Multi-agent collaborative calling method and system based on big data

Abstract

The invention relates to the technical field of data processing, in particular to a multi-agent collaborative calling method and system based on big data, by sending supply chain data to a decision-making agent, the real-time state of supply chain operation can be comprehensively captured, data support is provided for decision-making of production plans and resource allocation, and the decision-making efficiency is improved. Through introduction of conflict detection of allocated resources and analysis of task completion degree, reference data is fed back and updated, so that a decision-making agent performs iterative optimization on first control data according to data dimensions and reference data with richer content, the resource conflict condition between execution agents is eliminated, and the decision-making efficiency of the decision-making agent is improved. The completion degree of the production condition of the supply chain in the dimensions of efficiency, energy consumption and capacity is optimized, so that the second control data output by the decision-making agent can realize multi-target collaborative optimal calling, and more efficient and reliable control data is provided for the management of the supply chain; therefore, the production efficiency and the productivity of the supply chain are improved, and the production energy consumption is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a multi-agent collaborative call method and system based on big data. Background Art

[0002] In the field of supply chain management, the multi-agent collaborative call technology based on big data has become an important means to optimize the production situation.

[0003] The prior art usually uses traditional planning algorithms or simple rule engines to perform collaborative scheduling on supply chain multi-agents. For example, a traditional linear programming algorithm is used to formulate a production plan, and resource allocation is performed through fixed priority rules. However, these methods usually determine and optimize the production plan and resource allocation only for a single target such as efficiency, energy consumption, or production capacity when formulating the production plan and resource allocation. It is difficult to achieve a balance among multiple task targets such as efficiency, energy consumption, and production capacity, and there is a lack of an optimization mechanism for the production plan and resource allocation. When resource competition occurs among the executing agents, the control strategies of the production plan and resource allocation cannot be quickly adjusted, resulting in a low reliability of the production plan and resource allocation strategies, thus affecting the production efficiency, production capacity, and energy consumption of the supply chain.

[0004] Therefore, how to improve the reliability of multi-agent collaborative calls and the completion among multiple task targets has become an urgent problem to be solved. Summary of the Invention

[0005] In view of the above technical problems, the technical solution adopted by the present invention is a multi-agent collaborative call method based on big data. The multi-agent collaborative call method based on big data includes the following steps: S1. Send supply chain data as reference data to a decision-making agent, where the supply chain data includes business system data, device status data of each executing agent, and external environment data.

[0006] S2. The decision-making agent obtains first control data according to the received reference data, where the first control data includes first production plan data and first resource allocation data, and the first resource allocation data includes a plurality of allocated resources corresponding to each executing agent and the usage time period corresponding to each allocated resource.

[0007] S3. Based on the first resource allocation data, the relevance between each allocated resource and each task objective type, and the candidate importance corresponding to each executing agent, obtain the set of conflicting agent combinations, the set of task completion degrees, and the target completion result corresponding to the first resource allocation data, where the task objective types include target efficiency, target energy consumption, and target production capacity. The set of conflicting agent combinations includes several conflicting agent combinations composed of two executing agents with resource conflicts. The set of task completion degrees includes the task completion degree of each task objective type, and the target completion result is either the target is completed or the target is not completed.

[0008] S4. If the set of conflicting agent combinations is a non-empty set or the target completion result is that the target is not completed, then send the set of conflicting agent combinations, the set of task completion degrees, and the supply chain data together as reference data to the decision-making agent, and repeat step S2 until the set of conflicting agent combinations is an empty set and the target completion result is that the target is completed. Determine the first control data at the time of terminating the repeated execution as the second control data.

[0009] S5. Control the behavior of each executing agent according to the second control data.

[0010] The present invention also provides a multi-agent collaborative invocation system based on big data. The multi-agent collaborative invocation system based on big data includes: A data sending module, configured to send the supply chain data as reference data to the decision-making agent, where the supply chain data includes business system data, the device status data of each executing agent, and external environment data.

[0011] A first control data acquisition module, configured to enable the decision-making agent to obtain the first control data according to the received reference data, where the first control data includes the first production plan data and the first resource allocation data, and the first resource allocation data includes several allocated resources corresponding to each executing agent and the usage time period corresponding to each allocated resource.

[0012] A control data analysis module, configured to obtain the set of conflicting agent combinations, the set of task completion degrees, and the target completion result corresponding to the first resource allocation data according to the first resource allocation data, the relevance between each allocated resource and each task objective type, and the candidate importance corresponding to each executing agent, where the task objective types include target efficiency, target energy consumption, and target production capacity. The set of conflicting agent combinations includes several conflicting agent combinations composed of two executing agents with resource conflicts. The set of task completion degrees includes the task completion degree of each task objective type, and the target completion result is either the target is completed or the target is not completed.

[0013] The second control data acquisition module is configured to, if the conflict agent combination set is a non-empty set or the target completion result is an uncompleted target, send the conflict agent combination set, the task completion degree set, and the supply chain data to the decision-making agent as reference data, and repeatedly execute the first control data acquisition module until the conflict agent combination set is an empty set and the target completion result is a completed target, and determine the first control data at the time of terminating the repeated execution as the second control data.

[0014] The execution agent control module is configured to control the behavior of each execution agent according to the second control data.

[0015] The present invention has at least the following beneficial effects: By sending business system data, device status data, and external environment data to the decision-making agent, the decision-making agent can comprehensively capture the real-time operation status of the supply chain, providing rich and accurate data support for the decision-making control of production planning and resource allocation; By introducing conflict detection based on resource allocation among execution agents and analysis of task completion degrees for task target types, the reference data received by the decision-making agent is fed back and updated, enabling the decision-making agent to iteratively optimize the output first control data based on reference data with richer data dimensions and content, not only eliminating resource conflict situations among execution agents, but also optimizing the completion degrees of the supply chain production in terms of efficiency, energy consumption, and production capacity, enabling the second control data finally output by the decision-making agent to achieve collaborative optimal invocation of multiple targets, providing more efficient and reliable control data for the management of the supply chain, thereby improving the production efficiency and production capacity of the supply chain and reducing production energy consumption. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a flowchart of a multi-agent collaborative invocation method based on big data provided in Embodiment 1 of the present invention; Figure 2 It is a schematic diagram of a multi-agent collaborative invocation system based on big data provided in Embodiment 2 of the present invention. Detailed Embodiments

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It can be understood that, under appropriate circumstances, the above-mentioned terms used to distinguish similar objects can be interchanged, so that the present invention can also implement other embodiments other than the above-mentioned illustrated embodiments or described embodiments. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0020] Embodiment 1 Embodiment 1 of the present invention provides a method for collaborative invocation of multiple agents based on big data, as Figure 1 shown. The method for collaborative invocation of multiple agents based on big data includes the following steps: S1. Send the supply chain data as reference data to the decision-making agent, where the supply chain data includes business system data, device status data of each execution agent, and external environment data.

