Group agent-based trusted man-machine hybrid intelligent decision-making method and system

Abstract

The invention provides a credible man-machine hybrid intelligent decision-making method and system based on a group agent, and relates to the technical field of information, the method is executed based on a hierarchical cascade architecture, and the method comprises the steps that a first-level decision-making agent generates a decision-making scheme and issues an execution plan, and the process comprises first quantitative evaluation of plan suitability and resource credibility controllability; receiving an enabling application initiated by a secondary decision-making agent due to the fact that the secondary decision-making agent cannot independently process the emergency event, and executing second quantitative evaluation including the influence on the event processing and plan execution effect; in response to the application and authorized by a commander, dynamically configuring a temporary command permission and / or supplementing a resource interface for the applicant, and updating a role identifier of the applicant after the permission is configured; after the event is processed, the allocated authority and resources are recovered, and the original role is recovered; and monitoring an evaluation result, and when the evaluation result is lower than a corresponding threshold value, interrupting the process and requesting a commander to intervene. According to the method, cross-level collaboration, decision permission dynamic safety adjustment and whole-process credibility and controllability are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a trusted human-machine hybrid intelligent decision-making method and system based on group agents. BACKGROUND

[0002] With the deep integration of all things intelligent interconnection and advanced model technology, the dependence of key fields such as emergency response and command decision on group agent collaborative systems is increasing.

[0003] At present, there are many researches and applications on multi-agent cooperation and task allocation. Specifically, the existing multi-agent cooperation scheme, in the architecture level, the architecture design of the existing scheme has diversity in cross-level linkage and resource integration, which leads to low overall collaborative efficiency of the system, and there is a data island phenomenon; in the dynamic response level, when responding to emergencies, the dynamic adjustment mechanism of decision-making authority is not perfect, there are authority handover loopholes and resource mismatch risks; in the trusted security level, for the trusted controllability of the human-machine hybrid decision-making process, there is still a lack of systematic quantitative evaluation framework, which makes it difficult to guarantee the controllability and result reliability of the whole decision-making process.

[0004] Therefore, there is an urgent need for a group agent intelligent decision-making scheme that can realize cross-level efficient collaboration, support dynamic and safe adjustment of decision-making authority, and ensure the whole process of trusted controllability. SUMMARY

[0005] The present application provides a trusted human-machine hybrid intelligent decision-making method and system based on group agents, to solve the defects of the existing group agent system, such as low cross-level cooperation efficiency, rigid decision-making authority adjustment, lack of security guarantee, and insufficient decision-making process and result reliability.

[0006] This invention provides a trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents. The method first executes based on a hierarchical cascaded architecture comprising primary and secondary decision-making agents. The primary decision-making agent generates a decision scheme and distributes the execution plan derived from the decomposition of the decision scheme to the secondary decision-making agents. The process of generating the decision scheme includes a first quantitative assessment, which at least includes a predictive assessment of the suitability of the plan and a matching assessment of the credibility and controllability of the resources involved in the decision scheme. The method receives an empowerment request initiated by the secondary decision-making agent during the execution of the execution plan after it autonomously determines that the agent cannot handle the emergency independently, and performs a second quantitative assessment, which at least includes an execution effect assessment of the impact of the emergency on the handling of the emergency and its impact on the execution effect of the plan. In response to the empowerment request and with authorization from the commander, the method dynamically configures temporary command and decision-making permissions and / or supplementary resource and capability interfaces for the secondary decision-making agent that submitted the empowerment request, and updates its role identifier after configuring the temporary command and decision-making permissions. After the emergency is handled, the temporary command and decision-making permissions and supplementary resource and capability interfaces dynamically configured for the secondary decision-making agent are revoked, and its role identification is restored to its original state; the evaluation results of the first or second quantitative assessment are monitored, and when the evaluation result is lower than the corresponding threshold, the current process is interrupted or suspended, and the commander is requested to intervene.

