensions of information when making decisions and form a more comprehensive judgment. In this embodiment, when the target site selection model predicts the candidate address, a sequence of reasoning steps is generated at the same time. The sequence of reasoning steps records in detail how the model derives the site selection decision result from the input data, including analysis of the characteristics of the input data, comparison with historical data, and the basis for strategy selection.

[0054] The specific location decision results are generated by combining the evaluation results of the target location model prediction output and the reasoning steps generated by the thought chain. The location decision results not only include the final recommended address of the model, but also come with detailed decision basis and explanation, making the decision process transparent and understandable, and facilitating manual review and iterative improvement of the results. For example, the target location model recommends a candidate address because it predicts large customer traffic, relatively low operating costs, and few surrounding competitors. These reasoning steps will be clearly displayed in the final location decision report.

[0055] Through the above series of steps, not only the efficiency and accuracy of bank branch site selection are improved, but also the transparency and explainability of the decision-making process are ensured, which helps to further optimize the site selection strategy and improve overall operational efficiency.

[0056] In the specific implementation process, a preset fine-tuning strategy is used to optimize the above-mentioned initial site selection model, including at least one of the following: a supervised fine-tuning step, in which the above-mentioned initial site selection model is supervised trained according to the fine-tuning task, and the parameters of the above-mentioned initial site selection model are adjusted; a rejection sampling step, in which the rejection sampling technology is used to exclude samples that are not within the preset standard range from the above-mentioned training sample set; a proximal strategy optimization step, in which the proximal strategy optimization algorithm is used to adjust the strategy selection of the above-mentioned initial site selection model in the decision-making process; a direct strategy optimization step, in which the direct strategy optimization technology is used to optimize the output result of the above-mentioned initial site selection model, and adjust the parameter configuration of the above-mentioned initial site selection model in the output strategy.

[0057] This embodiment uses a variety of technical means to fine-tune the initial site selection model. Supervised fine-tuning (SFT) introduces a labeled data set for a specific task and conducts further training to optimize the performance of the model on the task. Through supervised fine-tuning, the model can learn from specific labeled instances to improve prediction accuracy and decision quality. Among them, supervised training is performed on the above-mentioned initial site selection model according to the fine-tuning task, and the parameters of the above-mentioned initial site selection model are adjusted, including: setting the above-mentioned fine-tuning task according to the above-mentioned training sample set, and the above-mentioned fine-tuning task includes site selection prediction under different preset scenarios, and the above-mentioned preset scenarios include financial life scenarios around residential areas, financial life scenarios around office buildings, financial life scenarios around commercial entities, and financial life scenarios around transportation hubs; the above-mentioned fine-tuning task is input into the above-mentioned initial site selection model for prediction, and a prediction result is obtained, the difference between the above-mentioned prediction result and the true label is calculated, and the parameters of the above-mentioned initial site selection model are adjusted according to the above-mentioned difference.

[0058] Specifically, the setting of fine-tuning tasks requires a set of known instance data related to the location selection of bank branches, which includes detailed information of the candidate addresses and their actual performance after operation (real labels). The preset scenarios of financial life scenarios around residential areas, financial life scenarios around office buildings, financial life scenarios around commercial entities, and financial life scenarios around transportation hubs further clarify the scope of fine-tuning tasks. This embodiment sets four fine-tuning tasks, which can be set as the simulation prediction of the income brought to the branch by residents and branches for personal financial activities around residential areas; the prediction of personal and corporate financial business interactions between corporate employees and branches due to work and life needs around office buildings; the prediction of the income of corporate groups such as merchants and enterprises conducting public business with branches around commercial entities; and the prediction of business interactions between tourists, passers-by, and non-fixed personnel and branches at transportation hubs.

[0059] The set fine-tuning tasks are input into the initial site selection model. The initial site selection model predicts each task based on its pre-trained parameters and outputs loss values, evaluation indicators, learning rates, weights, gradients, training progress information, and deductive reasoning processes. Among them, the loss value reflects the gap between the model prediction results and the actual results, which is used to evaluate the accuracy of the site selection results; the evaluation indicators include accuracy, precision, recall, etc., which measure the performance of the entire model; the learning rate is the learning rate of the current training process; the weights, gradients, and training progress information are used to analyze the current model training progress and effect; the deductive reasoning process visualizes the model training and reasoning process to assist in dynamic monitoring.

[0060] The difference between the predicted result and the true label can be quantified by a loss function (such as mean square error, cross entropy loss, etc.). Through the back propagation algorithm, the parameters of the model are adjusted according to the gradient of the loss function to reduce the prediction error and improve the prediction accuracy of the model.

[0061] The above process constitutes a training iteration, and the initial site selection model continuously performs predictions and parameter adjustments on each fine-tuning task until the error between the prediction results and the true labels in all preset scenarios reaches the set optimization standard, or reaches the preset maximum number of iterations. This process not only optimizes the model's prediction ability in specific scenarios, but also enhances the model's generalization ability to different financial life scenarios.

[0062] By setting fine-tuning tasks covering a variety of preset scenarios and using supervised training to adjust the parameters of the initial site selection model, we can learn the key factors of site selection decisions through simulation and prediction of different scenarios, and apply this knowledge in future site selection tasks to improve the efficiency and effectiveness of branch operations. Supervised fine-tuning quantifies the prediction error and adjusts the parameters to ensure that the model can learn from historical data and continuously optimize its decision-making strategy, which is crucial to improving the accuracy and reliability of the model.

[0063] Rejection sampling is a data screening technique used to ensure the quality of training data, especially when the original dataset contains a lot of noise or irrelevant samples. In bank branch location selection, rejection sampling can help the model avoid learning unrepresentative samples, thereby improving the stability and accuracy of decision-making. Use rejection sampling techniques to exclude samples that are not within the preset criteria from the training sample set. By training with high-quality samples, the model can avoid learning wrong or irrelevant patterns, thereby improving its performance on real site selection tasks.

[0064] Proximal strategy optimization is a reinforcement learning algorithm that enables the model to make better choices in the decision-making process. In the site selection task, the proximal strategy optimization algorithm can help the model optimize the site selection strategy and improve the economic and social benefits of the decision. The proximal strategy optimization algorithm is used to adjust the strategy selection of the above-mentioned initial site selection model in the decision-making process, including: a setting step, setting a strategy function, and using the output of the above-mentioned initial site selection model as the strategy parameter of the above-mentioned strategy function; an evaluation step, using the above-mentioned strategy parameters to generate a series of site selection decision samples, and evaluating the above-mentioned site selection decision samples, and generating a reward signal through a preset reward mechanism; an update step, based on the above-mentioned site selection decision samples and the above-mentioned reward signal, adjusting the above-mentioned strategy parameters according to the preset update rules to optimize the strategy selection of the above-mentioned initial site selection model in the decision-making process; repeating the above-mentioned evaluation step and the above-mentioned update step until the preset optimization standard is reached.