[0021] In a specific implementation manner, S1 includes the following steps: S11. Collect the business data of each supply link in the supply chain, the device status data of each execution agent, and the external environment data through the data agent. Among them, the business data includes order data, inventory data, production process data, and logistics status. The device status data includes operating status, current task load, production capacity utilization rate, and device failure rate. The external environment data includes weather data, traffic data, and demand data.

[0022] S12. Compose the business system data according to the business data of all supply links in the supply chain.

[0023] Among them, the supply chain refers to a functional network chain structure that surrounds the core enterprise and connects suppliers, manufacturers, logistics providers, distributors, and end users into an integrated whole through the control of information flow, logistics, and capital flow from raw material procurement, through production manufacturing, product distribution, and other links. The supply chain includes multiple types of agents such as decision-making agents, data agents, and multiple execution agents.

[0024] Specifically, the decision-making agent is a software entity or system based on algorithms, responsible for integrating relevant data such as the input supply chain data, generating global control instructions to control each execution agent to execute relevant tasks, so as to complete the overall functional network chain of the supply chain.

[0025] The execution agent is the underlying execution unit in the supply chain. For example, several production devices corresponding to manufacturers, as well as several logistics devices and several warehousing devices corresponding to suppliers, manufacturers, logistics providers, and distributors, have the ability to receive control instructions and execute relevant tasks. For example, they can execute tasks such as production line assembly, goods sorting, and transportation route planning respectively. Among them, production devices can include intelligent machine tools, devices on automated production lines, industrial robots, etc., which are used to execute specific production tasks such as processing and assembly according to the production plan; logistics devices can include transportation vehicles, transportation drones, intelligent distribution terminals, etc., which are used to execute the physical movement tasks of raw materials, semi-finished products, and finished products among suppliers, manufacturers, logistics providers, distributors, and users; warehousing devices can include intelligent shelves, automatic sorting machines, automatic guided vehicles, etc., which are used to execute material storage, picking, and handling tasks.

[0026] The data agent is used to deploy data collection points at supply chain procurement, production, warehousing, logistics and other supply links, at each execution agent, and select several locations on the logistics route to deploy data collection points. At the deployed data collection points, business data, equipment status data, and external environment data are obtained through technologies such as API interfaces, web crawlers, vibration sensors, temperature sensors, pressure sensors, current sensors, voltage sensors, AGV / AMR status sensors, intelligent shelf sensors, vehicle on-board OBD systems, positioning systems, and meteorological sensors.

[0027] Among them, business system data is structured data from various supply links within the supply chain, such as order data (order quantity, submission time, etc.), inventory data (inventory data, production time, etc.), production process data (raw material consumption, process flow, production work orders, etc.), and logistics status (vehicle scheduling, route planning, loading rate), etc., which are used to characterize the business logic and requirements of the supply chain.

[0028] The equipment status data of the execution agent is real-time status information such as the running status, current task load, production capacity utilization rate, and equipment failure rate of the execution agent, which is used to characterize the production capacity and production situation of the supply chain.

[0029] External environment data is external factor data that affects the supply chain, such as weather data, traffic data, demand data, etc., which are used to characterize the risk resistance ability and market response speed of the supply chain.

[0030] As described above, by sending business system data, device status data, and external environment data to the decision-making intelligent agent as reference data, the real-time status of the supply chain operation can be comprehensively captured, providing rich and accurate data support for decision-making and improving the reliability of the system control of the subsequent execution intelligent agent.

[0031] S2. The decision-making intelligent agent obtains first control data according to the received reference data. The first control data includes first production plan data and first resource allocation data. The first resource allocation data includes several allocated resources corresponding to each execution intelligent agent and the usage time period corresponding to each allocated resource.

[0032] Among them, the first production plan data includes a first production task list corresponding to each execution intelligent agent, the first production quantity and the first production time period corresponding to each first production task in the first production task list, and is used to control each execution intelligent agent to execute the corresponding production task.

[0033] In a specific embodiment, the decision-making intelligent agent can be composed of an optimization algorithm (such as a genetic algorithm, a particle swarm optimization), a machine learning model (such as reinforcement learning, a neural network), or an intelligent decision-making platform (such as a supply chain management system (Supply Chain Management System, SCM), an AI middle platform), and has the capabilities of data processing, modeling analysis, policy generation, and dynamic adjustment. Those skilled in the art know that any optimization algorithm, machine learning model, and intelligent decision-making platform in the prior art fall within the protection scope of the present invention, and will not be elaborated herein.

[0034] Taking the decision-making intelligent agent based on the reinforcement learning architecture as an example, specifically, the first production plan data is obtained by analyzing multi-source supply chain data, and the continuous resource allocation problem (such as resource usage duration) is transformed into a discrete action space to reduce the control complexity. By maximizing the long-term cumulative reward (such as target efficiency, target energy consumption, target production capacity) through the trained policy model, a first resource allocation data scheme is mapped and selected to guide resource allocation.

[0035] In a specific embodiment, S2 includes the following steps: S21. Map the reference data into a state vector. The state vector includes a business system feature vector, a device status feature vector corresponding to each execution intelligent agent, an external environment feature vector, a conflict intelligent agent feature vector, and a task completion degree feature vector. When the decision-making intelligent agent has not received the conflict intelligent agent combination set and the task completion degree set, the conflict intelligent agent feature vector and the task completion degree feature vector are zero vectors.

[0036] S22. Divide the value range corresponding to each allocated resource into corresponding value intervals according to the preset discrete level number corresponding to each allocated resource.

