An AI agent-based digital employee valuation, accounting and reconciliation method

Abstract

The application discloses an evaluation, accounting and reconciliation method based on AI Agent digital employees, which comprises the following steps: according to the evaluation, accounting and reconciliation process, the digital employees deployed on the virtual machine are divided into digital inspection posts, daily settlement posts and reconciliation posts; an AI Agent digital employee is constructed, the AI Agent digital employee manages and allocates the digital inspection posts, the daily settlement posts and the reconciliation posts, and completes the whole process of automatic evaluation, accounting and reconciliation. The application can effectively solve the efficiency and quality problems in the evaluation, accounting and reconciliation process, the AI Agent robot can dynamically and reasonably allocate product batches, is the planner and manager of the whole work process, and the digital employee robots at various posts can more accurately and efficiently complete the evaluation, accounting and reconciliation work. The work personnel can focus on the abnormal processing and management of the business process, thereby saving time and labor cost for the enterprise and improving the accuracy and standardization of the evaluation, accounting and reconciliation work.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of valuation settlement, in particular to a valuation and accounting method based on AI Agent digital employees. BACKGROUND

[0002] With the continuous increase of fund asset custody and outsourcing valuation business, it is increasingly difficult to accurately complete all valuation settlement businesses on time under limited manpower. Introducing more professional securities accounting talents also has problems such as high cost, lack of talents and low efficiency. During the manual actual valuation operation of fund assets, there are factors such as batch copying, fuzzy batch account running, re-account running after account running failure, no quality inspection, inability to accurately locate specific product execution steps and status, and no detailed comparison process for custody outsourcing comparison. The overall quality control is lacking, and it is often necessary to find the reason after the account running fails, which brings a large time cost to re-run the account, affecting the valuation output ratio and timeliness. SUMMARY

[0003] The purpose of the present application is to overcome the shortcomings and deficiencies of the prior art and provide a valuation and accounting method based on AI Agent digital employees.

[0004] The purpose of the present application is achieved by the following technical solutions:

[0005] A valuation and accounting method based on AI Agent digital employees, comprising the following steps:

[0006] S1, according to the valuation and accounting process, the digital employees deployed on the virtual machine are divided into digital inspection posts, daily settlement posts and reconciliation posts;

[0007] Among them, the digital inspection post is used for the inspection of data of transactions inside and outside the market; the daily settlement post is used for the accounting; the reconciliation post is used for the inspection of data of reconciliation statements, internal reconciliation and custody outsourcing reconciliation;

[0008] S2, construct AI Agent digital employees, AI Agent digital employees manage and distribute digital inspection posts, daily settlement posts and reconciliation posts, and complete the whole process of automatic valuation and accounting reconciliation;

[0009] The AI Agent digital employee processing logic is as follows:

[0010] Store the historical valuation processing time of a plurality of fund product pool products, divide them into four levels of fast account, relatively fast account, slow account and relatively slow account according to the concurrent processing capacity of the Hang Seng system, and set the corresponding concurrent number of each level;

[0011] The AI Agent plans the products to be processed each day, dynamically groups the products according to the product processing time, self-evaluates the historical grouping data and processing efficiency before grouping, self-optimizes the grouping plan, and plans the best grouping combination for the day.

[0012] The AI Agent monitors the state of the digital employee on the virtual machine, whether it is an idle state or a working state; when it is monitored that there is a digital employee on the virtual machine in an idle state, the AI Agent digital employee will automatically allocate batches to the idle digital employee.

[0013] The AI Agent digital employee in dynamic grouping, the actual product quantity allocated to the four levels is calculated as follows:

[0014] First, the theoretical product quantity that the four levels can process in unit time is calculated, denoted as R fast , R faster , R slow , R slower , the calculation formula is as follows:

[0015] R fast = C fast ×T;

[0016] R faster = C faster ×T;

[0017] R slow = C slow ×T;

[0018] R slower = C slower ×T;

[0019] Wherein, C fast , C faster , C slow , C slower are the concurrency numbers corresponding to fast account, faster account, slow account, and slower account respectively; T represents the total processing time.

