Enterprise agent low-code orchestration and version management platform and method

Abstract

The application provides an enterprise intelligent agent low-code arrangement and version management platform and method, which comprises the following steps: implanting intelligent agents in each node, acquiring registration information by an intelligent agent management system, generating or associating intelligent agent scheduling process nodes, forming an intelligent agent scheduling process node set, configuring a transmission instruction ID, downloading the instruction ID block by the intelligent agent management system node, arranging intelligent agent workflows for the instruction ID, triggering the instruction ID, uploading task order information to a consortium chain task order block, downloading block data by the intelligent agent management system, acquiring intelligent agent scheduling process nodes and order types, and screening available intelligent agent sets, generating a task allocation block by using an improved adaptive genetic algorithm, downloading the task allocation block, acquiring interface data by corresponding intelligent agents to execute tasks, and uploading results to a task result block. The application solves the problems of efficient intelligent agent scheduling management and effective application of cloud and edge resources in complex business scenarios of enterprises.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, in particular to an enterprise intelligent agent low-code orchestration and version management platform. BACKGROUND

[0002] Currently, in the wave of enterprise digital transformation, intelligent agent technology, as an important pillar to promote business automation and intelligent upgrading, has shown irreplaceable value. Although intelligent agent technology has made certain progress in enterprise applications, many solutions often struggle to adapt flexibly to dynamic and changing business environments, especially in task allocation and resource coordination. Existing methods often cannot effectively deal with the uneven distribution of resources between different nodes, and it is also difficult to reasonably split and schedule tasks according to their complexity. This limitation leads to resource waste or task delay in peak periods or complex scenarios, affecting the smoothness of overall business flow. SUMMARY

[0003] To solve the above problems, the present application provides an enterprise intelligent agent low-code orchestration and version management platform and method, which can comprehensively improve production efficiency, resource utilization, product quality and system flexibility.

[0004] To achieve the above purpose, the present application provides the following technical scheme:

[0005] On the one hand, an enterprise intelligent agent low-code orchestration and version management method is proposed, and the specific steps are as follows:

[0006] The cloud node or edge node implants an intelligent agent, and uploads the intelligent agent registration information to the alliance chain block, and the intelligent agent management system obtains the registration information to generate or associate intelligent agent scheduling process nodes, forming an intelligent agent scheduling process node set;

[0007] The application system node configures a transmission instruction ID and its order type, and uploads it to the alliance chain instruction ID block, and the intelligent agent management system node downloads the instruction ID block, and arranges intelligent agent workflow for instruction ID based on the intelligent agent scheduling process node set;

[0008] The application system node triggers the instruction ID, uploads the task order information to the alliance chain task order block, and the intelligent agent management system downloads the block data, parses the instruction ID, obtains the intelligent agent scheduling process node and the order type, and filters the available intelligent agent set;

[0009] The intelligent agent scheduling management module uses an improved adaptive genetic algorithm to optimize scheduling to generate a task allocation block based on the task order and the available intelligent agent set;

[0010] Each node downloads a task allocation block, a corresponding agent acquires interface data to execute a task, and uploads a result to a task result block.

[0011] The cloud node or the edge node implants an agent, uploads agent registration information to a consortium chain block, and an agent management system acquires the registration information to generate or associate an agent scheduling process node, forms an agent scheduling process node set, and further includes the following steps.

[0012] The cloud node or the edge node implants an agent, and the agent management system generates a corresponding agent scheduling process node for the same agent implanted in each node to form an agent scheduling process node set.

[0013] The agent management system dynamically evaluates the availability and scalability of the agent of each node based on real-time acquired remaining resource information and agent working state information of each node, automatically registers or deregisters the agent corresponding to each node, and completes automatic reduction and expansion of the agent of each node.

[0014] The application system node configures a transmission instruction ID and an order type thereof, and uploads to a consortium chain instruction ID block. The agent management system node downloads the instruction ID block, and arranges an agent workflow for the instruction ID based on the agent scheduling process node set. Further includes the following steps.

[0015] The application system node configures a transmission instruction ID, defines the order type corresponding to the current instruction ID including a regular order and a shiftable order based on the scene of a business instruction, and in addition, the shiftable order further includes resource threshold-based shunting and feature intensity threshold-based shunting.

[0016] The agent management system node arranges an agent workflow for the task order type corresponding to the instruction ID, and uploads a process configuration block. Each node consensus minimizes the data block, improves the efficiency of data transmission, and saves network resources.

[0017] Specifically, the application system node triggers the instruction ID, uploads task order information to a consortium chain task order block, the agent management system downloads block data, parses the instruction ID, acquires an agent scheduling process node and an order type, and screens an available agent set. Further includes the following steps.

[0018] The cloud node and the edge node server real-time statistics of the latest remaining resource information such as remaining computing power and remaining memory. If there is fluctuation, upload the latest remaining resource information to a remaining resource information block, and consensus to each node of the consortium chain.

[0019] The cloud node and the edge node server real-time acquire all agent working state information of the node, and generate an agent state block, and consensus to each node of the consortium chain.

[0020] The intelligent agent management system node obtains the latest task order block, the latest remaining resource information block and the latest intelligent agent state block in real time, analyzes the instruction ID in the task order block, obtains the intelligent agent scheduling process node and the order type, and screens the available intelligent agent set.

[0021] Further, the intelligent agent management system node obtains the latest task order block, the latest remaining resource information block and the latest intelligent agent state block in real time, analyzes the instruction ID in the task order block, obtains the intelligent agent scheduling process node and the order type, and screens the available intelligent agent set, and further comprises:

[0022] When the order type is a regular order, the available intelligent agent set is screened, and when the order type is a shiftable order, before screening the available intelligent agent set, a data shunting processing module of the intelligent agent management system obtains the to-be-processed data of the task order based on the data interface corresponding to the pre-configured front-end input module of the instruction ID and the condition information corresponding to the current task order.