[0037] S23. Obtain the discrete actions corresponding to each value range for each allocated resource, and obtain the action set corresponding to each allocated resource, where the action set is obtained from all the corresponding discrete actions.

[0038] S24. According to the action set corresponding to each allocated resource, obtain all the discrete action combinations, where the total number of discrete action combinations is the sum of the preset discrete level numbers corresponding to all the allocated resources.

[0039] S25. Input the state vector and all the discrete action combinations into the trained policy model corresponding to the decision-making agent, and obtain the first control data output by the decision-making agent.

[0040] Among them, according to the vector conversion method, the data of each dimension in the reference data is converted into the corresponding feature vector to obtain the business system feature vector, the device state feature vector corresponding to each execution agent, the external environment feature vector, the conflict agent feature vector, and the task completion degree feature vector. Specifically, the data included in the business data of each supply link is vector-converted, and the converted vectors are spliced to obtain the business system feature vector. For each execution agent, the data included in the corresponding device state data is vector-converted, and the converted vectors are spliced in a set order to obtain the device state feature vector corresponding to the execution agent. The data included in the external environment data is vector-converted, and the converted vectors are spliced in a set order to obtain the external environment feature vector. The set order can be set by the implementer according to the actual situation.

[0041] Since when the decision-making agent obtains the first control data for the first time, it has not received the conflict agent combination set and the task completion degree set data, correspondingly, the conflict agent feature vector and the task completion degree feature vector are zero vectors, and the length of each zero vector is the same as the length of the corresponding conflict agent feature vector or task completion degree feature vector.

[0042] Specifically, in this embodiment, a suitable vector conversion method is selected according to the data type. Those skilled in the art know that any vector conversion method in the prior art falls within the protection scope of the present invention. For example, the Word2Vec algorithm, the One-Hot algorithm, the convolutional neural network, etc. will not be elaborated here.

[0043] Too many discrete levels will lead to an explosion of the action space, while too few discrete levels will result in low accuracy of resource allocation. The specific number of preset discrete levels can be set by the implementer according to the actual situation. For example, different discrete precisions are set according to the importance of the allocated resources, and then the preset discrete levels are set.

[0044] By presetting the number of discrete levels, the value range corresponding to each allocated resource is divided into the corresponding number of value intervals, so as to transform the continuous resource allocation problem into a discrete action set and reduce the solution complexity. For example, for the allocated resource of parking lot A, the corresponding value range is [0h - 24h] per day, representing the range of continuous available time. Then, according to the preset discrete level number 6 corresponding to parking lot A, the value range is divided into 6 value intervals, corresponding to the [0h - 4h), [4h - 8h), [8h - 12h), [12h - 16h), [16h - 20h), [20h - 24h) per day. For the allocated resource of generator C, the corresponding value range is [8h - 20h] per day. Then, according to the discrete level number 4 corresponding to generator C, this value range is divided into 4 value intervals, corresponding to the [8h - 11h), [11h - 14h), [14h - 17h), [17h - 20h] per day.

[0045] Then, according to the value interval [0h - 4h) corresponding to parking lot A, the discrete action corresponding to parking lot A is obtained as (using parking lot A in the [4h - 8h) per day). Similarly, for the value interval [4h - 8h) corresponding to parking lot B, the discrete action corresponding to parking lot B is obtained as (using parking lot B in the [4h - 8h) per day). For the value interval [11h - 14h] corresponding to generator C, the discrete action corresponding to generator C is obtained as (using generator C in the [11h - 14h] per day).

[0046] For each discrete action corresponding to each value range according to each allocated resource, an action set corresponding to each allocated resource is statistically obtained. For example, the action set obtained for parking lot A is {(Use parking lot A during the period [0h - 4h) every day), (Use parking lot A during the period [4h - 8h) every day), (Use parking lot A during the period [8h - 12h) every day), (Use parking lot A during the period [12h - 16h) every day), (Use parking lot A during the period [16h - 20h) every day), (Use parking lot A during the period [20h - 24h) every day)}. The action set obtained for parking lot B is {(Use parking lot B during the period [0h - 4h) every day), (Use parking lot B during the period [4h - 8h) every day), (Use parking lot B during the period [8h - 12h) every day), (Use parking lot B during the period [12h - 16h) every day), (Use parking lot B during the period [16h - 20h) every day), (Use parking lot B during the period [20h - 24h) every day)}. The action set obtained for generator C is {(Use generator C during the period [8h - 11h) every day), (Use generator C during the period [11h - 14h) every day), (Use generator C during the period [14h - 17h) every day), (Use generator C during the period [17h - 20h] every day)}.

[0047] Furthermore, randomly select a discrete action from the action set corresponding to each allocated resource, and combine the discrete actions randomly selected from all allocated resources into a discrete action combination, that is, the number of discrete actions in each discrete action combination is the same as the total number of allocated resources. For example, randomly select the discrete action (Use parking lot A during the period [4h - 8h) every day) from the action set corresponding to parking lot A, randomly select the discrete action (Use parking lot B during the period [0h - 4h) every day) from the action set corresponding to parking lot B, and randomly select the discrete action (Use generator C during the period [8h - 11h) every day) from the action set corresponding to generator C, then a discrete action combination {(Use parking lot A during the period [4h - 8h) every day), (Use parking lot B during the period [0h - 4h) every day), (Use generator C during the period [8h - 11h) every day)} can be obtained.

[0048] Similarly, other discrete action combinations can also be obtained, such as {(Use parking lot A during the period [4h - 8h) every day), (Use parking lot B during the period [8h - 12h) every day), (Use generator C during the period [11h - 14h) every day)}, {(Use parking lot A during the period [0h - 4h) every day), (Use parking lot B during the period [8h - 12h) every day), (Use generator C during the period [14h - 17h) every day)}, etc. By traversing all discrete actions, all discrete action combinations can be obtained.

[0049] Input the state vector and all discrete action combinations into the trained policy model, where the trained policy model is obtained by training with historical data or a simulation environment, and the reward function in the training process designs three task objectives: efficiency, energy consumption, and production capacity, so as to output the first control data that simultaneously optimizes multi-dimensional objectives of efficiency, energy consumption, and production capacity. Among them, the trained policy model can be a deep Q-network (DQN), proximal policy optimization (PPO), or other models. Those skilled in the art know that the policy model and training method fall within the protection scope of the present invention and will not be elaborated here.