[0020] Then, according to the historical grouping data and processing efficiency, the product quantity proportion of each level is self-evaluated and optimized to obtain the product quantity proportion P fast , P faster , P slow , P slower of fast account, faster account, slow account, and slower account.

[0021] Finally, the actual product quantity allocated to each level is calculated, denoted as n fast , n faster , n slow , n slower, the calculation formula is as follows:

[0022] n fast = N×P fast ;

[0023] n faster = N×P faster ;

[0024] n slow = N×P slow ;

[0025] n slower = N×P slower ;

[0026] Wherein, N represents the total number of products to be processed.

[0027] The AI Agent considers the processing time fluctuation, and the calculation process of the actual product quantity to be allocated in four grades is as follows:

[0028] (1) Calculate the minimum value R min-fast , R min-faster , R min-slow , R min-slower and the maximum value R max-fast , R max-faser , R max-slow , R max-slower of the product quantity processed in four grades per unit time:

[0029] R min-fast ×C fast ;

[0030] R max-fast = ×C fast ;

[0031] R min-faster = ×C faster ;

[0032] R max-faser = ×C faster ;

[0033] R min-slow ×C slow ;

[0034] R max-slow = ×C slow ;

[0035] R min-slower ×Cslower ;

[0036] R max-slower = ×C slower ;

[0037] wherein t fast-min , t fast-max are the minimum and maximum values of the fast account product processing time, t faster-min , t faster-max are the minimum and maximum values of the faster account product processing time, t slow-min , t slow-max are the minimum and maximum values of the slow account product processing time, t slower-min , t slower-max are the minimum and maximum values of the slower account product processing time;

[0038] (2) Calculate the four levels of comprehensive unit time product processing number R com-fast , R com-faster , R com-slow , R com-slower combined with the historical processing efficiency weight:

[0039] R com-fast =W fast × +(1-W fast )× R fast ;

[0040] R com-faster =W faster × +(1-W faster )× R faster ;

[0041] R com-slow =W slow × +(1-W slow )× R slow ;

[0042] R com-slower =W slower × +(1-W slower )× R slower ;

[0043] wherein W fast , W faster , W slow , W slower are the processing efficiency weights of the four levels based on historical data;

[0044] (3) Recalculate the product quantity distribution:

[0045] The proportion of the number of products in the four grades P com-fast , P com-faster , P com-slow , P com-slower The calculation is as follows:

[0046] P com-fast = ;

[0047] P com-faster = ;

[0048] P com-slow = ;

[0049] P com-slower = ;

[0050] (4) The number of products in the final four grades n com-fast , n com-faster , n com-slow , n com-slower The calculation is as follows:

[0051] n com-fast = N×P com-fast ;

[0052] n com-faster = N×P com-faster ;

[0053] n com-slow = N×P com-slow ;

[0054] n com-slower = N×P com-slower .

[0055] After the number of products in the four grades is calculated, the AI Agent groups according to the calculated number of products, or groups according to the processing time, or groups according to the product type, or adjusts the grouping in real time.

[0056] Meanwhile, the present application provides:

[0057] A server, comprising a processor and a memory, at least one program is stored in the memory, the program is loaded and executed by the processor to realize the above-mentioned valuation and accounting method based on AI Agent digital employee.

[0058] A computer-readable storage medium, the storage medium has at least one program stored therein, the program is loaded and executed by a processor to realize the above-mentioned valuation accounting method based on AI Agent digital employee.

[0059] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0060] 1、The present application can effectively solve the efficiency and quality problems in the valuation accounting process by constructing the valuation accounting method based on AI Agent digital employee, the AI Agent robot can dynamically and reasonably allocate product batches, and is the planner and manager of the whole workflow, the digital employee robot of each post can more accurately and efficiently complete the valuation accounting work, the operation personnel focus their efforts on the abnormal processing and management of the business process, save time and labor cost for the enterprise, and improve the accuracy and standardization of valuation accounting operation.