[0023] Further, according to the shunting factor of the shiftable order being a resource demand threshold or a feature intensity threshold, adaptive window or feature intensity threshold analysis is applied to data sharding on the to-be-processed data of the task order, condition or primary key information corresponding to the data shard is obtained, a split task order is generated, associated with the original task order, and uploaded to the task order block.

[0024] Specifically, according to the shunting factor of the shiftable order being a resource demand threshold or a feature intensity threshold, adaptive window or feature intensity threshold analysis is applied to data sharding on the to-be-processed data of the task order, condition or primary key information corresponding to the data shard is obtained, a split task order is generated, associated with the original task order, and uploaded to the task order block, and further comprising:

[0025] The shunting factor of the shiftable order is a resource demand threshold, and the to-be-processed data of the task order is sharded by adaptive window based on time dimension or other feature dimension. When the data resource demand of the window interval is equal to the threshold, the task order of the interval and the corresponding condition of the interval are obtained, and are associated with the current task order corresponding to the instruction ID, and are uploaded to the task order block.

[0026] Specifically, according to the shunting factor of the shiftable order being a resource demand threshold or a feature intensity threshold, adaptive window or feature intensity threshold analysis is applied to data sharding on the to-be-processed data of the task order, condition or primary key information corresponding to the data shard is obtained, a split task order is generated, associated with the original task order, and uploaded to the task order block, and further comprising:

[0027] The shunting factor of the shuntable order is a feature intensity threshold, a data feature curve is generated, a baseline curve is compared, and a threshold interval set T0 exceeding the threshold is extracted as D_11 time, and a remaining set T1 is used as a D_12 time condition to associate the original task order, and uploaded to the task order block.

[0028] Specifically, according to the shuntable order, the shunting factor is a resource demand threshold or a feature intensity threshold, and a self-adaptive window or a feature intensity threshold analysis is applied to data sharding on the task order to be processed data, to obtain condition or primary key information corresponding to the data sharding, to generate a split task order, to associate the original task order, and to upload to the task order block, and further comprising:

[0029] The shunting factor of the shuntable order is a feature intensity threshold, and the key features and derived features of the task order processing data are statistically clustered, and the boundary box is generated by deviating from the historical center and exceeding the threshold, and the high-intensity interval of the features is extracted as D_21 to D_2N, and the original task order is associated and distributed to the cloud node, and the low-intensity interval of the remaining features is associated and distributed to the edge node.

[0030] Further, the intelligent agent scheduling management module adopts an improved adaptive genetic algorithm, and generates a task distribution block based on the task order and the available intelligent agent set, and further comprising:

[0031] The task order is encoded as a chromosome length m*n to generate an initial population, and the fitness function Fitness is R minus the completion time Fj, wherein R is k times the maximum Fj;

[0032] A group wheel selection two-point crossover reverse mutation is adopted, a nonlinear adaptive operator pc and pm are combined according to the population fitness change, an elite reservation iteration is combined, and an optimal scheduling scheme is output to upload the task distribution block.

[0033] On the other hand, the embodiment of the present application also discloses an enterprise intelligent agent low-code arrangement and version management platform, which is applied to any one of the enterprise intelligent agent low-code arrangement and version management methods, and comprises:

[0034] The alliance chain management module is used for accessing and managing the cloud node, the edge node, the intelligent agent management system node, the data acquisition node of each Internet of Things device, and each application system node;

[0035] The intelligent agent registration module is used for automatically installing various intelligent agents on each cloud or edge, and registering and uploading the intelligent agent registration block;

[0036] The intelligent agent registration module is used for automatically installing various intelligent agents on each cloud or edge, and registering and uploading the intelligent agent registration block;

[0037] The data shunting processing module is used for adaptive data segmentation of task order to-be-processed data.

[0038] The intelligent agent management module is used for generation management of intelligent agent scheduling flow nodes.

[0039] The intelligent agent scheduling module is used for intelligent dispatching of task orders.

[0040] The instruction configuration module is used for configuration generation of instructions triggered by an application system in the future.

[0041] The intelligent agent arrangement module is used for intelligent agent arrangement management of instructions generated by an application system based on business requirements.

[0042] The intelligent agent version management module is used for management and viewing of arranged intelligent agent flows corresponding to instructions.

[0043] Compared with the prior art, the present application has the following beneficial effects:

[0044] The present application aims at how to realize efficient intelligent agent scheduling management and effective application of cloud and edge resources in complex business scenarios of enterprises. By constructing the alliance chain underlying management module of each node, the fast transmission and consensus of the information of each node are realized, and the intelligent contract realizes the orderly allocation of the resources of each node. Based on the intelligent agent management system, adaptive data segmentation of tasks can be realized, a large task is allocated to multiple nodes for execution, and the utilization rate of the resources of each node is improved. By using the improved adaptive genetic algorithm for intelligent agent scheduling optimization of the edge node intelligent agent, an approximately optimal scheduling scheme is quickly found under complex constraints and multiple targets. The availability and scalability indicators of the intelligent agent are calculated by monitoring the resource usage state of each node and the intelligent agent scheduling frequency, dynamic analysis is performed, and the related intelligent agent is automatically registered or unregistered at each stage, the resources of each node and intelligent agent are effectively applied, and enterprise server resources are saved. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 A flowchart of an enterprise intelligent agent low-code arrangement and version management platform and method.