[0065] Specifically, the use of the proximal policy optimization algorithm to adjust the strategy selection of the initial site selection model is a process of improving the model's decision-making ability by iteratively optimizing the policy parameters. First, the policy function is set. The policy function is the core component in the proximal policy optimization algorithm, which defines the probability of the model choosing different decisions (i.e., choosing a certain address as an outlet) under a given environmental state (such as the characteristics of a candidate address and its surrounding financial life scenes). The parameters of the policy function, i.e., the policy parameters, are initialized based on the output of the initial site selection model. This means that the model's preliminary prediction results will serve as the starting point for the policy function, which will use these parameters to guide the decision-making process.

[0066] The evaluation step involves using the current policy parameters to generate a series of location decision samples, which represent the model's location decisions under the current policy. Each generated decision sample is then evaluated based on a preset reward mechanism that quantifies the quality of the decision sample. For example, if a decision sample predicts a high branch operating efficiency, it will receive a higher reward signal. The reward signal reflects the expected performance of the decision sample in the preset environment (financial life scenario) and is key information to guide the adjustment of policy parameters.

[0067] In the update step, based on the site selection decision samples and the corresponding reward signals, the policy parameters are adjusted according to the preset update rules. The proximal policy optimization algorithm limits the update amplitude of the policy parameters through a method called "clipping target" to ensure that the new policy does not deviate too far from the old policy, thereby ensuring the stability and convergence of the model update. This process optimizes the policy parameters to make the model's policy selection in the decision-making process more optimized, which can generate higher reward signals, that is, better site selection decisions.

[0068] The whole process is an iterative optimization cycle. The evaluation step and the update step are repeated until the strategy selection reaches the preset optimization criterion. The preset optimization criterion can be a specific reward threshold, or after a certain number of iterations, the update amplitude of the strategy parameters is less than a preset threshold, indicating that the model's strategy selection has stabilized and optimized. By continuously generating decision samples, evaluating reward signals, and adjusting strategy parameters, the model can gradually learn the strategy for making the most optimal location decision in complex financial life scenarios.

[0069] In summary, the use of the proximal policy optimization algorithm to adjust the strategy selection of the initial site selection model is a process of converting the model prediction output into strategy parameters, generating decision samples, evaluating their performance under the preset reward mechanism, and adjusting the strategy parameters accordingly to optimize the model's decision-making ability. This algorithm emphasizes the iterative optimization of the strategy, and only makes small adjustments with each update, which ensures the stability and efficiency of the model learning, allowing the model to better adapt to the ever-changing financial life scenarios and improve the accuracy and efficiency of bank branch site selection decisions.

[0070] Direct strategy optimization is a reinforcement learning method that directly optimizes the model's strategy output. It improves the model's decision-making performance by directly modifying the model's parameter configuration on the output strategy. In bank branch site selection, direct strategy optimization can ensure that the model can take all relevant factors into account and make the most optimized decision when outputting site recommendations.

[0071] By combining fine-tuning strategies such as supervised fine-tuning, rejection sampling, proximal strategy optimization, and direct strategy optimization, the initial site selection model can more accurately evaluate the potential of candidate sites and make data-based decisions. These preset fine-tuning strategies not only improve the model's predictive ability, but also enhance its adaptability and generalization ability in complex financial life scenarios, thereby achieving better performance in site selection tasks.

[0072] The above preset fine-tuning strategy is used to optimize the initial site selection model to obtain the target site selection model, and the test sample set around the alternative address is input into the above target site selection model for prediction, and the thinking chain is used to generate reasoning steps to obtain the site selection decision result, including: setting a prediction task according to the above test sample set, and the above prediction task includes site selection based on cost and benefit; using the above prediction task as the input of the above target site selection model, performing site selection prediction through the above target site selection model, and using the thinking chain technology to generate reasoning steps related to the above prediction task; analyzing and processing the above reasoning steps to obtain the above site selection decision result, and the above site selection decision result includes the predicted site selection result of the above target site selection model and an analysis report on the above predicted site selection result.

[0073] Specifically, based on the test sample set around the candidate address, the prediction task is clarified. The prediction task is site selection prediction, with special attention to predicting the economic costs and benefits of site selection. This means that the target site selection model must not only be able to predict the potential benefits of each candidate address (such as customer traffic, business volume, operating income, etc.), but also be able to evaluate the economic costs of site selection (such as rent, construction cost, operating cost, etc.) to help banks make more reasonable and economical site selection decisions. The prediction task is used as a criterion for selecting outlets, including address selection based on economic benefits, social benefits and cost factors.

[0074] The set prediction task is input into the target site selection model, and the target site selection model makes site selection predictions for each candidate address based on the learned strategies and parameters. At the same time, the thinking chain begins to play a role. The thinking chain is a method that can decompose the decision-making process into a series of logical reasoning steps, which enables the model to generate detailed reasoning steps when making predictions, that is, how the model derives the prediction results from the input data. These reasoning steps include analysis of the costs and benefits of the candidate addresses, comparison with historical data, and strategy selection based on comprehensive evaluation of multiple factors.

[0075] Finally, the reasoning steps generated by the chain of thought are analyzed and processed to obtain the site selection decision results. The site selection decision results include both the model's predicted site selection results and a detailed explanation of the prediction process and results, that is, the analysis report. The analysis report will record in detail the model's decision-making basis, the cost and benefit evaluation process, the reasons for the site selection recommendation, and even the model's uncertainty and possible risk points in the decision-making process. This explanatory output not only provides recommended sites, but also provides sufficient background information and decision logic to help decision makers understand the model's recommendations and conduct manual review or adjustments when necessary.

[0076] In order to comprehensively and systematically obtain training sample sets related to financial life scenarios and provide rich data support for training an accurate site selection model, a training sample set related to financial life scenarios is obtained, the above training sample set includes multiple training samples, each of the above training samples includes feature information and label information, the above feature information includes map information, personnel information, enterprise information and traffic information, and the above label information is the result of site selection and scoring using preset standards, including: through the electronic map data interface, the above map information is captured, wherein the above map information includes street layout, building distribution and location of traffic facilities; the above personnel information is collected by combining automated survey plans and data collection, wherein the above personnel information includes population size, age distribution, occupation type and working hours; using data analysis tools to extract the above enterprise information, wherein the above enterprise information includes enterprise business type, enterprise scale and enterprise working hours; based on map data and traffic analysis algorithms, the above traffic information is obtained, wherein the above traffic information includes travel mode and commuting time.

[0077] Specifically, regarding the acquisition of map information, in this embodiment, through the connection with the electronic map data interface, map information can be automatically captured. The map information includes street layout, building distribution and transportation facility location, which can help the model understand the environment of the alternative address, such as the type of surrounding buildings (residential, office, commercial, etc.) and the distance and type of public transportation stations. These are key factors in evaluating the site selection of outlets.