[0061] 2、The present application can greatly ensure the accuracy, stability and timeliness of daily valuation of all products by using AI Agent type digital employee for valuation automation, and ensure that the daily valuation of each fund asset is traceable, manageable and controllable. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 The flow chart for valuation accounting reconciliation.

[0063] Figure 2 The schematic diagram of batch processing time.

[0064] Figure 3 The flow chart for realizing automatic valuation accounting reconciliation of digital employees on virtual machines.

[0065] Figure 4 The schematic diagram of batch processing time of digital employees on virtual machines.

[0066] Figure 5 The AI Agent digital employee architecture diagram.

[0067] Figure 6 The AI Agent digital employee work schematic diagram. DETAILED DESCRIPTION

[0068] The present application will be further described in detail below in combination with embodiments and drawings, but the embodiments of the present application are not limited thereto.

[0069] The valuation accounting reconciliation process is as follows Figure 1As shown, it mainly includes five main links of in-out transaction data inspection, accounting link, statement data inspection, internal reconciliation, and custody outsourcing reconciliation, wherein the accounting link includes five parts of public class detection, daily clearing, generation of vouchers, generation of valuation table and error prevention class detection.

[0070] The valuation and accounting is operated in the Heng Seng system, and is completed in batches, each batch containing concurrent accounting of multiple products. Only when each product completes the accounting, the batch can end and the next batch operation is performed. This rule also applies to each node in the process. Figure 2

[0071] As shown in Figures 3 to 6 The present application provides a valuation and accounting reconciliation method based on AI Agent digital employees, comprising the following steps:

[0072] S1, according to the valuation and accounting reconciliation process, the digital employees deployed on the virtual machine are divided into digital inspection post, daily clearing post and reconciliation post;

[0073] Among them, the digital inspection post is used for in-out transaction data inspection link; the daily clearing post is used for accounting link; the reconciliation post is used for statement data inspection link, internal reconciliation link and custody outsourcing reconciliation link;

[0074] S2, constructing AI Agent digital employees, AI Agent digital employees manage and distribute the digital inspection post, daily clearing post and reconciliation post, and complete the full process of automatic valuation and accounting reconciliation;

[0075] As shown in Figure 3 The daily valuation and accounting is performed through the digital employees of the data inspection post, daily clearing post and reconciliation post from the daily cut, and then enters the distribution-accounting to reconciliation. Daily cut operation will be performed at 2:30 am every trading day, and the daily cut will perform key operations such as initialization of resource pool data to be processed, running detection on resource pool robots, and synchronization of valuation and accounting system product accounting number.

[0076] The specific implementation process of the three robots of the digital inspection post, the daily clearing post and the reconciliation post:

[0077] a) Data inspection post: data inspection mainly involves product transaction data type data and statement fund data information type data.

[0078] ​File type data is pre-configured path and file name rule through RPA robot frequency inspection to monitor the data completion. Based on business logic, each product will have multiple fund accounts, each fund account has two paths for storing transaction data files of hosting and outsourcing, and different file name matching rules are configured based on the inherent properties of fund accounts. When all fund accounts of a product complete data completion, the current product data is complete, and the data inspection post notifies the daily settlement post.

[0079] The RPA robot checks the reconciliation fund data and the data completion, and automatically imports the data. Based on the same business logic as described above, when all fund account reconciliation fund data is complete and imported, the data inspection post notifies the reconciliation post.

[0080] b) Daily settlement post: RPA robots here are divided into two categories, one is task assignment machine, and the other is task processing machine. Each functional node has only one corresponding task assignment machine. Task processing machine has multiple, each can receive different tasks issued by different assignment machines, and process different processes according to the different tasks. Here, the daily settlement post mainly refers to the RPA robot playing the role of task processing machine. After receiving the relevant task instructions, such as daily settlement processing instructions, the RPA robot will execute the pre-designed daily settlement process, complete the machine state pre-detection, data extraction, accounting, and end detection. When data extraction, only one processing machine will access the resource pool to be processed at the same time to avoid excessive consumption. Pre-detection, accounting and end detection are developed according to the actual needs of the current process. After the daily settlement post completes all node processes, it will notify the reconciliation post.