[0046] Figure 2 A genetic algorithm flowchart of an enterprise intelligent agent low-code arrangement and version management platform and method. DETAILED DESCRIPTION

[0047] The technical solutions of the present application will be described below in conjunction with embodiments, obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0048] The enterprise intelligent agent low-code orchestration and version management platform and method of the embodiment can specifically include:

[0049] First of all, it needs to be pointed out that the conventional order: the task execution of an order can only be executed by one intelligent agent, and one intelligent agent can only execute one order task at the same time; the shiftable order: usually, the data volume needs to be dispersed to various intelligent agents for task execution, or there is a feature intensity that is relatively large and needs to be allocated to the cloud for task execution, and the same type of intelligent agent of multiple service nodes can be selected to simultaneously divide and work on data processing, and one intelligent agent can only execute one order task at the same time. The shiftable order can be based on resource threshold shunting and feature intensity threshold shunting to perform data slicing processing in advance; the task order: the order as mentioned below refers to an instruction order initiated by the system to require intelligent agents to execute business; each intelligent agent scheduling flow node can be called by multiple instruction ID corresponding task orders at the same time.

[0050] In addition, the alliance chain is a blockchain network jointly managed and maintained by a plurality of organizations that are pre-screened, limited in number, and have business associations with each other. The permissions (such as accounting, reading, and writing) in the network are not open to everyone, but are distributed according to the rules negotiated by the alliance members. The essence of the alliance chain is a "cooperative" distributed ledger technology, and the smart contract in the alliance chain realizes the high performance, fine control, and compliance required by the application. It sacrifices part of the openness and complete decentralization, and the application of the smart contract realizes higher efficiency, better privacy protection, stronger controllability, and compliance.

[0051] The cloud node, the various edge nodes, the intelligent agent management system, the data acquisition nodes of various Internet of Things devices, and various application system nodes form an alliance chain. The technical architecture based on the alliance chain system can realize the rapid transmission and consensus of information of various nodes, and the application of the blockchain smart contract can realize the orderly allocation of resources of various nodes, avoiding the scheduling conflict problem of node and intelligent agent resources due to data transmission delay. The orchestration technology of the present application scheme uses intelligent agents that preferentially use edge nodes, and starts cloud intelligent agents to realize instruction task orders that cannot be realized by edge intelligent agents, effectively applying cloud and edge resources.

[0052] The cloud node or the edge node implants an intelligent agent, and uploads intelligent agent registration information to a block of an alliance chain, and an intelligent agent management system obtains the registration information to generate or associate intelligent agent scheduling flow nodes, to form a set of intelligent agent scheduling flow nodes, specifically including:

[0053] The cloud node or each edge node implants a new intelligent agent according to the business needs of the enterprise. When the cloud node or each edge node has a new intelligent agent implanted, the node that adds the intelligent agent automatically uploads the intelligent agent registration information of the node IP, intelligent agent name, intelligent agent type and corresponding function description and the like deployed by the intelligent agent to the alliance chain block and consensus to each alliance chain node. After the intelligent agent management system obtains the intelligent agent registration information block, the intelligent agent scheduling process node based on the intelligent agent name is generated, and is associated with the current intelligent agent registration information. If the intelligent agent scheduling process node with the intelligent agent name already exists, the intelligent agent scheduling process node is not generated, and the current intelligent agent registration information is directly associated with the corresponding intelligent agent scheduling process node, thereby forming an intelligent agent scheduling process node set.

[0054] The intelligent agent management system dynamically evaluates the availability and scalability of the intelligent agent of each node based on the real-time obtained remaining resource information and intelligent agent working state information of each node, automatically registers or deregisters the intelligent agent corresponding to each node, and completes the automatic reduction and expansion of the intelligent agent of each node, specifically including:

[0055] The intelligent agent management system counts the use frequency of each type of intelligent agent, the remaining resources of each node corresponding to the intelligent agent, and the use frequency of other intelligent agents of each node corresponding to the intelligent agent, obtains the availability index of the current intelligent agent in a node, and specifically as follows:

[0056] The test time period is in units of S hours;

[0057] P is the number of times of calling the current intelligent agent in a time unit of S hours;

[0058] The set H is a set H1, H2, H3...Hn of the minimum remaining resources of each edge node where the current intelligent agent is located in S hours;

[0059] The set L is a set L1, L2, L3...Ln of the average number of times of using other intelligent agents of each edge node where the current intelligent agent is located in S hours;

[0060] The availability set K of the current intelligent agent in each node is (P / L)*H, and K is a set K1, K2, K3...Kn;

[0061] When there is a value less than the pre-set reduction threshold in K1, K2, K3...Kn, the number of values less than the pre-set reduction threshold in the statistical set K is counted Z, such as Z equal to K, then the intelligent agent of one node is reserved, and the intelligent agents of other nodes are automatically logged out, each node uploads the current intelligent agent logout information to the intelligent agent logout block, and is consensus to each node of the alliance chain; if Z is greater than 0 and less than K, the intelligent agents of the nodes less than the reduction threshold in the set K are automatically logged out, each node uploads the current intelligent agent logout information to the intelligent agent logout block, and is consensus to each node of the alliance chain; if it is equal to 0, then each node of the current intelligent agent does not need to log out the current intelligent agent.

[0062] In addition, the intelligent agent management system also analyzes the scalability of each intelligent agent, specifically as follows:

[0063] Y1 is the minimum number of nodes that can be expanded;

[0064] The set N is the set N1, N2, N3...Nn of the minimum remaining resources of each edge node that has not registered the current intelligent agent in S hours;

[0065] When there is a value greater than the pre-set expansion threshold in K1, K2, K3...Kn, the number of values greater than the pre-set expansion threshold in the statistical set K is counted Y, such as Y less than Y1, the current intelligent agent scalability is 0, and there is no need to register the current intelligent agent in each node; if Y is greater than or equal to Y1, then Y1 nodes that have not registered the current intelligent agent can be registered, the platform automatically selects the largest node from the set N to automatically register the current intelligent agent, and each node uploads the current intelligent agent registration information to the intelligent agent logout block and is consensus to each node of the alliance chain.