[0050] As described above, by analyzing multi-source supply chain data through a decision-making agent, the continuous resource allocation problem is transformed into a discrete action space, and three task objectives of balancing efficiency, energy consumption, and production capacity are designed in the policy model to obtain the first production plan data and the first resource allocation data scheme, thereby improving the balance and reliability of the first control data among multi-task objectives.

[0051] S3. According to the first resource allocation data, the correlation degree between each allocated resource and each task objective type, and the candidate importance degree corresponding to each executing agent, obtain the conflict agent combination set, task completion degree set, and target completion result corresponding to the first resource allocation data, where the task objective types include target efficiency, target energy consumption, and target production capacity. The conflict agent combination set includes several conflict agent combinations composed of two executing agents with resource conflicts, the task completion degree set includes the task completion degree of each task objective type, and the target completion result is either completed or not completed.

[0052] Among them, the target efficiency corresponds to a preset efficiency value, the target energy consumption corresponds to a preset energy consumption value, and the target production capacity corresponds to a preset production capacity size. The preset efficiency value, preset energy consumption value, and preset production capacity value can be set by the implementer according to actual production requirements and are used to represent the target requirements for the supply chain.

[0053] The influence of the allocation situations of different allocated resources on the completion degrees of target requirements such as target efficiency, target energy consumption, and target production capacity is different, which is reflected as the different correlation degrees between each allocated resource and each task objective type. Among them, the completion degree of target efficiency refers to the degree to which the production efficiency of the supply chain is greater than the preset efficiency value; the completion degree of target energy consumption refers to the degree to which the energy consumption of the supply chain is less than the preset energy consumption value; the completion degree of target production capacity refers to the degree to which the production capacity of the supply chain is greater than the preset production capacity value.

[0054] For example, different allocation scenarios of raw materials and electricity will significantly affect the waiting time and production time of the executing agents, thereby significantly affecting the production line utilization rate, and further significantly affecting the production efficiency and production capacity of the supply chain. Therefore, the degree of correlation between raw materials, electricity and the target efficiency and target production capacity is relatively high; while different raw material allocation scenarios have less impact on energy consumption, and different electricity allocation scenarios have less impact on energy consumption when the total electricity consumption is determined. Therefore, the degree of correlation between raw materials, electricity and the target energy consumption is relatively low.

[0055] The influence of different executing agents on the achievement levels of target requirements such as target efficiency, target energy consumption, and target production capacity varies, which is reflected as different candidate importance levels corresponding to each executing agent.

[0056] For example, production equipment with a new model and comprehensive functions has higher production speed, resource utilization efficiency, output, and yield rate, so it has a greater positive impact on optimization goals such as target efficiency, target energy consumption, and target production capacity. That is, its importance is higher when making the supply chain more efficient, with lower energy consumption, and higher production capacity. Therefore, the candidate importance level of production equipment with a new model and comprehensive functions is higher than that of old and single-function production equipment. Transport vehicles with strong load capacity and long cruising range have higher single transport volume, higher transport efficiency, longer maintenance cycle, and lower transport energy consumption, so they have a greater positive impact on optimization goals such as target efficiency, target energy consumption, and target production capacity. That is, their importance is higher when making the supply chain more efficient, with lower energy consumption, and higher production capacity. Therefore, the candidate importance level of transport vehicles with strong load capacity and long cruising range is higher than that of transport vehicles with weak load capacity and short cruising range. Further, the resource allocation situation of executing agents with a high candidate importance level and the corresponding task completion situation have a greater positive impact on the achievement levels of various optimization goal types.

[0057] The degree of correlation between each allocated resource and each task objective type, and the specific values of the candidate importance corresponding to each executing agent can be set by the implementer according to the actual situation. For example, by collecting the resource allocation records corresponding to each executing agent in the past and the completion data of the corresponding task objectives (such as efficiency indicators, energy consumption data, production capacity data), the correlation can be calculated through the Pearson correlation coefficient and the Spearman rank correlation coefficient, which is used to quantify the degree of correlation between the allocated resource and the task objective type, as well as the candidate importance corresponding to the executing agent. Or the domain experts can score the degree of association between the resource and the objective and the candidate importance corresponding to the executing agent, and the final degree of correlation and candidate importance can be obtained by correcting the results through multiple rounds of feedback. Or through a controlled variable experiment, under the condition that other conditions remain unchanged, the allocation of a certain type of allocated resource can be adjusted separately or the executing agent can be replaced separately, and the change in the completion degree of each task objective type can be observed, which is used to quantify the degree of correlation between the allocated resource and the task objective type, as well as the candidate importance corresponding to the executing agent. Those skilled in the art know that the calculation methods of the above-mentioned degree of correlation and candidate importance in the prior art fall within the protection scope of the present invention, and will not be elaborated here.

[0058] In this embodiment, according to the first resource allocation data, the degree of correlation between each allocated resource and each task objective type, and the candidate importance corresponding to each executing agent, the completion degree of the first allocation data for the task objectives of target efficiency, target energy consumption, and target production capacity is measured, and the task completion degree set and target completion result corresponding to the first resource allocation data are obtained and sent to the decision-making agent as feedback information, so as to improve the balance of resource allocation among multiple task objectives and improve the completion of multiple task objectives.

[0059] When multiple executing agents compete for the same resource, for example, when two production devices apply for the same batch of raw materials at the same time, or two logistics devices use the same shipping time window in the same warehouse, or two logistics devices use the same time window at the same parking point, or two production devices use the same power connection socket at the same time window, etc., resource conflicts will occur. Then, the conflict situation of the allocated resources among the executing agents is analyzed through the first resource allocation data, and the conflict agent combination set corresponding to the first resource allocation data is obtained and sent to the decision-making agent as feedback information, so as to improve the reliability of the control data.

[0060] In a specific embodiment, S3 includes the following steps: S31, obtain the degree of correlation between the resource type corresponding to each allocated resource and each task objective type, where the resource type includes production and manufacturing resources, warehousing and logistics resources, and human resources.

[0061] S32. Obtain the resource conflict results between any two executing agents according to the first resource allocation data, where the resource conflict results are either in conflict or not in conflict.