[0081] c) Reconciliation post: After receiving the data inspection post's notification of the completion of the reconciliation and the daily settlement post's notification of the completion of the accounting, the reconciliation post will perform the corresponding internal reconciliation operation. Internal reconciliation is the comparison of internal data between hosting and outsourcing. When the internal data of both parties is consistent, an automatic reconciliation between hosting and outsourcing can be performed. If the reconciliation results are inconsistent or there is data that needs to be adjusted, the business personnel will initiate a recheck request.

[0082] For example, Figure 4Load balancing of business processes is achieved by AI Agent digital employees. The business needs to automatically account for 5000+ products every day, and reconcile them. As mentioned earlier, products are processed in batches, and the distribution of processing time for products within a batch significantly affects the efficiency of each batch, which in turn affects the overall efficiency of the business. To better solve this problem, AI Agent robots are used to reasonably plan 5000+ products into different product groups, with the processing time of products in the same product group being concentrated. This reduces waiting time and improves the efficiency of each batch processing and the overall processing efficiency.

[0083] As Figure 5 , the AI Agent digital employee architecture is:

[0084] a) Memory: AI Agent digital employees store the historical valuation processing time of all 5000+ products, and according to the concurrent processing capacity of the Hang Seng system, they are divided into four different levels: fast accounting, relatively fast accounting, slow accounting, and relatively slow accounting. The number of concurrent processes corresponding to each level is 5, 10, 30, and 60, respectively.

[0085] b) Planning: AI Agent plans the products to be processed daily, dynamically groups them according to product processing time, and self-evaluates historical grouping data and processing efficiency before grouping to optimize the grouping plan and plan the best combination for the day.

[0086] c) Tools: AI Agent monitors the status of digital employees (data inspection posts, end-of-day clearing posts, and reconciliation posts) on virtual machines, whether they are idle or working.

[0087] d) Execution: When AI Agent detects that a digital employee is idle, it will automatically assign batches to the idle digital employee.

[0088] The actual number of products to be allocated to the four levels in dynamic grouping is calculated as follows:

[0089] First, calculate the number of products that the four levels can theoretically process in a unit of time, denoted as R fast , R faster , R slow , R slower , the calculation formula is as follows:

[0090] R fast = C fast ×T;

[0091] R faster = C faster ×T;

[0092] Rslow = C slow ×T;

[0093] R slower = C slower ×T;

[0094] Among them, C fast C faster C slow C slower These represent the concurrency levels for four categories: fast, relatively fast, slow, and relatively slow (i.e., 5, 10, 30, and 60 respectively); T represents the total processing time (which can be determined based on the actual operation of the Hengsheng system, such as the length of a day's work).

[0095] Then, based on historical grouping data and processing efficiency, the product quantity ratio of each level was self-evaluated and optimized to obtain the product quantity ratio P of four levels: fast account, relatively fast account, slow account, and relatively slow account. fast P faster P slow P slower (The sum of these percentages should be 1);

[0096] Finally, calculate the actual number of products that should be allocated to each level, and let it be n. fast n faster n slow n slower The calculation formula is as follows:

[0097] n fast = N×P fast ;

[0098] n faster = N×P faster ;

[0099] n slow = N×P slow ;

[0100] n slower = N×P slower ;

[0101] Where N represents the total number of products that need to be processed (in this case, more than 5,000 products).