[0007] This invention provides a trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents. The first-level decision-making intelligence agents include a planning intelligence agent, a preparation intelligence agent, an execution monitoring intelligence agent, and an evaluation intelligence agent. First-level decision-making intelligence agents at the same level support sequential collaboration in a ring-like cascade, while first-level decision-making intelligence agents at different levels support cross-level vertical linkage collaboration. Second-level decision-making intelligence agents are used to: execute the execution plan issued by the first-level decision-making intelligence agents; or, when detecting a sudden event that cannot be handled independently, initiate an empowerment request to obtain temporary command and decision-making authority and / or supplementary resource and capability interfaces required for parallel collaborative execution with other second-level decision-making intelligence agents at the same level; or, after obtaining temporary command and decision-making authority and updating their role identifier to that of a temporary first-level decision-making intelligence agent, organize other intelligence agents to collaboratively handle the sudden event.

[0008] This invention provides a trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents. The evaluation agent has a first working mode and a second working mode. The first working mode is an evaluation sub-module, which is embedded in the planning agent, the preparation agent, and the execution monitoring agent, respectively. The second working mode is an independent evaluation agent.

[0009] The application provides a trusted man-machine hybrid intelligent decision-making method based on swarm agents, wherein in a first working mode, an evaluation submodule embedded in a planning agent is used for performing a prediction evaluation of the adaptability of a preplan in a first quantitative evaluation; an evaluation submodule embedded in a preparation agent is used for performing a matching evaluation of the trustworthiness, controllability and task execution controllability of resources involved in a decision scheme in the first quantitative evaluation; and an evaluation submodule embedded in an execution monitoring agent is used for performing an evaluation of the influence of a sudden event on disposal and the influence of the sudden event on the execution effect of a plan in a second quantitative evaluation.

[0010] The application provides a trusted man-machine hybrid intelligent decision-making method based on swarm agents, wherein the planning agent comprises a superior planning agent and an inferior planning agent. The generation of the decision scheme comprises the following steps: the superior planning agent receives a task and issues a pre-order to the inferior planning agent in combination with the intention of a commander; and the superior planning agent and the inferior planning agent generate multiple preplans in parallel, simulate and deduce each preplan, and determine a decision scheme according to the order of the commander.

[0011] The application provides a trusted man-machine hybrid intelligent decision-making method based on swarm agents, wherein the preparation agent comprises a superior preparation agent and an inferior preparation agent. The generation of the execution plan comprises the following steps: the superior preparation agent groups and evaluates resources according to the decision scheme, generates an execution plan of the current level and issues the execution plan; and the inferior preparation agent receives the execution plan of the current level, generates an adapted branch plan and issues the branch plan to a corresponding secondary decision agent.

[0012] The application provides a trusted man-machine hybrid intelligent decision-making method based on swarm agents, and the dynamic configuration of a secondary decision agent for a power application comprises the following steps: the execution monitoring agent evaluates the power application and generates an authorization proposal; after receiving a confirmation instruction of the commander on the authorization proposal, the secondary decision agent for the power application is dynamically configured with temporary command decision-making authority and / or an additional resource and capability interface; and after the temporary command decision-making authority is configured, the role identifier is updated.

[0013] The application provides a trusted man-machine hybrid intelligent decision-making method based on swarm agents, and the method further comprises the following steps: based on an inter-agent communication protocol with added information fields of trust values and controllable values, the cooperation state of the agents is monitored and interactively verified; based on a resource calling protocol with added identity authentication and calling audit mechanisms, the trustworthiness of external resources is verified; wherein the external resources comprise external models, external databases and external tools.

[0014] According to the present application, a trusted man-machine hybrid intelligent decision-making system based on swarm agents is provided. The system comprises: a primary decision-making agent configured to perform the operations performed by the primary decision-making agent in any of the preceding methods; a secondary decision-making agent configured to perform the operations performed by the secondary decision-making agent in any of the preceding methods; a trusted and controllable module deployed between each agent, configured to monitor and interactively verify the cooperation state of the agents based on an inter-agent communication protocol with added trusted value and controllable value information fields; and a resource management module deployed between the agents and external resources, configured to verify the trustworthiness of the external resources based on a resource calling protocol with added identity authentication and calling audit mechanisms.

[0015] According to the present application, a trusted man-machine hybrid intelligent decision-making system based on swarm agents is provided. The system comprises: a primary decision-making agent configured to perform the operations performed by the primary decision-making agent in any of the preceding methods; a secondary decision-making agent configured to perform the operations performed by the secondary decision-making agent in any of the preceding methods; a trusted and controllable module deployed between each agent, configured to monitor and interactively verify the cooperation state of the agents based on an inter-agent communication protocol with added trusted value and controllable value information fields; and a resource management module deployed between the agents and external resources, configured to verify the trustworthiness of the external resources based on a resource calling protocol with added identity authentication and calling audit mechanisms.