[0062] S33. Obtain the task completion degree corresponding to each task objective type according to the first resource allocation data, the relevance between each allocated resource and each task objective type, and the candidate importance corresponding to each executing agent.

[0063] S34. Obtain the conflict agent combination set, the task completion degree set, and the target completion result corresponding to the first resource allocation data according to the resource conflict results between any two executing agents and the task completion degree corresponding to each task objective type.

[0064] Among them, production and manufacturing resources can include resources such as raw materials and electricity, which are used for product production; warehousing and logistics resources can include resources such as parking lots, traffic lanes, real-time traffic condition APIs, TMS transportation management, vehicle-mounted weighing sensors, etc., which are used for route planning and determination of product loading rates; human resources can include resources such as worker resources, which are used to manually complete corresponding tasks.

[0065] In a specific embodiment, S32 includes the following steps: S321. According to the several allocated resources corresponding to each executing agent, determine the executing agent set corresponding to each allocated resource for each allocated resource.

[0066] S322. For any allocated resource, obtain the resource conflict results between any two executing agents in the executing agent set corresponding to the current allocated resource according to the usage time periods of any two executing agents in the executing agent set corresponding to the current allocated resource for the current allocated resource.

[0067] S323. Traverse all the allocated resources and all the executing agents in the executing agent set corresponding to each allocated resource to obtain the resource conflict results between any two executing agents.

[0068] Among them, if there is an overlap in the usage time periods of any two executing agents in the executing agent set corresponding to the current allocated resource for the current allocated resource, it means that the two executing agents have a resource conflict for the current allocated resource. Then determine that the resource conflict result between the two executing agents is in conflict, and this conflict situation needs to be considered and resolved when optimizing the control data later to improve the reliability of the control data.

[0069] In a specific embodiment, S33 includes the following steps: S331. Vectorize the first resource allocation data to obtain the allocation feature vector corresponding to the first resource allocation data.

[0070] S332. Obtain the correlation degree matrix according to the correlation degree between each allocated resource and each task objective type.

[0071] S333. Obtain the importance degree vector according to the candidate importance degree corresponding to each executing agent.

[0072] S334. Input the allocation feature vector, the correlation degree matrix, and the importance degree vector into the trained prediction model to obtain a predicted numerical sequence, where the predicted numerical sequence includes the predicted numerical value corresponding to each task objective type.

[0073] S335. For any task objective type, obtain the task completion degree of each task objective type according to the predicted numerical value corresponding to the current task objective type and the preset target numerical value corresponding to the current task objective type.

[0074] Among them, according to the vector conversion method, the first resource allocation data is vectorized to obtain the allocation feature vector corresponding to the first resource allocation data.

[0075] Take each allocated resource as the row keyword of the correlation degree matrix, take each task objective type as the column keyword of the correlation degree matrix, and use the correlation degree between each allocated resource and each task objective type as the value at the corresponding position in the correlation degree matrix to construct the correlation degree matrix.

[0076] The importance degree vector is a 1*K-dimensional vector, where K is the total number of executing agents. Correspondingly, the implementer pre-sets the position of each executing agent in the importance degree vector in advance, so as to use the candidate importance degree corresponding to each executing agent as the value at the corresponding position in the importance degree vector to construct the importance degree vector.

[0077] The trained prediction model can take the convolutional network model as an example, including a convolutional layer and a fully connected layer. Among them, the convolutional layer is used to extract features from the input allocation feature vector, correlation degree matrix, and importance degree vector, and the fully connected layer maps the extracted features to obtain the predicted numerical value corresponding to each task objective type and form a predicted numerical sequence to represent the task completion degree of each task objective type by the supply chain based on the first control data. The number of neurons in the fully connected layer is the same as the number of task objective types. It can be known that in order to ensure the generalization and accuracy of the prediction model, the implementer can train the prediction model. The training method of the prediction model belongs to the prior art and will not be elaborated here.

[0078] The preset target values include a preset efficiency value, a preset energy consumption value, and a preset production capacity value.

[0079] For example, for the target efficiency, calculate the first difference between the predicted value corresponding to the target efficiency and the preset efficiency value, and then calculate the first ratio of the first difference to the preset efficiency value. Then, the first ratio can represent the degree to which the production efficiency of the supply chain is greater than the preset efficiency value, that is, the degree of task completion corresponding to the target efficiency.

[0080] For the target energy consumption, calculate the second difference between the preset energy consumption value and the predicted value corresponding to the target energy consumption, and then calculate the second ratio of the second difference to the preset energy consumption value. Then, the second ratio can represent the degree to which the energy consumption of the supply chain is less than the preset energy consumption value, that is, the degree of task completion corresponding to the target energy consumption.

[0081] For the target production capacity, calculate the third difference between the predicted value corresponding to the target production capacity and the preset production capacity value, and then calculate the third ratio of the third difference to the preset production capacity value. Then, the third ratio can represent the degree to which the production capacity of the supply chain is greater than the preset production capacity value, that is, the degree of task completion corresponding to the target production capacity.

[0082] As described above, by combining the first resource allocation data, the correlation degree between each task target type, and the candidate importance degree corresponding to each executing agent, the task completion degree corresponding to each task target type is quantitatively characterized and sent to the decision-making agent as feedback information, thereby improving the balance of resource allocation among multiple task targets.

[0083] In a specific embodiment, S34 includes the following steps: S341, Determine two executing agents with a resource conflict result of conflict as a group of conflict agent combinations.

[0084] S342, According to all the conflict agent combinations, form a conflict agent combination set corresponding to the first resource allocation data.

[0085] S343, For any task target type, if the task completion degree of the current task target type is greater than the completion degree threshold corresponding to the current task target type, determine that the intermediate completion result of the current task target type is to complete the target.

[0086] S344, If the task completion degree of the current task target type is less than or equal to the completion degree threshold corresponding to the current task target type, determine that the intermediate completion result of the current task target type is not to complete the target.

[0087] Among them, the specific value of the completion degree threshold can be flexibly set by the implementer in combination with actual scenarios such as industry characteristics, enterprise strategic goals, and resource constraint conditions. For example, the completion degree threshold corresponding to the target efficiency is 0, the completion degree threshold corresponding to the target energy consumption is 10%, and the completion degree threshold corresponding to the target production capacity is 5%.