[0102] The AI ​​Agent takes into account processing time fluctuations, and the calculation process for the actual number of products to be allocated to the four levels is as follows:

[0103] (1) Calculate the minimum value R of the number of products processed per unit time for the four levels. min-fast R min-faster R min-slow R min-slowerand maximum value R max-fast , R max-faser , R max-slow , R max-slower :

[0104] R min-fast ×C fast ;

[0105] R max-fast = ×C fast ;

[0106] R min-faster = ×C faster ;

[0107] R max-faser = ×C faster ;

[0108] R min-slow ×C slow ;

[0109] R max-slow = ×C slow ;

[0110] R min-slower ×C slower ;

[0111] R max-slower = ×C slower ;

[0112] wherein t fast-min , t fast-max are minimum and maximum values of fast account product processing time, t faster-min , t faster-max are minimum and maximum values of relatively fast account product processing time, t slow-min , t slow-max are minimum and maximum values of slow account product processing time, t slower-min , t slower-max are minimum and maximum values of relatively slow account product processing time;

[0113] (2) combined with historical processing efficiency weight, calculate four levels of comprehensive unit time product number R com-fast , R com-faster , R com-slow , R com-slower :

[0114] R com-fast =Wfast × + (1 - W fast ) × R fast ;

[0115] R com-faster = W faster × + (1 - W faster ) × R faster ;

[0116] R com-slow = W slow × + (1 - W slow ) × R slow ;

[0117] R com-slower = W slower × + (1 - W slower ) × R slower ;

[0118] Wherein, W fast , W faster , W slow , W slower are four levels of processing efficiency weight based on historical data;

[0119] (3) Recalculate the product quantity distribution:

[0120] The four levels of new product quantity ratio P com-fast , P com-faster , P com-slow , P com-slower are calculated as follows:

[0121] P com-fast = ;

[0122] P com-faster = ;

[0123] P com-slow = ;

[0124] P com-slower = ;

[0125] (4) The final four levels of product quantity n com-fast , n com-faster , n com-slow , n com-slower are calculated as follows:

[0126] n com-fast= N x P com-fast ;

[0127] n com-faster = N x P com-faster ;

[0128] n com-slow = N x P com-slow ;

[0129] n com-slower = N x P com-slower .

[0130] After calculating the number of products that should be allocated to each level, the AI Agent groups the products according to the calculated number, or sorts the products by processing time, or combines product types, or adjusts the grouping in real time.

[0131] 1. Allocate products according to the calculated number

[0132] According to the number of products calculated above, the products are allocated to the fast account, faster account, slow account, and slower account groups in turn. For example, from the list of all products, the first n com-fast products are selected and placed in the fast account group, then n com-faster products are selected and placed in the faster account group, and so on.

[0133] 2. Auxiliary grouping considering product characteristics

[0134] Sort by processing time: If more accurate processing time estimates for each product can be obtained, all products can be sorted by processing time from small to large. Then, according to the product number range calculated above, the products are divided from the product with the shortest processing time. For example, the product with the shortest processing time is divided into the fast account group, and the next products are divided into the faster account group, slow account group, and slower account group in turn.

[0135] Combine product types: If different product types have different processing time rules, the products can be classified by product type first. For example, products are divided into A, B, C, and other types, and it is known that most A type products historically belong to the faster account processing level. Therefore, when grouping, the products that meet the faster account processing time range in the A type product are placed in the faster account group first.

[0136] 3. Real-time adjustment of grouping

[0137] During the grouping process, if it is found that the actual processing time of a product does not match the expected processing time range of the group it belongs to, or if the system load changes (for example, there is a backlog of concurrent processing tasks for a group), the product can be adjusted to other more suitable groups in real time. For example, a product originally assigned to the fast account group may be adjusted to the slow account group if it is found that its processing time is longer than the upper limit time of the fast account group.

[0138] The following is a Python example code framework for implementing product grouping according to the above grouping idea. Assume that there is a list of products containing all product information, and each product has an attribute (such as processing_time) representing its processing time. The example code is as follows:

[0139] # Define the concurrency of each level.

[0140] CONCURRENCY_FAST = 5;

[0141] CONCURRENCY_FASTER = 10;

[0142] CONCURRENCY_SLOW = 30;

[0143] CONCURRENCY_SLOWER = 60;

[0144] # Total processing time (assuming an example value is set here, actual value needs to be determined according to the situation).

[0145] TOTAL_PROCESSING_TIME = 3600 # in seconds, here assuming 1 hour;

[0146] # Processing time range for each level (example values, need to be adjusted according to actual situation).