[0016] The present application also provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the trusted man-machine hybrid intelligent decision-making method based on swarm agents according to any of the preceding methods when executing the computer program.

[0017] The present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the trusted man-machine hybrid intelligent decision-making method based on swarm agents according to any of the preceding methods.

[0018] The present application also provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the trusted man-machine hybrid intelligent decision-making method based on swarm agents according to any of the preceding methods.

[0019] The application provides a trusted man-machine hybrid intelligent decision-making method and system based on a group agent, which realizes full-link enhancement from static planning to dynamic execution through the combination of hierarchical architecture and quantitative evaluation; the first quantitative evaluation is simultaneously performed in the scheme generation stage, which can exclude schemes with poor plan adaptability or unreliable resources in advance, thereby guaranteeing the reliability of the decision-making basis from the source; when a sudden event occurs during execution, the second quantitative evaluation is triggered, which can quantitatively evaluate the event in real time and provide accurate and objective decision-making basis for subsequent dynamic adjustment of the authority; on this basis, the system is supported by the evaluation results, and temporary authority and interface configuration of the execution body and post-recovery are completed by personnel authorization, so that the authority adjustment is no longer a rigid preset, but becomes a data-driven and safely controlled flexible response capability; finally, the continuous monitoring of the two evaluation results and the setting of the intervention threshold constitute a double-trust insurance, which ensures that the credibility decay of any link can be captured in time and handed over to personnel for takeover, thereby realizing efficient cooperation across levels, dynamic safety adjustment of decision-making authority and trusted and controllable decision-making process. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0021] Figure 1 is a structural schematic diagram of a trusted man-machine hybrid intelligent decision-making system based on a group agent provided by the application.

[0022] Figure 2 is a flowchart of a trusted man-machine hybrid intelligent decision-making method based on a group agent provided by the application.

[0023] Figure 3 is a complete flowchart of a trusted man-machine hybrid intelligent decision-making method based on a group agent provided by the application.

[0024] Figure 4 is a schematic diagram of the physical structure of an electronic device provided by the application. DETAILED DESCRIPTION

[0025] In order to make the objects, technical solutions and advantages of the application clearer, the technical solutions in the application will be described clearly and completely below with reference to the drawings in the application. Obviously, the described embodiments are some embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0026] In a common group agent decision system architecture, multiple agents are usually deployed on computing nodes with a hierarchical relationship, and exchange information and cooperate with each other through a specific communication protocol, such as Agent to Agent Protocol (A2A), Autonomous Network Protocol (ANP), etc. The commander accesses the system through a dedicated command and control terminal, issues tasks and monitors the execution.

[0027] When dealing with complex tasks, although multiple agents are logically subordinate to the same task system, they are usually operated as independent modules following a fixed process in application logic. Their decision-making process is either independently calculated based on their own model or requires task scheduling and result aggregation through a unified control node.

[0028] Under this architecture, task execution and permission allocation usually follow static, predefined rules. When an execution-type agent encounters an unexpected event outside its permission or capability range during task execution, the existing collaboration process cannot support it to dynamically obtain higher decision-making authority and additional resources. The commander or superior system must manually analyze and operate to re-decompose tasks, grant permissions, and allocate resources, resulting in low efficiency.

[0029] In addition, some existing technologies involve multi-agent collaboration, and the design focus is mainly on task allocation optimization or communication efficiency improvement. They do not quantitatively evaluate the credibility and controllability of the agent's own state, interaction behavior, and external resource calls in the execution process, and deeply couple them with the permission management mechanism, resulting in a lack of active and adaptive security intervention capabilities when the system faces abnormal or untrusted situations.

[0030] At the same time, if the complex collaborative decision-making and trust evaluation logic are all concentrated in the top layer or the cloud, all perception data and decision instructions need to go through a long transmission and processing link, which is difficult to meet the stringent requirements of real-time, autonomy, and reliability in emergency response scenarios.