[0088] Above, by quantitatively measuring the task completion degrees of target efficiency, target energy consumption, and target production capacity based on the first allocation data, as well as the conflict situation of the allocated resources among the executing agents, a closed-loop feedback mechanism is provided for the control decision of the decision-making agent, thereby improving the balance of the supply chain production situation among multiple task goals based on the first control data.

[0089] S4. If the set of conflict agent combinations is a non-empty set or the target completion result is an uncompleted target, then the set of conflict agent combinations, the set of task completion degrees, and the supply chain data are jointly used as reference data and sent to the decision-making agent, and step S2 is repeatedly executed until the set of conflict agent combinations is an empty set and the target completion result is a completed target. The first control data at the time of terminating the repeated execution is determined as the second control data.

[0090] Among them, by jointly using the set of conflict agent combinations, the set of task completion degrees, and the supply chain data as reference data and sending them to the decision-making agent, the decision-making agent receives new reference data, and thus updates to obtain new first control data based on the more abundant reference data in terms of data dimension and content, so as to optimize the supply chain production situation, improve the completion degrees of target efficiency, target energy consumption, and target production capacity, and optimize the conflict situation of the allocated resources among the executing agents until the set of conflict agent combinations is an empty set, that is, there is no conflict situation of the allocated resources among the executing agents, and the target completion result is a completed target, that is, the supply chain production situation meets the preset requirements in terms of efficiency, energy consumption, and production capacity dimensions.

[0091] S5. Control the behaviors of each executing agent according to the second control data.

[0092] In a specific implementation manner, the second control data includes second production plan data and second resource allocation data. The second production plan data includes a production task list corresponding to each executing agent and the production quantity and production time period corresponding to each production task in the production task list. The second resource allocation data includes several target allocated resources corresponding to each executing agent and the target time period corresponding to each target allocated resource. S5 includes the following steps: S51. Allocate resources for each executing agent according to the second resource allocation data.

[0093] S52. Control the production tasks of each executing agent according to the second production plan data.

[0094] As described above, by sending business system data, device status data, and external environment data to the decision-making intelligent agent, the decision-making intelligent agent can comprehensively capture the real-time operation status of the supply chain, providing rich and accurate data support for the decision-making control of production planning and resource allocation. By introducing the analysis of conflict detection for resource allocation among execution intelligent agents and the task completion degree of each task objective type, the reference data received by the decision-making intelligent agent is fed back and updated, enabling the decision-making intelligent agent to iteratively optimize the first control data output based on the more abundant reference data in terms of data dimension and content. This not only eliminates the resource conflict situation among execution intelligent agents but also optimizes the completion degree of the supply chain production in terms of efficiency, energy consumption, and production capacity, enabling the second control data finally output by the decision-making intelligent agent to achieve the collaborative optimal invocation of multiple objectives, providing more efficient and reliable control data for the management of the supply chain, thereby improving the production efficiency and production capacity of the supply chain and reducing production energy consumption.

[0095] Embodiment 2 Embodiment 2 provides a multi-agent collaborative invocation system based on big data, as Figure 2 shown. The multi-agent collaborative invocation system based on big data includes: A data sending module 21, configured to send supply chain data as reference data to the decision-making intelligent agent, where the supply chain data includes business system data, device status data of each execution intelligent agent, and external environment data.

[0096] A first control data acquisition module 22, configured to obtain first control data from the decision-making intelligent agent based on the received reference data, where the first control data includes first production plan data and first resource allocation data, and the first resource allocation data includes a plurality of allocated resources corresponding to each execution intelligent agent and the usage time period corresponding to each allocated resource.

[0097] A control data analysis module 23, configured to obtain a conflict intelligent agent combination set, a task completion degree set, and a target completion result corresponding to the first resource allocation data according to the first resource allocation data, the relevance between each allocated resource and each task objective type, and the candidate importance degree corresponding to each execution intelligent agent, where the task objective types include target efficiency, target energy consumption, and target production capacity, the conflict intelligent agent combination set includes a plurality of conflict intelligent agent combinations composed of two execution intelligent agents with resource conflicts, the task completion degree set includes the task completion degree of each task objective type, and the target completion result is either completed or not completed.

[0098] The second control data acquisition module 24 is configured to, if the conflict agent combination set is a non-empty set or the target completion result is an uncompleted target, use the conflict agent combination set, the task completion degree set, and the supply chain data as reference data and send them to the decision-making agent, and repeatedly execute the first control data acquisition module 22 until the conflict agent combination set is an empty set and the target completion result is a completed target, and determine the first control data at the time of terminating the repeated execution as the second control data.

[0099] The execution agent control module 25 is configured to control the behavior of each execution agent according to the second control data.

[0100] In a specific embodiment, the data sending module 21 includes: The first data acquisition sub-module is configured to collect the business data of each supply link in the supply chain, the device status data of each execution agent, and the external environment data through the data agent. Among them, the business data includes order data, inventory data, production process data, and logistics status, the device status data includes the running status, the current task load, the production capacity utilization rate, and the device failure rate, and the external environment data includes weather data, traffic data, and demand data.

[0101] The second data acquisition sub-module is configured to form business system data according to the business data of all supply links in the supply chain.

[0102] In a specific embodiment, the first control data acquisition module 22 includes: The state vector mapping sub-module is configured to map the reference data into a state vector. Among them, the state vector includes a business system feature vector, a device status feature vector corresponding to each execution agent, an external environment feature vector, a conflict agent feature vector, and a task completion degree feature vector. When the decision-making agent does not receive the conflict agent combination set and the task completion degree set, the conflict agent feature vector and the task completion degree feature vector are zero vectors.

[0103] The value range division sub-module is configured to divide the value range corresponding to each allocated resource into corresponding numbers of value intervals according to the preset discrete level numbers corresponding to each allocated resource.

[0104] The action set acquisition sub-module is configured to obtain the discrete actions corresponding to each allocated resource for each value interval, and obtain the action set corresponding to each allocated resource, where the action set is obtained from all the corresponding discrete actions.

[0105] The discrete action combination sub-module is configured to obtain all the discrete action combinations according to the action sets corresponding to each allocated resource, where the total number of discrete action combinations is the sum of the preset discrete level numbers corresponding to all allocated resources.