[0147] FAST_MIN_TIME, FAST_MAX_TIME = 10, 30;

[0148] FASTER_MIN_TIME, FASTER_MAX_TIME = 30, 60;

[0149] SLOW_MIN_TIME, SLOW_MAX_TIME = 60, 120;

[0150] SLOWER_MIN_TIME, SLOWER_MAX_TIME = 120, 240;

[0151] # Each tier has a weight based on historical data for processing efficiency (example values, actual values should be determined)

[0152] WEIGHT_FAST = 0.3;

[0153] WEIGHT_FASTER = 0.3;

[0154] WEIGHT_SLOW = 0.2;

[0155] WEIGHT_SLOWER = 0.2;

[0156] def calculate_group_sizes(products):

[0157] """

[0158] Calculate the sizes of each group based on the list of products

[0159] """

[0160] num_products = len(products).

[0161] # Calculate the minimum and maximum number of products that can be processed per unit time for each tier

[0162] R_MIN_FAST=TOTAL_PROCESSING_TIME / FAST_MAX_TIME * CONCURRENCY_FAST;

[0163] R_MAX_FAST=TOTAL_PROCESSING_TIME / FAST_MIN_TIME * CONCURRENCY_FAST;

[0164] R_MIN_FASTER=TOTAL_PROCESSING_TIME / FASTER_MAX_TIME*CONCURRENCY_FASTER;

[0165] R_MAX_FASTER=TOTAL_PROCESSING_TIME / FASTER_MIN_TIME*CONCURRENCY_FASTER;

[0166] R_MIN_SLOW=TOTAL_PROCESSING_TIME / SLOW_MAX_TIME * CONCURRENCY_SLOW;

[0167] R_MAX_SLOW = TOTAL_PROCESSING_TIME / SLOW_MIN_TIME * CONCURRENCY_SLOW;

[0168] R_MIN_SLOWER = TOTAL_PROCESSING_TIME / SLOWER_MAX_TIME * CONCURRENCY_SLOWER;

[0169] R_MAX_SLOWER = TOTAL_PROCESSING_TIME / SLOWER_MIN_TIME * CONCURRENCY_SLOWER;

[0170] # Calculate the total number of products processed per unit time.

[0171] R_COM_FAST = (WEIGHT_FAST * ((R_MIN_FAST + R_MAX_FAST) / 2) + (1 - WEIGHT_FAST) * ((R_MIN_FAST + R_MAX_FAST) / 2));

[0172] R_COM_FASTER = (WEIGHT_FASTER * ((R_MIN_FASTER + R_MAX_FASTER) / 2) + (1 - WEIGHT_FASTER) * ((R_MIN_FASTER + R_MAX_FASTER) / 2));

[0173] R_COM_SLOW = (WEIGHT_SLOW * ((R_MIN_SLOW + R_MAX_SLOW) / 2) + (1 - WEIGHT_SLOW) * ((R_MIN_SLOW + R_MAX_SLOW) / 2));

[0174] R_COM_SLOWER = (WEIGHT_SLOWER * ((R_MIN_SLOWER + R_MAX_SLOWER) / 2) + (1 - WEIGHT_SLOWER) * ((R_MIN_SLOWER + R_MAX_SLOWER) / 2));

[0175] # Calculate the new proportion of product quantities for each level.

[0176] total_com = R_COM_FAST + R_COM_FASTER + R_COM_SLOW + R_COM_SLOWER; P_COM_FAST = R_COM_FAST / total_com;

[0177] P_COM_FASTER = R_COM_FASTER / total_com;

[0178] P_COM_SLOW = R_COM_SLOW / total_com;

[0179] P_COM_SLOWER = R_COM_SLOWER / total_com;

[0180] # Calculate the number of products in each group.