[0078] The collection of personnel information adopts a combination of automated survey solutions and data collection, which involves obtaining data such as population size, age distribution, occupation type, and working hours around the candidate address. This information is crucial for predicting the potential customers and customer behavior of the outlets. For example, age distribution can help understand the target customer group, and occupation type and working hours can be used to predict customer traffic during non-working hours and weekdays, thereby evaluating the operating potential of the outlets.

[0079] Business information is extracted from multiple sources using data analysis tools, including business type, business size, and business hours. This information is critical to understanding the business environment and potential corporate business volume around the candidate address. Business type and size can reflect the economic vitality of the area, while business hours can help predict the business volume of the outlet during the weekday, especially the demand for corporate business.

[0080] Traffic information is obtained through map data and traffic analysis algorithms, including travel modes (public transportation, private cars, walking, etc.) and commuting time. The acquisition of traffic information can help the model evaluate the convenience of transportation and customer flow at the candidate address, especially during peak hours in the morning and evening and on weekends. Changes in traffic conditions have a significant impact on the business volume of outlets. By analyzing the distribution of transportation facilities, the rules of traffic flow, and the frequency of use of different travel modes, the model can more accurately predict the service coverage and potential customer groups of outlets.

[0081] The feature information (map information, personnel information, enterprise information and traffic information) obtained above and the label information together constitute the training sample set. The label information is the result of site selection and scoring according to the preset standards, which helps the model understand which feature combinations can bring higher operational benefits and social value. The label information is obtained through the following steps: Based on the above training sample set obtained, a multi-dimensional evaluation is performed using the above preset standards to obtain multiple pre-site selection results and the scoring scores corresponding to the pre-site selection results; the label information is generated based on the pre-site selection results and the scoring scores.

[0082] Specifically, based on the collected training sample set, each candidate address is evaluated in multiple dimensions using preset standards. The preset standards are based on standards including economic benefits, social benefits, operating costs and risk assessment. Among them, economic benefits can be reflected by evaluating the potential customer flow, operating income and return on investment of the candidate address; social benefits can be reflected by the contribution of the candidate address to the surrounding community, such as improving the coverage of financial services and promoting employment; operating costs can be reflected by analyzing the construction cost, rental cost and operating cost of the candidate address; regarding risk assessment, the financial risks of the candidate address can be evaluated, such as potential negative impacts such as market competition and economic fluctuations. Through the above preset standards, each sample is comprehensively scored to generate multiple site selection and scoring results, which reflect the site selection potential of the candidate address under different evaluation dimensions. The scoring score can be a comprehensive indicator calculated by the weighted average method, and the weight of each evaluation dimension is set according to the actual situation of the bank. For example, if the bank currently pays more attention to economic benefits, the weight of the economic benefit indicator is higher.

[0083] The pre-selected site results and scores obtained above are the label information, which is used to train the site selection model. Label information is the goal of model learning, allowing the model to understand which addresses are evaluated as better site selection options under given feature information, and the specific scores of these sites under the preset standards. In this way, the model can learn the relationship between feature information and preset standard evaluation results, so as to make more accurate site selection predictions on new data.

[0084] In short, by using preset standards to conduct multi-dimensional evaluation of the acquired feature information, pre-selection results and scoring scores are generated, and then these evaluation results are converted into label information required for training the model to build a site selection model based on historical data and actual operating experience. This process not only ensures the pertinence and effectiveness of model training, but also enables the model to comprehensively consider economic benefits, social benefits, operating costs, risk assessment and other factors when making new site selection decisions, and make more scientific and comprehensive site selection recommendations.

[0085] The embodiment of the intelligent site selection method for bank outlets of the present application realizes effective training and fine-tuning optimization of the large language model by comprehensively acquiring training sample sets closely related to financial life scenarios and adopting preset fine-tuning strategies. The target site selection model finally obtained can accurately predict the alternative addresses. At the same time, with the help of thinking chain technology, the decision-making reasoning steps are automatically generated, which significantly improves the scientificity, accuracy and transparency of the site selection decision, effectively guides the reasonable site selection of bank outlets to achieve the best economic and social benefits, greatly optimizes the traditional site selection process, and improves the efficiency of bank outlet site selection.

[0086] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the bank branch intelligent site selection method of the present application will be described in detail below in conjunction with specific embodiments.

[0087] This embodiment relates to a specific method for intelligent site selection of bank outlets, such as Figure 3 As shown, it includes the construction of a financial life simulation model (equivalent to the above-mentioned target site selection model), branch selection and branch operation benefit prediction.

[0088] The first step is to build a financial life simulation model. There are three main steps to build this model: Figure 4 As shown, it includes selecting a base model, data preparation, and model fine-tuning, and finally forming a large model that is suitable for the current task. Among them, the location selection principle of the model base is to support Chinese, be lightweight, have strong contextual connection ability, and have strong adaptability to specific tasks. The model is based on the Transformer architecture and chooses to use the Decoder-only method, which is more adaptable to training data and has lower computing power requirements. The model puts the normalization layer in front to improve the stability and efficiency of the model training process. The self-attention layer uses the group query self-attention method. For the processing flow chart, see Figure 5. The data processing process starts with the input link, which includes various information closely related to financial life scenarios, such as map data, demographic data, corporate information, traffic flow information, etc. These input data are the basis for the model to make predictions and analyses. RMSNorm (RMSNorm LayerNormalization) is a layer used to normalize input data in deep learning models. It aims to control the range of activation values ​​within the model, avoid gradient vanishing or exploding problems, and improve the stability and efficiency of training. RMSNorm performs normalization by calculating the root mean square value of the input. It is particularly effective for processing the input of the Attention layer because it can better adapt to input data of different scales. The self-attention mechanism is a core component of the Transformer architecture, allowing the model to pay attention to the importance of different parts of the input sequence, so that it can focus more on key information when processing variable-length sequence data. In the financial life simulation model, the self-attention layer is used to capture the correlation and dependency between input data, providing a more detailed and complex contextual understanding for subsequent predictions. After the Attention layer is processed, the data is normalized again by the RMS Norm layer to ensure that the data remains stable when flowing through the multi-layer neural network, avoiding the influence of the model's learning effect due to the drastic changes in the data range. MLP (Multilayer Perceptron) is a fully connected neural network composed of multiple layers of linear and nonlinear transformation layers. In the model, MLP is used to further process and transform the features processed by the previous layers, and transform the high-dimensional and complex feature information into the prediction results of the final output of the model. MLP multi-layer perceptron can learn nonlinear data relationships, which is crucial to improving the prediction accuracy and generalization ability of the model. Based on the input feature information and the processed intermediate results, the model generates prediction results for financial life scenarios, such as the recommendation of branch location selection and the prediction of operating benefits.