[0066] As described above, the resource usage state of each node and the intelligent agent scheduling frequency are monitored, the availability and scalability indicators of the intelligent agent are calculated, dynamic analysis is performed, and the related intelligent agent is automatically registered or logged out at each stage, so that the node and intelligent agent resources are effectively applied, and the enterprise server resources are saved.

[0067] Further, the application system node configures a transmission instruction ID and its order type, and uploads it to the alliance chain instruction ID block, the intelligent agent management system node downloads the instruction ID block, and arranges an intelligent agent workflow for the instruction ID based on the intelligent agent scheduling process node set, specifically including:

[0068] Based on the needs of enterprise business links, configure transmission instructions at the business operation link of the application system node, and generate a unique instruction ID for it. Based on the data generated by the business, mark the order type corresponding to the current instruction ID as a regular order or a shuntable order. If it is a shuntable order, it also includes the configuration of the shunt factor, which includes resource demand threshold-based shunting and feature intensity threshold-based shunting. Upload the instruction ID related information to the alliance chain instruction ID block.

[0069] The agent management system node downloads the instruction ID block. Based on the needs of enterprise business instruction ID processing, the corresponding agent workflow is arranged on the agent management system, the instruction ID is selected, the name of the agent scheduling process node called by the current to-be-processed instruction ID order is configured, the data input interface name of the current to-be-processed business is selected as the front-end input of the agent, and the data output interface name after data processing of the current business is selected as the back-end output of the agent. The agent management system node uploads this configuration information and configuration time information to the process configuration block and consensus to each node of the alliance chain. The data blocks of each node consensus are minimized before and after the processing of the specific task order, which improves the efficiency of data transmission and saves network resources.

[0070] As described above, based on business needs, the corresponding agent call configuration is arranged for each instruction ID on the agent management system node. When calling the process block corresponding to the instruction ID, the system automatically calls the instruction ID process configuration block information generated at the latest time.

[0071] Further, the application system node triggers the instruction ID and uploads the task order information to the alliance chain task order block. The agent management system downloads the block data, parses the instruction ID, obtains the agent scheduling process node and order type, and filters the available agent set. Specifically, it includes:

[0072] The application system node triggers the instruction at a certain business link, for example, triggers instruction ID1. The application system node uploads instruction ID1, judges the task order information corresponding to the current instruction ID1, such as the current order being a regular order. Upload the current instruction ID1 and the key information corresponding to the current task order, such as the data primary key that needs to be processed by the agent, or the corresponding time range, etc. These key information is used as the condition for obtaining the interface data of the above-mentioned to-be-processed business data input interface. If the current order is a shuntable order, in addition to uploading the above-mentioned key information corresponding to the task order, the capacity of the data corresponding to the task order also needs to be uploaded, and then the required computing power demand or resource quantity demand of the agent processing this task order is calculated. The application system node uploads the above-mentioned information to the task order block and consensus to each node of the alliance chain.

[0073] As described above, based on the block data uploaded by the instruction ID, the agent management system node parses the key features of the agent to-be-processed data instead of the data itself, reduces the frequency of large batch data transmission, improves efficiency, and reduces the waste of invalid resources.

[0074] In addition, it should be noted that the cloud node and the edge node server real-time statistics of the latest remaining computing power, remaining memory and other remaining resource information, if there is fluctuation, and upload the latest remaining resource information to the remaining resource information block, and consensus to each node of the alliance chain; if there is no fluctuation, no processing is performed.

[0075] It should be further noted that the agent of the cloud node and the edge node uploads the latest state information of the corresponding agent to the agent state block when starting execution or ending execution of the task, including the IP of the cloud node or the edge node, the agent name, the working state information of the agent starting execution or ending execution, and the like, and consensus to each node of the alliance chain.

[0076] Further, the agent management system node real-time acquires the latest task order block, the latest remaining resource information block and the latest agent state block, parses the instruction ID in the task order block, acquires the agent scheduling flow node and the order type, and screens the available agent set. Specifically, it includes:

[0077] The agent management system node downloads the latest task order block, the latest remaining resource information block and the latest agent state block, real-time parses the downloaded task order block data, acquires the corresponding instruction ID1 and the information corresponding to the current task order, and acquires the pre-configured agent scheduling flow node corresponding to the instruction ID1 and the order type marked by the instruction ID1 according to the instruction ID1. According to the cloud node or the edge node to which the agent associated with the agent scheduling flow node belongs, and the latest remaining resource information of the cloud node and the edge node and the working state of the corresponding agent are acquired. The working state of the agent starting execution is marked as 1, the working state of the agent ending execution is marked as 0, and the agent whose node remaining resource is less than the pre-set threshold is marked as 1. Set as set M1, set M1 includes the corresponding cloud node IP or edge node IP, and the agent name. If set M1 is empty, the instruction ID1 waits for allocation; if set M1 is not empty, select the node agent set that meets the current task order demand, realize adaptive data segmentation of the task based on the agent management system, allocate a large task to multiple nodes for execution, and improve the utilization rate of each node resource.

[0078] If the order type corresponding to the current instruction ID is marked as a regular order, if the agent set M1 contains edge node agents, then all edge node agents in set M1 are the candidate set M. If the agent set M1 does not contain edge node agents, then the cloud node agents in set M1 are the candidate set M.