[0181] n_com_fast = int(num_products * P_COM_FAST);

[0182] n_com_faster = int(num_products * P_COM_FASTER);

[0183] n_com_slow = int(num_products * P_COM_SLOW);

[0184] n_com_slower = int(num_products * P_COM_SLOWER);

[0185] return n_com_fast, n_com_faster, n_com_slow, n_com_slower;

[0186] def group_products(products):

[0187] """

[0188] Group the products according to the calculated group sizes;

[0189] """

[0190] n_com_fast, n_com_faster, n_com_slow, n_com_slower = calculate_group_sizes(products);

[0191] fast_group = products[:n_com_fast];

[0192] faster_group = products[n_com_fast:n_com_fast + n_com_faster];

[0193] slow_group=products[n_com_fast+n_com_faster:n_com_fast + n_com_faster+n_com_slow];

[0194] slower_group=products[n_com_fast+n_com_faster+n_com_slow:];

[0195] return fast_group,faster_group,slow_group,slower_group.

[0196] # Assuming there is a sample product list here, it needs to be replaced with real data in actual applications.

[0197] products_example=[{"processing_time": 20}, {"processing_time": 40},{"processing_time": 80};

[0198] {"processing_time": 150},{"processing_time": 200}];

[0199] fast_group,faster_group,slow_group,slower_group=group_products(products_example);

[0200] print("Quantity of products in Fast Account Group:", len(fast_group));

[0201] print("Number of products in the faster account group:", len(faster_group));

[0202] print("Number of products in the slow account group:", len(slow_group));

[0203] print("Number of products in the slower account group:", len(slower_group)).

[0204] In the code above:

[0205] The `calculate_group_sizes` function calculates the number of products that should be allocated to each group based on the product list and the set parameters (concurrency, processing time range, processing efficiency weight, etc.).

[0206] The group_products function performs actual grouping operation on the input product list according to the group sizes calculated by the calculate_group_sizes function. The AI Agent written in Python implements the global man-machine collaborative automatic valuation reconciliation method of the present application, and the AI Agent written in other languages is also within the protection scope of the present application.

[0207] The present application combines AI Agent digital employee management and distribution, automatically divides 5000+ products into multiple groups of a small number of product groups, runs on all accounting robots as much as possible, and is assisted by a sound monitoring and recycling mechanism to fully utilize load balancing, high concurrency, high reliability and high scalability to improve accounting efficiency and accounting quality. Through file inspection and risk control of each reconciliation link, and completely separate accounting steps, the overall accounting efficiency, stability and success rate are improved.

[0208] Meanwhile, the present application provides:

[0209] A server, comprising a processor and a memory, at least one program is stored in the memory, the program is loaded and executed by the processor to realize the above-mentioned valuation reconciliation method based on AI Agent digital employee.

[0210] A computer readable storage medium, at least one program is stored in the storage medium, the program is loaded and executed by the processor to realize the above-mentioned valuation reconciliation method based on AI Agent digital employee.

[0211] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited by the above embodiments, any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present application are equivalent replacement methods, which are all included in the protection scope of the present application.