[0089] After the base is built, it is necessary to prepare predictions for training. The "prediction" here refers to the pre-processed data set prepared for model training, which is the basis for model learning and fine-tuning. Predictions are mainly collected and processed manually, and the content is mainly related to financial life scenarios, including map information, population size, basic population information (age, work content, working hours, etc.), traffic information, and corporate information (business type, scale, working hours, etc.). Banking business is now very extensive, and the operation of outlets is easily affected by many factors. Therefore, models are constructed from multiple dimensions such as population, enterprise, and transportation to analyze the impact of operations in an all-round and three-dimensional manner. Data processing mainly consists of three steps: data capture, data processing, and data mixing. See Figure 6First, data is captured to capture map information, personnel information, enterprise information and traffic information. Map information is collected in a fine-grained manner through electronic maps; personnel information is collected by combining automated research programs with data collection; enterprise information is obtained by using data analysis tools; and traffic information is obtained based on map data and traffic analysis algorithms.

[0090] After the data is captured, it needs to be processed. Since the collected data is relatively rough, it needs to be processed initially to obtain predictions related to financial life, and different expected results are obtained according to task division to form a data set. In order to ensure that the data distribution conforms to the state of the real world, the data needs to be mixed. This embodiment uses the Monte Carlo algorithm for sampling, and performs sampling cross-section on the above map information, personnel information, enterprise information and traffic information.

[0091] This embodiment adopts a combined fine-tuning algorithm of supervised fine-tuning (SFT), rejection sampling, proximal strategy optimization and direct strategy optimization, and uses the processed data as a fine-tuning task to train the model, thereby improving the performance of the model after alignment. Among them, the fine-tuning task content is divided according to the scene, specifically divided into: Fine-tuning task 1: around residential areas, residents and outlets conduct personal financial activities, and simulate the prediction of the income brought to the outlets; fine-tuning task 2: around office buildings, the prediction of personal and corporate financial business interactions between corporate employees and outlets due to work and life needs; fine-tuning task 3: around commercial entities, merchants, enterprises and other corporate groups, and outlets conduct public business income prediction; fine-tuning task 4: transportation hubs, tourists, passers-by, non-fixed personnel, and outlets conduct business interactions.

[0092] Based on the real environment around the candidate address, the intelligent site selection decision is completed by analyzing the economic benefits and social benefits generated by the site selection. The model training task is difficult, and the design that conforms to the downstream tasks of the large language model can improve the model accuracy. In addition, different analysis reports can be generated according to different prediction tasks to assist users in understanding and judgment. Prediction tasks can be arranged and combined to make judgments with different focuses according to user needs.

[0093] This embodiment uses a combination of big model and thought chain technology to achieve the site selection and analysis report generation of network points. The big model generates reasoning steps through thought chain during prediction. This step is the analysis process of network point selection. The intermediate results are sorted out to form the final analysis report. At the same time, the reasoning process is output as the selection criteria to form experience and manual evaluation criteria.

[0094] like Figure 7As shown in the figure, the thinking chain is a method of decomposing a complex problem into multiple simple sub-problems. The decomposition process is the problem analysis process. Through directional stimulation prompts, users can provide more requirements to help judge the results. The prediction task is the standard for predicting and analyzing the results. Presetting the prediction task in advance can have a directional impact on the thinking chain and the generated results, and guide the problem to the final result. The realization of the prediction task is mainly reflected in the form of parameters, which is reflected in two parts of work. First, as part of the data set preparation, the data preparation adopts the model processing method, and the prediction task is used as one of the data set training standards. Secondly, the prediction task is used in the thinking chain as part of the problem decomposition and generation prompts, which affects the feedback and decision-making. Among them, the prediction task is set as the standard for selecting outlets: Prediction task 1: address selection based on economic benefits; Prediction task 2: address selection based on social benefits; Prediction task 3: address selection based on cost factors.

[0095] See also Figure 8 The problem decomposition process of the thinking chain will generate decision information, which will be recorded in text form and combined with the prediction task to form a text result through the large model, and at the same time generate a prediction of the network operation efficiency. The report is formed by inputting the preset prediction task, decision information and model feedback, generating a text model, and finally forming a decision report of a specified style, which will analyze the prediction results one by one and output an analysis report.

[0096] The following is a specific example based on the site selection of a bank:

[0097] A bank plans to set up a new branch in a newly developed commercial area, which includes multiple residential areas, office buildings, shopping centers and transportation hubs. The bank needs to evaluate the comprehensive operational benefits of different locations to determine the best branch location. The bank branch intelligent site selection method of this application is used to achieve this goal.

[0098] First, we obtain detailed map data of the commercial area, including roads, buildings, residential areas, commercial facilities, etc.; through research and data analysis, we obtain the population density, age distribution, occupation type, working hours, etc. of the area; collect corporate information of office buildings and commercial entities, including corporate type, scale, industry, working hours, etc.; analyze the traffic flow, distribution of public transportation stations, traffic congestion, etc. in the area. We clean, annotate and format the collected data to ensure data quality and consistency, and form a format suitable for model training.

[0099] A large language model based on the Transformer architecture that supports Chinese processing, is lightweight, and has strong context association capabilities is selected as the base model. The above preprocessed data is used as the fine-tuning task dataset for supervised fine-tuning, and algorithms such as rejection sampling, proximal strategy optimization, and direct strategy optimization are combined to improve the model's specific task performance. The fine-tuned model is used to simulate the financial life scenarios of the new commercial district, including the economic benefits and social benefits of branch operations. Benefit predictions are made for different site selection schemes, including analysis based on economic benefits, social benefits, and cost factors to generate evaluation results with different focuses. The model outputs prediction results and analysis reports. The report records the site selection analysis process and reasoning steps in detail, including the results of different prediction tasks, to provide comprehensive reference information for decision makers.

[0100] Using the thinking chain technology, complex problems are decomposed into multiple sub-problems. When the model performs the prediction task, it simultaneously generates reasoning steps and corresponding decision information. According to the preset prediction task and model feedback information, the large model automatically generates an analysis report, which analyzes the outlet operation benefits of each prediction task in detail and provides decision-making suggestions. The report also contains the evaluation criteria for model generation, which is used for subsequent model iteration and manual review to ensure the scientificity and rationality of the decision.

[0101] Through this embodiment, a comprehensive and objective branch location analysis report is obtained. The report not only includes economic benefit forecasts, but also covers social benefits and cost factors, providing strong data support and decision-making basis for the scientific location selection of bank branches. In addition, the report also provides detailed reasoning process and evaluation criteria, which helps banks understand the logic behind model decisions and enhances the transparency and explainability of decisions.