[0079] If the order type corresponding to the current instruction ID is marked as a shiftable order, the data shunting processing module of the agent management system obtains the condition information corresponding to the task order of the instruction ID and the data interface corresponding to the pre-configured access front-end input, obtains the data to be processed of the task order based on the condition, and simultaneously, according to the shunting factor information corresponding to the current instruction ID, first adopts adaptive window slicing to perform shunting processing on the data of the task order corresponding to the current instruction ID, specifically including:

[0080] If the shunting factor of the current instruction ID is a resource demand threshold, the information of the time dimension or other characteristic dimensions is obtained, for example, data slicing is performed through an adaptive time window, for example, the starting time of the time condition in the block information of the task order corresponding to the current instruction ID is taken as the starting time of the time interval, a time interval is gradually intercepted by moving backward, and the data capacity, processing data and other resource demands of the time interval are statistically calculated in real time, for example, if the resource demand is equal to the pre-set threshold, the moving backward expansion of the time interval is stopped, and the time interval is taken as the time condition of the first task order D1 split out by the first split task order D1 corresponding to the instruction ID, and the first task order D_01 is associated with the current task order corresponding to the instruction ID. The end time node in the time condition of the instruction ID is taken as the starting time node to continue moving backward, and similarly, the split task orders D_02, D_03...D_0N are obtained, and the splitting and shunting of the task order corresponding to the current instruction ID are completed, and the split task orders and the corresponding time condition information are uploaded to the task order block and are consensus to each node of the alliance chain.

[0081] If the shunting factor of the current instruction ID is a characteristic intensity threshold, the techniques for shunting the task order corresponding to the current instruction ID include but are not limited to:

[0082] The data shunting processing module of the agent management system generates a curve of the data features or derived features corresponding to the task order corresponding to the current instruction ID based on the time dimension, and generates a baseline curve based on historical data. For example, based on the time dimension data segmentation, when the value of the data features or derived features corresponding to the task order corresponding to the current instruction ID exceeds the pre-set threshold at this time point, the time point is taken as the starting time point, and when the value of the data features or derived features corresponding to the task order corresponding to the current instruction ID returns to the baseline time point, the time point is taken as the ending time point, and the time interval is intercepted and added to the feature intensity threshold time interval set T0, and then all time intervals in which the value of the data features or derived features corresponding to the task order corresponding to the current instruction ID exceeds the threshold are obtained and added to T0, T0 is taken as the time condition corresponding to the task order D_11 split by the current instruction ID, and the task order D_11 is associated with the current task order corresponding to the instruction ID. For the time interval corresponding to the instruction ID and not added to T0, the remaining time interval is taken as T1, T0 is taken as the time condition corresponding to the task order D_12 split by the current instruction ID, and the task order D_11 is associated with the current task order corresponding to the instruction ID. As described above, the data slicing processing is performed on the task order information corresponding to the current instruction ID, the task order D_11 and the corresponding time condition T0 after the task order corresponding to the instruction ID is split, and the task order D_12 and the corresponding time condition T1 are generated, the new task order block associated with the current task order corresponding to the instruction ID is generated, and is consensus to each node of the alliance chain.

[0083] In addition, the data shunting processing module of the intelligent agent management system obtains the region or data feature with abnormal feature intensity. The data shunting processing performs clustering statistics on the key features in the task order corresponding to the current instruction ID and the key features of the historical data or the model training data, judges whether the features in the current task order deviate from the set center of the key features of the historical data or the model training data beyond a threshold, and generates a bounding box with the features as the center in the task order of the image recognition type, obtains the pixels in the bounding box as the orders D_21 split from the current task order, associates the orders D_21 with the current task order corresponding to the instruction ID, and further splits the orders D_22, D_23,..., D_2N and associates them with the current task order corresponding to the instruction ID, and distributes the order information D_21, D_22, D_23,..., D_2N with the feature intensity greater than the threshold to the intelligent agents of the cloud nodes for queuing processing. The remaining information of the region or feature without abnormal feature intensity in the task order corresponding to the current instruction ID is taken as the order D_20, associated with the current task order corresponding to the instruction ID, and distributed to the intelligent agents of the edge nodes for processing. As described above, the intelligent agent management system uploads the split order information and the primary key information of the feature data and the distribution information distributed to the edge nodes or the cloud nodes to the task order block and consensus to each node of the alliance chain.

[0084] Further, the cloud node intelligent agent calls a large-scale pre-trained deep learning model to perform fine analysis on the local high-dimensional data uploaded to the cloud node. The cloud model has stronger feature expression and pattern recognition capability and can identify subtle and complex defects that the edge model cannot distinguish. The cloud model not only gives the final classification but also outputs the prediction uncertainty measure. For samples with high uncertainty, they can be automatically pushed to the artificial expert interface for review, and the review result is taken as new labeled data. Based on the difficult samples accumulated by the cloud and the expert labeling, a more powerful next-generation detection model is generated through collaborative training in a federated learning framework.

[0085] It should be noted that the split original task order no longer participates in the intelligent agent distribution and execution of the current process scheduling node, and when the split task orders associated with the original task order are all completed based on the task execution of this process node, the original task order is completed based on the task execution of this process node.

[0086] Further, based on the above instruction task order and split instruction task order and the state information of the current intelligent agent, the intelligent agent scheduling management module uses an improved adaptive genetic algorithm to perform intelligent agent scheduling optimization on the edge node intelligent agent, quickly finds an approximately optimal scheduling scheme under complex constraints and multiple objectives.

[0087] The following mathematical symbols are explained: M: node agent; T: execution time; Sch: scheduling plan; Optsch: shortest execution time scheme; ST: start execution time; ET: end execution time; Jobid: task order number; Procid: agent scheduling process node number; Fitness: fitness function; k: proportional coefficient; R: normalized intermediate value.