Claims

1. A valuation, accounting, and reconciliation method for digital employees based on AI Agents, characterized in that, Includes the following steps: S1. Based on the valuation accounting and reconciliation process, the digital employees deployed on the virtual machine are divided into digital inspection posts, end-of-day settlement posts, and reconciliation posts; Among them, the digital inspection post is used for the inspection of on-exchange and off-exchange transaction data; the end-of-day clearing post is used for the bookkeeping process, which includes five parts: public inspection, end-of-day clearing, voucher generation, valuation table generation, and error prevention inspection; and the reconciliation post is used for the reconciliation statement data inspection, internal reconciliation, and outsourced reconciliation. S2. Build AI Agent digital employees. AI Agent digital employees manage and allocate digital inspection posts, end-of-day settlement posts, and reconciliation posts to complete the entire process of automated valuation, accounting, and reconciliation. The AI ​​Agent digital employee processing logic is as follows: Store the historical valuation processing time of several fund product pools, and divide them into four levels according to the concurrent processing capacity: fast account, relatively fast account, slow account, and relatively slow account, and set the corresponding concurrency number for each level; The AI ​​Agent plans the products to be processed each day, dynamically groups them according to four levels of product processing time, and performs a self-evaluation of historical grouping data and processing efficiency before grouping, self-optimizing the grouping plan and planning the best combination of plans for the day. The dynamic grouping AI Agent takes into account processing time fluctuations, and the calculation process for the actual number of products to be allocated to the four levels is as follows: (1) Calculate the minimum value R of the number of products processed per unit time for the four levels. min-fast R min-faster R min-slow R min-slower and maximum value R max-fast R max-faser R max-slow R max-slower : R min-fast ×C fast ; R max-fast = ×C fast ; R min-faster = ×C faster ; R max-faser = ×C faster ; R min-slow ×C slow ; R max-slow = ×C slow ; R min-slower ×C slower ; R max-slower = ×C slower ; Among them, t fast-min t fast-max t represents the minimum and maximum processing time for the Quick Account product. faster-min t faster-max t represents the minimum and maximum processing times for faster account products. slow-min t slow-max Minimum and maximum processing times for slow account products, t slower-min t slower-max These are the minimum and maximum processing times for slower account products; (2) Calculate the number of products processed per unit time R for the four levels by combining the historical processing efficiency weights. com-fast R com-faster R com-slow R com-slower : R com-fast =W fast × +(1-W fast )× R fast ; R com-faster =W faster × +(1-W faster )× R faster ; R com-slow =W slow × +(1-W slow )× R slow ; R com-slower =W slower × +(1-W slower )× R slower ; Among them, W fast W faster W slow W slower The processing efficiency weights are assigned to four levels based on historical data. (3) Recalculate the product quantity allocation: The percentage of new products in the four levels (P) com-fast P com-faster P com-slow P com-slower Calculated as follows: P com-fast = ； P com-faster = ； P com-slow = ； P com-slower = ； (4) The final number of products n in the four grades com-fast n com-faster n com-slow n com-slower The calculation method is as follows: n com-fast = N×P com-fast ; n com-faster = N×P com-faster ; n com-slow = N×P com-slow ; n com-slower = N×P com-slower ; The AI ​​Agent monitors the status of digital employees on virtual machines, whether they are idle or working. When it detects that a digital employee on a virtual machine is idle, the AI ​​Agent will automatically assign groups of digital employees to the idle digital employees.

2. The valuation, accounting, and reconciliation method for AI Agent-based digital employees according to claim 1, characterized in that, The calculation process for the actual number of products to be allocated to the four levels of the AI ​​Agent digital employee in dynamic grouping is as follows: First, calculate the theoretical number of products that the four levels can process per unit time, denoted as R. fast R faster R slow R slower The calculation formula is as follows: R fast = C fast ×T; R faster = C faster ×T; R slow = C slow ×T; R slower = C slower ×T; Among them, C fast C faster C slow C slower These represent the concurrency levels for four categories: fast, relatively fast, slow, and relatively slow; T represents the total processing time. Then, based on historical grouping data and processing efficiency, the product quantity ratio of each level was self-evaluated and optimized to obtain the product quantity ratio P of four levels: fast account, relatively fast account, slow account, and relatively slow account. fast P faster P slow P slower ; Finally, calculate the actual number of products that should be allocated to each level, and let it be n. fast n faster n slow n slower The calculation formula is as follows: n fast = N×P fast ; n faster = N×P faster ; n slow = N×P slow ; n slower = N×P slower ; Where N represents the total number of products that need to be processed.

3. The valuation, accounting, and reconciliation method for AI Agent-based digital employees according to claim 1 or 2, characterized in that, After calculating the actual number of products to be allocated to the four levels, the AI ​​Agent groups the products according to the calculated number, or sorts them by processing time, or combines them with product type, or adjusts the groups in real time.

4. A server, the server comprising a processor and a memory, characterized in that, The memory stores at least one program, which is loaded and executed by the processor to implement the valuation and accounting method for AI Agent-based digital employees as described in any one of claims 1 to 2.

5. A computer-readable storage medium storing at least one program, characterized in that, The program is loaded and executed by a processor to implement the valuation and accounting method for AI Agent-based digital employees as described in any one of claims 1 to 2.

Images