[0102] The embodiment of the present application also provides a bank branch intelligent site selection device. It should be noted that the bank branch intelligent site selection device of the embodiment of the present application can be used to execute the bank branch intelligent site selection method provided by the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0103] The following is an introduction to the bank branch intelligent site selection device provided in the embodiment of the present application.

[0104] Fig. 9 : is a structural block diagram of the bank branch intelligent site selection device according to the embodiment of the present application. Fig. 9As shown, the device includes an acquisition unit 10, a training unit 20, an optimization unit 30 and a prediction unit 40. The acquisition unit is used to acquire a training sample set related to a financial life scenario, the training sample set includes multiple training samples, each of the training samples includes feature information and label information, the feature information includes map information, personnel information, enterprise information and traffic information, and the label information is the result of site selection and scoring using a preset standard; the training unit is used to train a large language model using the training sample set to obtain an initial site selection model; the optimization unit is used to optimize the initial site selection model using a preset fine-tuning strategy to obtain a target site selection model; the prediction unit is used to input the test sample set around the candidate address into the target site selection model for prediction, and use a thinking chain to generate reasoning steps to obtain a site selection decision result.

[0105] Specifically, before starting model training, we first need to obtain data, namely the training sample set, which is closely related to the location decision of bank branches in financial life scenarios. Financial life scenarios can be understood as the environment and conditions around bank branches, including commercial areas, residential areas, etc., as well as the operation of bank branches and customer behavior in these scenarios.

[0106] The training sample set is a collection of multiple training samples, each of which consists of feature information and label information. Each sample represents the potential location of a bank branch in a specific location, as well as the environmental characteristics and evaluation results related to the location.

[0107] The label information is the site selection and scoring result of the training samples based on a set of preset evaluation criteria, including market potential, cost-benefit analysis, customer convenience, competition analysis, etc. The label information is the goal of model training, which helps the model learn how to judge and predict the site selection effect of bank outlets given the feature information.

[0108] In general, the process of obtaining a training sample set involves collecting detailed information about the environment around bank branches, namely feature information, and selecting and scoring these feature information according to the site selection evaluation criteria formulated by experts to obtain label information, thereby obtaining a training sample set containing feature information and label information, which will be used to train the large language model. This process ensures that the input data for model training is both comprehensive and targeted.

[0109] To train the large language model using the training sample set, it is first necessary to preprocess the obtained training sample set, including data cleaning and standardization. The purpose of preprocessing is to ensure that the model can effectively learn from the data. The large language model is trained using the pre-trained training sample set. The training process aims to adjust the parameters of the large language model so that it can minimize the difference between the predicted results and the true labels. Through a large number of iterations, the large language model gradually learns how to make location decisions similar to expert evaluations based on feature information. After training, the large language model can predict the location and score based on the input feature information, and obtain the initial location model, which can preliminarily understand the financial life scenario and make location decisions.

[0110] By using a large amount of data related to financial life scenarios and through training, the initial site selection model learned how to predict bank branch location decisions based on map information, personnel information, corporate information and traffic information.

[0111] The initial site selection model already has the basic ability to predict bank branch location decisions based on feature information, but it still needs to be further optimized to better meet specific business scenarios and goals.

[0112] The purpose of the preset fine-tuning strategy is to optimize the performance of the initial site selection model on a specific task. In this embodiment, a combination of multiple algorithms including multiple algorithms is used to fine-tune the initial site selection model to obtain a target site selection model. The target site selection model can more effectively understand the complexity of financial life scenarios, more accurately predict the potential and risks of different locations as bank outlets, and generate more reasonable site selection suggestions based on the input feature information.

[0113] Specifically, the test sample set refers to a data set of environmental characteristics (such as population, traffic information, etc.) around the candidate address. These data are used to simulate the financial life scenarios of different candidate addresses so that the model can perform predictive analysis. The test sample set is ingested into the trained and optimized target site selection model, which comprehensively analyzes the potential benefits and social impact of the candidate address and outputs the prediction result, which is the result of site selection and scoring.

[0114] Thinking chain is a method to decompose complex decision-making processes into a series of simple reasoning steps, which can help the model consider more dimensions of information when making decisions and form a more comprehensive judgment. In this embodiment, when the target site selection model predicts the candidate address, a sequence of reasoning steps will be generated at the same time. The sequence of reasoning steps records in detail how the model derives the site selection decision result from the input data, including analysis of the characteristics of the input data, comparison with historical data, and the basis for strategic selection.

[0115] The specific location decision results are generated by combining the evaluation results of the target location model prediction output and the reasoning steps generated by the thought chain. The location decision results not only include the final recommended address of the model, but also come with detailed decision basis and explanation, making the decision process transparent and understandable, and facilitating manual review and iterative improvement of the results. For example, the target location model recommends a candidate address because it predicts large customer traffic, relatively low operating costs, and few surrounding competitors. These reasoning steps will be clearly displayed in the final location decision report.

[0116] In the specific implementation process, the above-mentioned optimization unit includes a supervised fine-tuning module, a rejection sampling module, a proximal strategy optimization module and a direct strategy optimization module. Among them, the supervised fine-tuning module is used to perform supervised training on the above-mentioned initial site selection model according to the fine-tuning task, and adjust the parameters of the above-mentioned initial site selection model; the rejection sampling module is used to exclude samples that are not within the preset standard range from the above-mentioned training sample set using the rejection sampling technology; the proximal strategy optimization module is used to use the proximal strategy optimization algorithm to adjust the strategy selection of the above-mentioned initial site selection model in the decision-making process; the direct strategy optimization module is used to optimize the output results of the above-mentioned initial site selection model using the direct strategy optimization technology, and adjust the parameter configuration of the above-mentioned initial site selection model in the output strategy.

[0117] Supervised fine-tuning (SFT) introduces annotated datasets for specific tasks and conducts further training to optimize the performance of the model on the task. Through supervised fine-tuning, the model can learn from specific annotated instances to improve prediction accuracy and decision quality. Among them, the supervised fine-tuning module includes a first setting submodule and an input submodule, wherein the first setting submodule is used to set the fine-tuning task according to the training sample set, and the fine-tuning task includes site selection prediction under different preset scenarios, and the preset scenarios include financial life scenarios around residential areas, financial life scenarios around office buildings, financial life scenarios around commercial entities, and financial life scenarios around transportation hubs; the input submodule is used to input the fine-tuning task into the initial site selection model for prediction, obtain the prediction result, calculate the difference between the prediction result and the true label, and adjust the parameters of the initial site selection model according to the difference.

[0118] Specifically, the setting of fine-tuning tasks requires a set of known instance data related to the location selection of bank branches, which includes detailed information of the candidate addresses and their actual performance after operation (real labels). The preset scenarios of financial life scenarios around residential areas, financial life scenarios around office buildings, financial life scenarios around commercial entities, and financial life scenarios around transportation hubs further clarify the scope of fine-tuning tasks. This embodiment sets four fine-tuning tasks, which can be set as the simulation prediction of the income brought to the branch by residents and branches for personal financial activities around residential areas; the prediction of personal and corporate financial business interactions between corporate employees and branches due to work and life needs around office buildings; the prediction of the income of corporate groups such as merchants and enterprises conducting public business with branches around commercial entities; and the prediction of business interactions between tourists, passers-by, and non-fixed personnel and branches at transportation hubs.