[0106] The first control data acquisition sub-module is used to input the state vector and all discrete actions into the trained policy model corresponding to the decision-making agent, and obtain the first control data output by the decision-making agent.

[0107] In a specific embodiment, the control data analysis module 23 includes: The correlation degree acquisition sub-module is used to obtain the correlation degree between the resource type corresponding to each allocated resource and each task objective type, where the resource types include production and manufacturing resources, warehousing and logistics resources, and human resources.

[0108] The resource conflict result acquisition sub-module is used to obtain the resource conflict result between any two execution agents according to the first resource allocation data, where the resource conflict result is that a conflict occurs or no conflict occurs.

[0109] The task completion degree acquisition sub-module is used to obtain the task completion degree corresponding to each task objective type according to the first resource allocation data, the correlation degree between each allocated resource and each task objective type, and the candidate importance degree corresponding to each execution agent.

[0110] The control data analysis sub-module is used to obtain the conflict agent combination set, the task completion degree set, and the target completion result corresponding to the first resource allocation data according to the resource conflict result between any two execution agents and the task completion degree corresponding to each task objective type.

[0111] In a specific embodiment, the resource conflict result acquisition sub-module includes: The execution agent set determination unit is used to determine, according to the several allocated resources corresponding to each execution agent, the several execution agents corresponding to each allocated resource as the execution agent set corresponding to each allocated resource.

[0112] The first resource conflict result acquisition unit is used to, for any allocated resource, obtain the resource conflict result between any two execution agents in the execution agent set corresponding to the current allocated resource according to the usage time periods of the current allocated resource by any two execution agents in the execution agent set corresponding to the current allocated resource.

[0113] The second resource conflict result acquisition unit is used to traverse all the allocated resources and all the execution agents in the execution agent set corresponding to each allocated resource, and obtain the resource conflict result between any two execution agents.

[0114] In a specific embodiment, the task completion degree acquisition sub-module includes: An allocation feature vector acquisition unit for vectorizing the first resource allocation data to obtain an allocation feature vector corresponding to the first resource allocation data.

[0115] A correlation degree matrix acquisition unit for obtaining a correlation degree matrix according to the correlation degree between each allocated resource and each task objective type.

[0116] An importance degree vector acquisition unit for obtaining an importance degree vector according to the candidate importance degree corresponding to each executing agent.

[0117] A predicted value acquisition unit for inputting the allocation feature vector, the correlation degree matrix and the importance degree vector into a trained prediction model to obtain a predicted value sequence, where the predicted value sequence includes the predicted value corresponding to each task objective type.

[0118] A task completion degree acquisition unit for, for any task objective type, obtaining the task completion degree of each task objective type according to the predicted value corresponding to the current task objective type and the preset target value corresponding to the current task objective type.

[0119] In a specific embodiment, the control data analysis sub-module includes: A conflicting agent combination acquisition unit for determining two executing agents with a resource conflict result of having a conflict as a group of conflicting agent combinations.

[0120] A conflicting agent combination set acquisition unit for forming a set of conflicting agent combinations corresponding to the first resource allocation data according to all the conflicting agent combinations.

[0121] A first completion result acquisition unit for, for any task objective type, if the task completion degree of the current task objective type is greater than the completion degree threshold corresponding to the current task objective type, determining that the intermediate completion result of the current task objective type is to complete the target.

[0122] A second completion result acquisition unit for, if the task completion degree of the current task objective type is less than or equal to the completion degree threshold corresponding to the current task objective type, determining that the intermediate completion result of the current task objective type is not to complete the target.

[0123] In a specific embodiment, the second control data includes second production plan data and second resource allocation data. The second production plan data includes a production task list corresponding to each executing agent and the production quantity and production time period corresponding to each production task in the production task list. The second resource allocation data includes a number of target allocated resources corresponding to each executing agent and the target time period corresponding to each target allocated resource. The executing agent control module 25 includes: The first control sub-module is used to allocate resources for each executing agent according to the second resource allocation data.

[0124] The second control sub-module is used to control the production tasks of each executing agent according to the second production plan data.

[0125] It should be noted that the information interaction, execution process, etc. between the above modules, because they are based on the same concept as the method embodiment of the present invention, for their specific functions and the technical effects brought, reference can be specifically made to the method embodiment part, and details will not be repeated here.

[0126] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments by using the disclosed technical content within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A multi-agent collaborative invocation method based on big data, characterized in that, The method includes the following steps: S1. Send the supply chain data as reference data to the decision-making agent, where the supply chain data includes business system data, device status data of each execution agent, and external environment data; S2. The decision-making agent obtains first control data according to the received reference data, where the first control data includes first production plan data and first resource allocation data, and the first resource allocation data includes several allocated resources corresponding to each execution agent and the usage time period corresponding to each allocated resource; S3. According to the first resource allocation data, the correlation degree between each allocated resource and each task objective type, and the candidate importance degree corresponding to each execution agent, obtain the conflict agent combination set, task completion degree set, and target completion result corresponding to the first resource allocation data, where the task objective types include target efficiency, target energy consumption, and target production capacity, the conflict agent combination set includes several conflict agent combinations composed of two execution agents with resource conflicts, the task completion degree set includes the task completion degree of each task objective type, and the target completion result is target completed or target not completed; S4. If the conflict agent combination set is a non-empty set or the target completion result is target not completed, then send the conflict agent combination set, the task completion degree set, and the supply chain data together as reference data to the decision-making agent, and repeat step S2 until the conflict agent combination set is an empty set and the target completion result is target completed, and determine the first control data at the end of the repeated execution as the second control data; S5. Control the behavior of each execution agent according to the second control data.

2. The multi-agent collaborative invocation method based on big data according to claim 1, wherein S1 includes the following steps: S11. The data agent collects business data of each supply link in the supply chain, device status data of each execution agent, and external environment data, where the business data includes order data, inventory data, production process data, and logistics status, the device status data includes operating status, current task load, production capacity utilization rate, and device failure rate, and the external environment data includes weather data, traffic data, and demand data; S12. Compose the business system data according to the business data of all supply links in the supply chain.