[0088] In the process of business execution, let M = {1, 2, 3, …, m} be the node agent, let N = {1, 2, 3, …, n} be the order, and let Sch = N = {1, 2, 3, …, m x n} be the set of all process nodes. The order of the process nodes before and after the constraint is given. For each order j, after the last step of processing is completed, and the next process i node is idle, the order can be processed. STj is the start time of j, ETj is the execution time of j, Cj is the completion time of j. The execution time of the order is different due to the different agent selection schemes. The best agent selection and execution process order scheme directly constitutes the optimal scheduling scheme.

[0089] The business execution agent scheduling problem generally needs to meet the following conditions: the execution order of all orders on the agent is given in advance, and the order should be executed on the agent according to the given order; at the same time, the execution of one process node of one order can only be executed on one agent, and one agent can only execute one process node business of one order. The execution priority of each order is the same, and the order can be completed only after all execution tasks are completed; except for special cases, the execution time of the order is determined and remains unchanged. Order j is executed on agent m, denoted as = 1, otherwise = 0. The specific mathematical model is:

[0090] (1)

[0091] (2)

[0092] (3)

[0093] (4)

[0094] (5)

[0095] Formula (1) represents the shortest completion time of Optsch; formula (2) represents that each order j must be executed continuously once started and cannot be terminated in advance; formula (3) represents that each agent performs at most one order at the same time; and constraint conditions formula (4) and formula (5) ensure the accuracy and rationality of the processing process.

[0096] In the business agent scheduling problem, the order execution sequence can be regarded as the chromosome arrangement in genetic behavior, and the algorithm mainly changes the order execution sequence through chromosome population iteration to complete the optimization task.

[0097] The design of the algorithm includes the following steps: encoding and decoding, generating an initial population, chromosome selection, crossover and mutation, and algorithm termination, which are as follows:

[0098] (1) Chromosome encoding and decoding. Since the dimensions of the genetic algorithm solution and the business agent scheduling solution are different, the scheduling scheme can be corresponded to a chromosome through encoding. The number of agent nodes is m, the number of orders is n, and therefore the length of the encoding is m x n. The encoding is randomly arranged through the randperm function as an initial feasible solution. pop initial solutions form a whole to constitute an initial population.

[0099] (2) Design of fitness function. In the business agent execution scheduling problem, the final solution target is to make the completion time shortest, and the fitness function is selected in the following manner: record each chromosome code, the completion time of chromosome i , select the maximum completion time max , and expand it to a certain proportion coefficient (design as 1.2-1.5 times). The proportion coefficient is k, and then obtain the intermediate reference value R, which is subtracted from the completion time of each chromosome. The greater the difference, the higher the fitness value of the chromosome, that is, the shorter the completion time. Therefore, the fitness function is designed as:

[0100] (6)

[0101] (7)

[0102] (3) Generation of initial population. The intergroup roulette strategy is adopted, so that individuals with higher fitness values are more likely to be selected as parents to generate new individuals, thereby guiding the optimization direction of the algorithm.

[0103] (4) Chromosome selection. The chromosome selection is in the form of normalized probability roulette. By generating a random number, it is determined which interval it falls into to decide which chromosome to copy. From the existing pop population, pop chromosomes are selected again through the roulette betting form to form a new population.

[0104] (5) Genetic operations. The crossover operation uses a two-point crossover method to randomly generate two points. , This results in two paternal chromosomes. , exist , The exchange of gene segments between the two points forms two new chromosomes. and and to and The algorithm performs validity processing by comparing the two exchanged segments, retaining the code that cannot be canceled out and replacing it sequentially. The mutation method is reverse mutation, which randomly generates two points on the chromosome, and reverses the chromosome segments between the two points to achieve the mutation effect. The algorithm terminates when the maximum number of iterations is reached.

[0105] (6) This invention employs a novel, nonlinearly varying adaptive crossover and mutation operator improvement strategy. The operator changes according to the fitness value of the population, maintaining diversity in the early stages of the population while preventing the destruction of superior individuals in the later stages. The crossover operator (pc) and mutation operator (pm) are designed as follows: (8)

[0106] Where Pc is the crossover probability of the current individual; Pc1 is the maximum crossover probability; Pc2 is the minimum crossover probability; f is the fitness value of the current individual; favg is the average fitness value of all individuals in the current population; fmax is the maximum fitness value in the current population; and β is the adaptive adjustment factor.

[0107] (9)

[0108] Where Pm is the mutation probability of the current individual; Pm1 is the maximum mutation probability; Pm2 is the minimum mutation probability; f, favg, fmax, and β have the same meaning as in formula (8).

[0109] Explanation of the function of the parameters:

[0110] (1) Pc1, Pc2, Pm1, Pm2: control the upper and lower bounds of crossover and mutation probabilities to prevent the probabilities from being too high or too low.

[0111] (2) F, favg, fmax: Reflect the degree of superiority or inferiority of an individual in the population, used to dynamically adjust probabilities:

[0112] If the individual fitness f is close to fmax (good individual), then β approaches 0, Pc and Pm decrease, thus protecting the good solution;

[0113] If the individual fitness f is low (poor individual), then beta tends to 1, and Pc and Pm increase, enhancing the search ability.

[0114] (3) : for judging the convergence state of the population:

[0115] If the value is less than or equal to 0.5, it indicates that the population has not converged, and self-adaptive adjustment is adopted.

[0116] If the value is greater than 0.5, it indicates that the population has tended to converge, and the maximum probability Pc1 or Pm1 is fixedly used to prevent falling into a local optimum.