[0119] The set fine-tuning tasks are input into the initial site selection model. The initial site selection model predicts each task based on its pre-trained parameters and outputs loss values, evaluation indicators, learning rates, weights, gradients, training progress information, and deductive reasoning processes. Among them, the loss value reflects the gap between the model prediction results and the actual results, which is used to evaluate the accuracy of the site selection results; the evaluation indicators include accuracy, precision, recall, etc., which measure the performance of the entire model; the learning rate is the learning rate of the current training process; the weights, gradients, and training progress information are used to analyze the current model training progress and effect; the deductive reasoning process visualizes the model training and reasoning process to assist in dynamic monitoring.

[0120] The difference between the predicted result and the true label can be quantified by a loss function (such as mean square error, cross entropy loss, etc.). Through the back propagation algorithm, the parameters of the model are adjusted according to the gradient of the loss function to reduce the prediction error and improve the prediction accuracy of the model.

[0121] The above process constitutes a training iteration, and the initial site selection model continuously performs predictions and parameter adjustments on each fine-tuning task until the error between the prediction results and the true labels in all preset scenarios reaches the set optimization standard, or reaches the preset maximum number of iterations. This process not only optimizes the model's prediction ability in specific scenarios, but also enhances the model's generalization ability to different financial life scenarios.

[0122] By setting fine-tuning tasks covering a variety of preset scenarios and using supervised training to adjust the parameters of the initial site selection model, we can learn the key factors of site selection decisions through simulation and prediction of different scenarios, and apply this knowledge in future site selection tasks to improve the efficiency and effectiveness of branch operations. Supervised fine-tuning quantifies the prediction error and adjusts the parameters to ensure that the model can learn from historical data and continuously optimize its decision-making strategy, which is crucial to improving the accuracy and reliability of the model.

[0123] Rejection sampling is a data screening technique used to ensure the quality of training data, especially when the original dataset contains a lot of noise or irrelevant samples. In bank branch location selection, rejection sampling can help the model avoid learning unrepresentative samples, thereby improving the stability and accuracy of decision-making. Use rejection sampling techniques to exclude samples that are not within the preset criteria from the training sample set. By training with high-quality samples, the model can avoid learning wrong or irrelevant patterns, thereby improving its performance on real site selection tasks.

[0124] Proximal strategy optimization is a reinforcement learning algorithm that enables the model to make better choices in the decision-making process. In the site selection task, the proximal strategy optimization algorithm can help the model optimize the site selection strategy and improve the economic and social benefits of the decision. The above-mentioned proximal strategy optimization module includes setting a second setting submodule, an evaluation submodule, an update submodule and a repetition submodule. Among them, the second setting submodule is used to set the strategy function, and use the output of the above-mentioned initial site selection model as the strategy parameter of the above-mentioned strategy function; the evaluation submodule is used to use the above-mentioned strategy parameters to generate a series of site selection decision samples, and evaluate the above-mentioned site selection decision samples, and generate a reward signal through a preset reward mechanism; the update submodule is used to adjust the above-mentioned strategy parameters according to the preset update rules based on the above-mentioned site selection decision samples and the above-mentioned reward signal to optimize the strategy selection of the above-mentioned initial site selection model in the decision-making process; the repetition submodule is used to repeatedly execute the above-mentioned evaluation step and the above-mentioned update step until the preset optimization standard is reached.

[0125] Specifically, the use of the proximal policy optimization algorithm to adjust the strategy selection of the initial site selection model is a process of improving the model's decision-making ability by iteratively optimizing the policy parameters. First, the policy function is set. The policy function is the core component in the proximal policy optimization algorithm, which defines the probability of the model choosing different decisions (i.e., choosing a certain address as an outlet) under a given environmental state (such as the characteristics of a candidate address and its surrounding financial life scenes). The parameters of the policy function, i.e., the policy parameters, are initialized based on the output of the initial site selection model. This means that the model's preliminary prediction results will serve as the starting point for the policy function, which will use these parameters to guide the decision-making process.

[0126] The evaluation submodule involves using the current strategy parameters to generate a series of location decision samples, which represent the model's location decisions under the current strategy. Each generated decision sample is then evaluated based on a preset reward mechanism that quantifies the quality of the decision sample. For example, if a decision sample predicts a high branch operation efficiency, it will receive a higher reward signal. The reward signal reflects the expected performance of the decision sample in the preset environment (financial life scenario) and is key information to guide the adjustment of strategy parameters.

[0127] In the update submodule, based on the site selection decision samples and the corresponding reward signals, the policy parameters are adjusted according to the preset update rules. The proximal policy optimization algorithm limits the update amplitude of the policy parameters through a method called "clipping target" to ensure that the new policy does not deviate too far from the old policy, thereby ensuring the stability and convergence of the model update. This process optimizes the policy parameters to make the model's policy selection in the decision-making process more optimized, and can generate higher reward signals, that is, better site selection decisions.

[0128] Repeating the submodules ensures that the whole process is an iterative optimization cycle. The preset optimization criterion can be a specific reward threshold, or after a certain number of iterations, the update amplitude of the strategy parameters is less than a preset threshold, indicating that the model's strategy selection has stabilized and optimized. By continuously generating decision samples, evaluating reward signals, and adjusting strategy parameters, the model can gradually learn the strategy for making the most optimal location decision in complex financial life scenarios.

[0129] In summary, the use of the proximal policy optimization algorithm to adjust the strategy selection of the initial site selection model is a process of converting the model prediction output into strategy parameters, generating decision samples, evaluating their performance under the preset reward mechanism, and adjusting the strategy parameters accordingly to optimize the model's decision-making ability. This algorithm emphasizes the iterative optimization of the strategy, and only makes small adjustments with each update, which ensures the stability and efficiency of the model learning, allowing the model to better adapt to the ever-changing financial life scenarios and improve the accuracy and efficiency of bank branch site selection decisions.

[0130] Direct strategy optimization is a reinforcement learning method that directly optimizes the model's strategy output. It improves the model's decision-making performance by directly modifying the model's parameter configuration on the output strategy. In bank branch site selection, direct strategy optimization can ensure that the model can take all relevant factors into account and make the most optimized decision when outputting site recommendations.

[0131] By combining the supervised fine-tuning module, the rejection sampling module, the proximal strategy optimization module, and the direct strategy optimization module, the initial site selection model can more accurately evaluate the potential of candidate sites and make data-based decisions. This not only improves the model's predictive ability, but also enhances its adaptability and generalization capabilities in complex financial life scenarios, thereby achieving better performance in site selection tasks.