3. The multi-agent collaborative invocation method based on big data according to claim 1, wherein S2 includes the following steps: S21. Map the reference data to a state vector, where the state vector includes a business system feature vector, a device status feature vector corresponding to each execution agent, an external environment feature vector, a conflict agent feature vector, and a task completion degree feature vector. When the decision-making agent has not received the conflict agent combination set and the task completion degree set, the conflict agent feature vector and the task completion degree feature vector are zero vectors; S22. Divide the value range corresponding to each allocated resource into corresponding numbers of value intervals according to the preset discrete level number corresponding to each allocated resource; S23. Obtain the discrete actions corresponding to each value range for each allocated resource, and obtain the action set corresponding to each allocated resource, where the action set is obtained from all the corresponding discrete actions; S24. According to the action sets corresponding to each allocated resource, obtain all the discrete action combinations, where the total number of discrete action combinations is the sum of the preset discrete level numbers corresponding to all the allocated resources; S25. Input the state vector and all the discrete action combinations into the trained policy model corresponding to the decision-making agent, and obtain the first control data output by the decision-making agent.

4. The multi-agent collaborative invocation method based on big data according to claim 1, characterized in that S3 includes the following steps: S31. Obtain the correlation degree between the resource type corresponding to each allocated resource and each task objective type, where the resource types include production and manufacturing resources, warehousing and logistics resources, and human resources; S32. According to the first resource allocation data, obtain the resource conflict results between any two execution agents, where the resource conflict results are either in conflict or not in conflict; S33. According to the first resource allocation data, the correlation degree between each allocated resource and each task objective type, and the candidate importance degree corresponding to each execution agent, obtain the task completion degree corresponding to each task objective type; S34. According to the resource conflict results between any two execution agents and the task completion degree corresponding to each task objective type, obtain the conflict agent combination set, the task completion degree set, and the target completion result corresponding to the first resource allocation data.

5. The multi-agent collaborative invocation method based on big data according to claim 4, wherein, S32 includes the following steps: S321. According to the several allocated resources corresponding to each execution agent, determine the execution agent set corresponding to each allocated resource for each of the several execution agents corresponding to each allocated resource; S322. For any one allocated resource, according to the usage time periods of any two execution agents in the execution agent set corresponding to the current allocated resource for the current allocated resource, obtain the resource conflict results between any two execution agents in the execution agent set corresponding to the current allocated resource; S323. Traverse all the allocated resources and all the execution agents in the execution agent set corresponding to each allocated resource, and obtain the resource conflict results between any two execution agents.

6. The multi-agent collaborative invocation method based on big data according to claim 5, wherein, S33 includes the following steps: S331. Vectorize the first resource allocation data to obtain the allocation feature vector corresponding to the first resource allocation data; S332. Obtain the correlation degree matrix according to the correlation degree between each allocated resource and each task objective type; S333. Obtain the importance degree vector according to the candidate importance degree corresponding to each execution agent; S334. Input the allocation feature vector, the correlation degree matrix, and the importance degree vector into the trained prediction model to obtain a prediction numerical sequence, where the prediction numerical sequence includes the prediction numerical values corresponding to each task objective type; S335. For any task target type, obtain the task completion degree of each task target type according to the predicted value corresponding to the current task target type and the preset target value corresponding to the current task target type.

7. The multi-agent collaborative invocation method based on big data according to claim 6, characterized in that S34 includes the following steps: S341. Determine two execution agents with a resource conflict result of "conflict" as a group of conflicting agent combinations. S342. According to all the conflicting agent combinations, form a set of conflicting agent combinations corresponding to the first resource allocation data. S343. For any task target type, if the task completion degree of the current task target type is greater than the completion degree threshold corresponding to the current task target type, determine that the intermediate completion result of the current task target type is "complete target". S344. If the task completion degree of the current task target type is less than or equal to the completion degree threshold corresponding to the current task target type, determine that the intermediate completion result of the current task target type is "incomplete target".

8. The multi-agent collaborative invocation method based on big data according to claim 1, wherein The second control data includes second production plan data and second resource allocation data. The second production plan data includes a production task list corresponding to each execution agent and the production volume and production time period corresponding to each production task in the production task list. The second resource allocation data includes several target allocation resources corresponding to each execution agent and the target time period corresponding to each target allocation resource. S5 includes the following steps: S51. Allocate resources for each execution agent according to the second resource allocation data. S52. Control the production tasks of each execution agent according to the second production plan data.

9. A multi-agent collaborative call system based on big data, characterized in that, The system includes: A data sending module, configured to send supply chain data as reference data to a decision-making agent, where the supply chain data includes business system data, device status data of each execution agent, and external environment data. A first control data acquisition module, configured to obtain first control data from the decision-making agent according to the received reference data, where the first control data includes first production plan data and first resource allocation data, and the first resource allocation data includes several allocation resources corresponding to each execution agent and the usage time period corresponding to each allocation resource. A control data analysis module, configured to obtain a set of conflicting agent combinations, a set of task completion degrees, and a target completion result corresponding to the first resource allocation data according to the first resource allocation data, the relevance between each allocation resource and each task target type, and the candidate importance degree corresponding to each execution agent, where the task target types include target efficiency, target energy consumption, and target production capacity, the set of conflicting agent combinations includes several conflicting agent combinations composed of two execution agents with resource conflicts, the set of task completion degrees includes the task completion degree of each task target type, and the target completion result is "complete target" or "incomplete target". The second control data acquisition module is configured to, if the conflict agent combination set is a non-empty set or the target completion result is an uncompleted target, use the conflict agent combination set, the task completion degree set, and the supply chain data as reference data and send them to the decision-making agent, and repeatedly execute the first control data acquisition module until the conflict agent combination set is an empty set and the target completion result is a completed target, and determine the first control data at the time of terminating the repeated execution as the second control data; The execution agent control module is configured to control the behavior of each execution agent according to the second control data.

10. The multi-agent collaborative call system based on big data according to claim 9, wherein, The data sending module includes: The first data acquisition sub-module is configured to collect the business data of each supply link in the supply chain, the device status data of each execution agent, and the external environment data through a data agent. Among them, the business data includes order data, inventory data, production process data, and logistics status, the device status data includes operating status, current task load, production capacity utilization rate, and device failure rate, and the external environment data includes weather data, traffic data, and demand data; The second data acquisition sub-module is configured to form business system data according to the business data of all supply links in the supply chain.

Images