[0117] In formula (8) and formula (9), no matter the size of the current mutation probability, the improved crossover mutation rate is not equal to 0, and in the early stage of evolution, it is basically impossible to fall into a local optimal solution. With the iteration of the population, the crossover mutation probability gradually increases, the search speed is accelerated, and the optimal solution can be found as soon as possible. When the current individual is close to the optimal individual, the value of the crossover mutation probability gradually decreases, avoiding the current individual from falling into a local optimal solution. The operator not only can retain excellent coding schemes in iteration, but also can jump out of a local optimum in solving the business intelligent agent scheduling problem, and has good effect in solving the business intelligent agent scheduling problem. Meanwhile, the elite reservation strategy is applied to the algorithm, that is, the best chromosome in the current generation is reserved, so as to prevent the optimal individual from being lost or damaged due to crossover and mutation, thereby accelerating the evolution process. The application process of the improved adaptive genetic algorithm is as shown in Figure 2 , and finally the optimal scheduling scheme is obtained, and then the allocation scheme of the real-time task instruction is determined.

[0118] The intelligent agent management system node uploads the final task order allocation data to the task allocation block, and consensus to each alliance chain node, each cloud node or edge node downloads the task allocation block data, and parses the corresponding task allocation data, such as the current task allocation data has the corresponding intelligent agent task information allocated to the current node, then the corresponding task order enters the current intelligent agent execution queue, waits for the execution of the previous task order to be completed, when the current task order enters the execution, the intelligent agent obtains the task order block information corresponding to the current task order, obtains the interface condition information of the corresponding to-be-processed data based on the task order block information, including but not limited to the time condition, the feature data primary key information in the above, and then based on the information, calls the interface data of the pre-configured data input interface, executes the task and outputs the execution result data to the pre-configured data output interface, and uploads the execution result together with the task instruction ID to the task instruction result block, and consensus to each alliance chain node. The corresponding node downloads the task instruction result block for the next business task.

[0119] The application adopts a genetic algorithm to allocate an agent capable of executing a current parallel task instruction order in real time, avoids queuing and waiting in an edge node, distributes resource requests in multiple nodes, effectively utilizes intelligent agent resources of each server, and improves service execution efficiency.

[0120] Further, each node downloads a task allocation block, a corresponding intelligent agent acquires interface data to execute a task, and uploads a result to a task result block, and specifically includes:

[0121] Each node downloads and parses the task allocation block, and if the allocation is to the current node, the task order enters an execution queue of the intelligent agent corresponding to the current node.

[0122] The execution intelligent agent of the current node acquires condition information and primary key information in the task order block, acquires data to be processed by the task order based on a data input interface pre-configured based on an instruction ID corresponding to the task order, the intelligent agent executes a task, outputs result data of the executed task to a data output interface pre-configured based on the instruction ID, and regularly clears processed data to avoid resource waste.

[0123] The execution intelligent agent of the current node uploads a processing result state to a corresponding task execution result block, and consensus to each node of the alliance chain. If the order is not associated with an original task order, a current task order result block is generated, and if it is associated, it is judged that all split task orders are completed, and an original task order result block is generated.

[0124] Specifically, based on a task order processed by an edge node intelligent agent or a cloud node intelligent agent, if the current order is not associated with an original task order of an instruction ID, the current order completes data processing based on the intelligent agent process, and if the current order is associated with the original task order of the instruction ID, it is judged whether all task orders associated with the original task order are completed based on the intelligent agent process, if all are completed, a task instruction result block of the original task order of the instruction ID is generated, and consensus to each node of the alliance chain, and if there is an incomplete task order, the original task order is not processed.

[0125] It should be noted that the above enumeration is only a few specific embodiments of the present application. Obviously, the present application is not limited to the above embodiments, and there are many variations. All variations that can be directly derived or inferred by those skilled in the art from the content disclosed by the present application should be considered as the protection scope of the present application.

Claims

1. A method for enterprise intelligent agent low-code orchestration and version management, characterized in that, The method comprises: The cloud node or edge node implants an intelligent agent, and uploads intelligent agent registration information to a consortium chain intelligent agent registration block, and an intelligent agent management system obtains the registration information to generate or associate intelligent agent scheduling process nodes, forming an intelligent agent scheduling process node set; The application system node configures a transmission instruction ID and its order type, and uploads it to a consortium chain instruction ID block, and the intelligent agent management system node downloads the instruction ID block, and arranges an intelligent agent workflow for the instruction ID based on the intelligent agent scheduling process node set; The application system node triggers the instruction ID, uploads task order information to a consortium chain task order block, the intelligent agent management system downloads the instruction ID block data, parses the instruction ID, obtains the intelligent agent scheduling process node and the order type, and screens an available intelligent agent set; The intelligent agent scheduling management module adopts an improved adaptive genetic algorithm, optimizes scheduling to generate a task allocation block based on a task order and a corresponding available intelligent agent set; wherein it further comprises: encoding the task order into a chromosome length m x n to generate an initial population, and an adaptability function Fitness is R minus a completion time Fj, wherein R is k times the maximum Fj, and k is a preset proportion coefficient; A two-point crossover reverse sequence mutation is adopted, a nonlinear adaptive operator pc and pm are combined with elite retention iteration according to population adaptability changes, and an optimal scheduling scheme is output to upload a task allocation block until termination; Each node downloads the task allocation block, the corresponding intelligent agent obtains interface data to execute a task, and uploads the result to a task result block.

2. The method of claim 1, wherein, The cloud node or edge node implants an intelligent agent, and uploads intelligent agent registration information to a consortium chain block, and an intelligent agent management system obtains the registration information to generate or associate intelligent agent scheduling process nodes, forming an intelligent agent scheduling process node set, comprising: The cloud node or edge node implants an intelligent agent, and the intelligent agent management system generates corresponding intelligent agent scheduling process nodes for the same intelligent agent implanted into each node, forming an intelligent agent scheduling process node set; The intelligent agent management system dynamically evaluates the availability and scalability of intelligent agents of each node based on real-time obtained remaining resource information and intelligent agent working state information of each node, automatically registers or deregisters intelligent agents corresponding to each node, and completes automatic reduction and expansion of intelligent agents of each node.