[0132] The above-mentioned prediction unit includes a setting module, a prediction module and an analysis and processing module. Among them, the setting module is used to set the prediction task according to the above-mentioned test sample set after obtaining the target site selection model, and the above-mentioned prediction task includes site selection based on cost and benefit; the prediction module is used to use the above-mentioned prediction task as the input of the above-mentioned target site selection model, perform site selection prediction through the above-mentioned target site selection model, and use the thinking chain technology to generate the reasoning steps related to the above-mentioned prediction task; the analysis and processing module is used to analyze and process the above-mentioned reasoning steps to obtain the above-mentioned site selection decision result, and the above-mentioned site selection decision result includes the predicted site selection result of the above-mentioned target site selection model and the analysis report of the above-mentioned predicted site selection result.

[0133] Specifically, based on the test sample set around the candidate address, the prediction task is clarified. The prediction task is site selection prediction, with special attention to predicting the economic costs and benefits of site selection. This means that the target site selection model must not only be able to predict the potential benefits of each candidate address (such as customer traffic, business volume, operating income, etc.), but also be able to evaluate the economic costs of site selection (such as rent, construction cost, operating cost, etc.) to help banks make more reasonable and economical site selection decisions. The prediction task is used as a criterion for selecting outlets, including address selection based on economic benefits, social benefits and cost factors.

[0134] The set prediction task is input into the target site selection model, and the target site selection model makes site selection predictions for each candidate address based on the learned strategies and parameters. At the same time, the thinking chain begins to play a role. The thinking chain is a method that can decompose the decision-making process into a series of logical reasoning steps, which enables the model to generate detailed reasoning steps when making predictions, that is, how the model derives the prediction results from the input data. These reasoning steps include analysis of the costs and benefits of the candidate addresses, comparison with historical data, and strategy selection based on comprehensive evaluation of multiple factors.

[0135] Finally, the reasoning steps generated by the chain of thought are analyzed and processed to obtain the site selection decision results. The site selection decision results include both the model's predicted site selection results and a detailed explanation of the prediction process and results, that is, the analysis report. The analysis report will record in detail the model's decision-making basis, the cost and benefit evaluation process, the reasons for the site selection recommendation, and even the model's uncertainty and possible risk points in the decision-making process. This explanatory output not only provides recommended sites, but also provides sufficient background information and decision logic to help decision makers understand the model's recommendations and conduct manual review or adjustments when necessary.

[0136] In order to comprehensively and systematically obtain training sample sets related to financial life scenarios and provide rich data support for training an accurate site selection model, the acquisition unit includes: a capture module, a collection module, an extraction module and an acquisition module. The capture module is used to capture the above map information through the electronic map data interface, wherein the above map information includes street layout, building distribution and transportation facility location; the collection module is used to collect the above personnel information by combining automated survey plans and data collection, wherein the above personnel information includes population size, age distribution, occupation type and working hours; the extraction module is used to use data analysis tools to extract the above enterprise information, wherein the above enterprise information includes enterprise business type, enterprise scale and enterprise working hours; the acquisition module is used to obtain the above traffic information based on map data and traffic analysis algorithms, wherein the above traffic information includes travel mode and commuting time.

[0137] Specifically, regarding the acquisition of map information, in this embodiment, through the connection with the electronic map data interface, map information can be automatically captured. The map information includes street layout, building distribution and transportation facility location, which can help the model understand the environment of the alternative address, such as the type of surrounding buildings (residential, office, commercial, etc.) and the distance and type of public transportation stations. These are key factors in evaluating the site selection of outlets.

[0138] The collection of personnel information adopts a combination of automated survey solutions and data collection, which involves obtaining data such as population size, age distribution, occupation type, and working hours around the candidate address. This information is crucial for predicting the potential customers and customer behavior of the outlets. For example, age distribution can help understand the target customer group, and occupation type and working hours can be used to predict customer traffic during non-working hours and weekdays, thereby evaluating the operating potential of the outlets.

[0139] Business information is extracted from multiple sources using data analysis tools, including business type, business size, and business hours. This information is critical to understanding the business environment and potential corporate business volume around the candidate address. Business type and size can reflect the economic vitality of the area, while business hours can help predict the business volume of the outlet during the weekday, especially the demand for corporate business.

[0140] Traffic information is obtained through map data and traffic analysis algorithms, including travel modes (public transportation, private cars, walking, etc.) and commuting time. The acquisition of traffic information can help the model evaluate the convenience of transportation and customer flow at the candidate address, especially during peak hours in the morning and evening and on weekends. Changes in traffic conditions have a significant impact on the business volume of outlets. By analyzing the distribution of transportation facilities, the rules of traffic flow, and the frequency of use of different travel modes, the model can more accurately predict the service coverage and potential customer groups of outlets.

[0141] The feature information (map information, personnel information, enterprise information and traffic information) acquired above and the label information together constitute a training sample set. The label information is the result of site selection and scoring according to preset standards, which helps the model understand which feature combinations can bring higher operational benefits and social value. The acquisition unit includes a multi-dimensional evaluation module and a generation module. Among them, the multi-dimensional evaluation module is used to perform a multi-dimensional evaluation based on the acquired training sample set using the preset standards to obtain multiple pre-site selection results and the scoring scores corresponding to the pre-site selection results; the generation module is used to generate the label information based on the pre-site selection results and the scoring scores.

[0142] Specifically, based on the collected training sample set, each candidate address is evaluated in multiple dimensions using preset standards. The preset standards are based on standards including economic benefits, social benefits, operating costs and risk assessment. Among them, economic benefits can be reflected by evaluating the potential customer flow, operating income and return on investment of the candidate address; social benefits can be reflected by the contribution of the candidate address to the surrounding community, such as improving the coverage of financial services and promoting employment; operating costs can be reflected by analyzing the construction cost, rental cost and operating cost of the candidate address; regarding risk assessment, the financial risks of the candidate address can be evaluated, such as potential negative impacts such as market competition and economic fluctuations. Through the above preset standards, each sample is comprehensively scored to generate multiple site selection and scoring results, which reflect the site selection potential of the candidate address under different evaluation dimensions. The scoring score can be a comprehensive indicator calculated by the weighted average method, and the weight of each evaluation dimension is set according to the actual situation of the bank. For example, if the bank currently pays more attention to economic benefits, the weight of the economic benefit indicator is higher.

[0143] The pre-selected site results and scores obtained above are the label information, which is used to train the site selection model. Label information is the goal of model learning, allowing the model to understand which addresses are evaluated as better site selection options under given feature information, and the specific scores of these sites under the preset standards. In this way, the model can learn the relationship between feature information and preset standard evaluation results, so as to make more accurate site selection predictions on new data.