3. The method of claim 1, wherein, The application system node configures a transmission instruction ID and its order type, and uploads it to a consortium chain instruction ID block, and the intelligent agent management system node downloads the instruction ID block, and arranges an intelligent agent workflow for the instruction ID based on the intelligent agent scheduling process node set, comprising: The application system node configures a transmission instruction ID, defines the order type corresponding to the current instruction ID based on the scene of the business instruction, including a regular order and a shiftable order, in addition, the shiftable order further includes resource threshold-based shunting and feature intensity threshold-based shunting; The intelligent agent management system node arranges an intelligent agent workflow for the task order type corresponding to the instruction ID, and uploads a process configuration block, and the data block consensus of each node is minimized, improving the efficiency of data transmission and saving network resources.

4. The method of claim 1, wherein, The application system node triggers the instruction ID, uploads the task order information to the alliance chain task order block, the intelligent agent management system downloads the instruction ID block data, analyzes the instruction ID, obtains the intelligent agent scheduling process node and the order type, and screens the available intelligent agent set, including: The cloud node and the edge node server real-time statistics of the latest remaining computing power, remaining memory and other remaining resource information, if there is fluctuation, and upload the latest remaining resource information to the remaining resource information block, and consensus to each node of the alliance chain; The cloud node and the edge node server real-time obtain all intelligent agent working state information of the node, and generate intelligent agent state block, and consensus to each node of the alliance chain; The intelligent agent management system node real-time obtains the latest task order block, the latest remaining resource information block and the latest intelligent agent state block, analyzes the instruction ID in the task order block, obtains the intelligent agent scheduling process node and the order type, and screens the available intelligent agent set.

5. The method of claim 4, wherein, The intelligent agent management system node real-time obtains the latest task order block, the latest remaining resource information block and the latest intelligent agent state block, analyzes the instruction ID in the task order block, obtains the intelligent agent scheduling process node and the order type, and screens the available intelligent agent set, including: When the order type is a regular order, the available intelligent agent set is screened, and when the order type is a shunt order, the data shunt processing module of the intelligent agent management system obtains the to-be-processed data of the task order based on the data interface corresponding to the pre-configured front input module of the instruction ID and the condition information corresponding to the current task order before screening the available intelligent agent set; Further, according to the shunt factor of the shunt order, the resource demand threshold or the feature intensity threshold is judged, the adaptive window or the feature intensity threshold analysis is applied, the data slicing of the to-be-processed data of the task order is performed, the condition or the primary key information corresponding to the data slicing is obtained, the split task order is generated, is associated with the original task order, and is uploaded to the task order block.

6. The method of claim 5, wherein, According to the shunt factor of the shunt order, the resource demand threshold or the feature intensity threshold is judged, the adaptive window or the feature intensity threshold analysis is applied, the data slicing of the to-be-processed data of the task order is performed, the condition or the primary key information corresponding to the data slicing is obtained, the split task order is generated, is associated with the original task order, and is uploaded to the task order block, including: The shunt factor of the shunt order is the resource demand threshold, the adaptive window is used to perform data slicing on the to-be-processed data of the task order based on the time dimension or other feature dimensions, when the data resource demand of the window interval is equal to the threshold, the task order of the interval and the corresponding condition of the interval are obtained, and are associated with the current task order corresponding to the instruction ID, and are uploaded to the task order block.

7. The method of claim 5, wherein, According to the shunt factor of the shunt order, the resource demand threshold or the feature intensity threshold is judged, the adaptive window or the feature intensity threshold analysis is applied, the data slicing of the to-be-processed data of the task order is performed, the condition or the primary key information corresponding to the data slicing is obtained, the split task order is generated, is associated with the original task order, and is uploaded to the task order block, including: The shunting factor of the shuntable order is a feature intensity threshold, a data feature curve is generated, a baseline curve is compared, T0 of a threshold interval set exceeding the threshold is extracted as D_11 time, T1 of a remaining set is extracted as D_12 time condition, the original task order is associated, and uploaded to the task order block.

8. The method of claim 5, wherein, The shunting factor of the shuntable order is a feature intensity threshold, a data feature curve is generated, a baseline curve is compared, T0 of a threshold interval set exceeding the threshold is extracted as D_11 time, T1 of a remaining set is extracted as D_12 time condition, the original task order is associated, and uploaded to the task order block. The shunting factor of the shuntable order is a feature intensity threshold, a data feature curve is generated, a baseline curve is compared, T0 of a threshold interval set exceeding the threshold is extracted as D_11 time, T1 of a remaining set is extracted as D_12 time condition, the original task order is associated, and uploaded to the task order block.

9. An enterprise intelligent agent low-code orchestration and version management platform, applying the method of any one of claims 1 to 8, characterized in that, Including: The alliance chain management module is used for accessing and managing cloud nodes, various edge nodes, agent management system nodes, various Internet of Things data acquisition nodes, and various application system nodes; The agent registration module is used for automatically installing various agents on each cloud or edge and registering and uploading the agent registration block; The agent registration module is used for automatically installing various agents on each cloud or edge and registering and uploading the agent registration block; The data shunting processing module is used for adaptive data segmentation of task order processing data; The agent management module is used for generating and managing the agent scheduling process node; The agent scheduling module is used for intelligent dispatching of task orders; The instruction configuration module is used for configuring and generating instructions triggered by the application system in the future; The agent scheduling module is used for intelligent dispatching of task orders; The agent version management module is used for managing and viewing the agent process arranged by the instructions.

Images