[0144] In short, by using preset standards to conduct multi-dimensional evaluation of the acquired feature information, pre-selection results and scoring scores are generated, and then these evaluation results are converted into label information required for training the model to build a site selection model based on historical data and actual operating experience. This process not only ensures the pertinence and effectiveness of model training, but also enables the model to comprehensively consider economic benefits, social benefits, operating costs, risk assessment and other factors when making new site selection decisions, and make more scientific and comprehensive site selection recommendations.

[0145] In this embodiment, the acquisition unit comprehensively acquires the training sample sets closely related to the financial life scenarios, and the optimization unit realizes the effective training and fine-tuning optimization of the large language model. The target site selection model finally obtained can accurately predict the alternative addresses. At the same time, with the help of thinking chain technology, the decision reasoning steps are automatically generated, which significantly improves the scientificity, accuracy and transparency of the site selection decision, effectively guides the reasonable site selection of bank outlets to achieve the best economic and social benefits, greatly optimizes the traditional site selection process, and improves the efficiency of bank branch site selection.

[0146] The above-mentioned bank branch intelligent site selection device includes a processor and a memory. The above-mentioned acquisition unit, training unit, optimization unit, prediction unit, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize the corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned modules are located in different processors in any combination.

[0147] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0148] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the bank branch intelligent site selection method.

[0149] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the program executes the bank branch intelligent site selection method when running.

[0150] An embodiment of the present invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned bank branch intelligent site selection method are implemented.

[0151] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes the steps of the above-mentioned bank branch intelligent site selection method.

[0152] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0153] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0154] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0155] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0156] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0157] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0158] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0159] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0160] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0161] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for intelligent site selection of bank outlets, characterized in that: include: Acquire a training sample set related to financial life scenarios, the training sample set including multiple training samples, each of the training samples including feature information and label information, the feature information including map information, personnel information, enterprise information and traffic information, and the label information is the result of site selection and scoring using preset standards; Using the training sample set to train the large language model to obtain an initial site selection model; The initial site selection model is optimized by using a preset fine-tuning strategy to obtain a target site selection model; The test sample set around the candidate address is input into the target site selection model for prediction, and the thinking chain is used to generate reasoning steps to obtain the site selection decision result.

2. The method according to claim 1, characterized in that The initial site selection model is optimized using a preset fine-tuning strategy, including at least one of the following: A supervised fine-tuning step, performing supervised training on the initial site selection model according to the fine-tuning task, and adjusting the parameters of the initial site selection model; A rejection sampling step, using a rejection sampling technique to exclude samples that are not within a preset standard range from the training sample set; A proximal strategy optimization step, using a proximal strategy optimization algorithm to adjust the strategy selection of the initial site selection model in the decision-making process; The direct strategy optimization step uses direct strategy optimization technology to optimize the output result of the initial site selection model and adjust the parameter configuration of the initial site selection model on the output strategy.

3. The method according to claim 2, characterized in that The initial site selection model is supervisedly trained according to the fine-tuning task, and the parameters of the initial site selection model are adjusted, including: The fine-tuning task is set according to the training sample set, and the fine-tuning task includes site selection prediction under different preset scenarios, and the preset scenarios include financial life scenarios around residential areas, financial life scenarios around office buildings, financial life scenarios around commercial entities, and financial life scenarios around transportation hubs; The fine-tuning task is input into the initial site selection model for prediction to obtain a prediction result, the difference between the prediction result and the true label is calculated, and the parameters of the initial site selection model are adjusted according to the difference.

4. The method according to claim 1, characterized in that The test sample set around the candidate address is input into the target site selection model for prediction, and the thinking chain is used to generate reasoning steps to obtain the site selection decision result, including: Setting a prediction task according to the test sample set, the prediction task including selecting a site according to cost and benefit; Taking the prediction task as the input of the target site selection model, performing site selection prediction through the target site selection model, and using the thought chain technology to generate reasoning steps related to the prediction task; The reasoning steps are analyzed and processed to obtain the site selection decision result, which includes the predicted site selection result of the target site selection model and an analysis report on the predicted site selection result.

5. The method according to claim 1, characterized in that A training sample set related to a financial life scenario is obtained, wherein the training sample set includes a plurality of training samples, each of which includes feature information and label information, wherein the feature information includes map information, personnel information, enterprise information, and traffic information, and the label information is a result of site selection and scoring using a preset standard, including: Capturing the map information through an electronic map data interface, wherein the map information includes street layout, building distribution and transportation facility locations; Collecting the personnel information by combining automated survey solutions and data collection, wherein the personnel information includes population size, age distribution, occupation type and working hours; Using a data analysis tool to extract the enterprise information, wherein the enterprise information includes the enterprise business type, enterprise scale and enterprise working hours; The traffic information is obtained based on map data and a traffic analysis algorithm, wherein the traffic information includes travel modes and commuting time.

6. The method according to claim 1, characterized in that A training sample set related to a financial life scenario is obtained, wherein the training sample set includes a plurality of training samples, each of which includes feature information and label information, wherein the feature information includes map information, personnel information, enterprise information, and traffic information, and the label information is a result of site selection and scoring using a preset standard, including: According to the acquired training sample set, a multi-dimensional evaluation is performed using the preset standard to obtain a plurality of pre-selected site results and scoring scores corresponding to the pre-selected site results; The label information is generated according to the pre-site selection result and the scoring score.

7. The method according to claim 2, characterized in that The proximal strategy optimization algorithm is used to adjust the strategy selection of the initial site selection model in the decision-making process, including: A setting step, setting a strategy function, and using the output of the initial site selection model as a strategy parameter of the strategy function; An evaluation step, using the strategy parameters to generate a series of site selection decision samples, and evaluating the site selection decision samples to generate a reward signal through a preset reward mechanism; An updating step, based on the site selection decision samples and the reward signal, adjusting the strategy parameters according to a preset updating rule to optimize the strategy selection of the initial site selection model in the decision-making process; The evaluation step and the updating step are repeatedly performed until a preset optimization criterion is reached.

8. A bank branch intelligent site selection device, characterized in that: include: An acquisition unit, used to acquire a training sample set related to a financial life scenario, wherein the training sample set includes a plurality of training samples, each of which includes feature information and label information, wherein the feature information includes map information, personnel information, enterprise information, and traffic information, and the label information is a result of site selection and scoring using a preset standard; A training unit, used to train the large language model using the training sample set to obtain an initial site selection model; An optimization unit, used to optimize the initial site selection model by using a preset fine-tuning strategy to obtain a target site selection model; The prediction unit is used to input the test sample set around the candidate address into the target site selection model for prediction, and use the thinking chain to generate reasoning steps to obtain the site selection decision result.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the bank branch intelligent site selection method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing the bank branch intelligent site selection method described in any one of claims 1 to 7.

Images