[0031] To solve the above technical problems, the present application provides a trusted man-machine hybrid intelligent decision-making method and system based on group agents, which builds a group agent collaboration system with clear responsibilities and ring-like cascading within a hierarchical architecture, and based on quantitative trust evaluation and threshold intervention mechanisms throughout the task generation, execution, and dynamic permission adjustment process, realizes cross-level collaboration, dynamic and safe adjustment of decision-making authority, and whole-process trust and controllability.

[0032] The application provides a trusted man-machine hybrid intelligent decision-making system based on swarm agents, which comprises a first-level decision-making agent, a second-level decision-making agent, a trusted controllable module and a resource management module.

[0033] In the embodiment of the application, the first-level decision-making agent is configured to perform the operations performed by the first-level decision-making agent in the method embodiments below; and the second-level decision-making agent is configured to perform the operations performed by the second-level decision-making agent in the method embodiments below.

[0034] Exemplarily, the first-level decision-making agent can generate a decision scheme and issue an execution plan; receive and process an empowerment application of the second-level decision-making agent; dynamically configure permissions and interfaces for the second-level decision-making agent and update the role identification thereof after authorization; recover the permissions and interfaces and restore the role of the second-level decision-making agent after the task is completed; monitor the quantitative evaluation result and request the intervention of the commander when the result is below a threshold.

[0035] In the embodiment of the application, the trusted controllable module is deployed between each agent. It is configured to monitor and interactively verify the cooperation state of the agents based on an inter-agent communication protocol with added trusted value and controllable value information fields.

[0036] Exemplarily, the trusted controllable module adds quantitative fields of trusted values and controllable values to A2A, ANP, Artificial Society Physical Information System Protocol (ACP) and other basic protocols, monitors the cooperation state and task execution progress of the agents in real time, and ensures the trusted verification and controllable takeover of the task results of the agents by the command team.

[0037] In the embodiment of the application, the resource management module is deployed between the agents and external resources, and is configured to verify the trustworthiness of the external resources based on a resource calling protocol with added identity authentication and calling audit mechanisms. The external resources include external models, external databases and external tools.

[0038] Exemplarily, the resource management module extends the trusted verification rules of the resource registration, calling and execution stages based on the Model Call Protocol (MCP), records and traces the access qualifications, calling process and execution result of the external resources throughout the whole process.

[0039] In this way, the application provides a trusted basis for the dynamic adjustment of decision-making permissions by deeply integrating the trusted controllable mechanism and the dynamic calling mechanism into the system architecture, constructing a security barrier from two dimensions of agent cooperation and external resource calling.

[0040] In the embodiment of the present application, the first-level decision-making agent comprises a planning agent, a preparation agent, an execution monitoring agent and an evaluation agent.

[0041] The evaluation agent has a first working mode and a second working mode.

[0042] Exemplarily, the first working mode is an evaluation sub-module embedded in the planning agent, the preparation agent and the execution monitoring agent respectively; and the second working mode is an independent evaluation agent.

[0043] In the embodiment of the present application, the sequential cooperation of the ring cascade is supported between the first-level decision-making agents of the same level, and the up-down linkage cooperation is supported between the first-level decision-making agents of different levels.

[0044] Exemplarily, the sequential cooperation of the ring cascade is embodied as follows: in the same level (such as a certain command level), the planning agent, the preparation agent, the execution monitoring agent and the evaluation agent can sequentially transmit and process information according to the task flow, and jointly generate and iteratively optimize the emergency support scheme; and the up-down linkage cooperation is embodied as follows: the execution monitoring agents of the superior and the subordinate can synchronously track the task state and directly exchange control instructions.

[0045] In the embodiment of the present application, the second-level decision-making agent can perform the following three operations in different scenarios.

[0046] 1) The second-level decision-making agent is used to execute the execution plan issued by the first-level decision-making agent.

[0047] Exemplarily, the second-level decision-making agent (such as a transportation unit agent) receives and analyzes the plan of transporting Y materials along the X route, automatically controls the vehicle to complete the driving, loading, unloading and other actions, and reports the position and state in real time.

[0048] 2) The second-level decision-making agent is used to initiate an empowerment application to obtain temporary command decision-making authority and / or additional resources and capability interfaces required for parallel cooperative execution with other second-level decision-making agents of the same level when a sudden event that cannot be independently handled is detected.

[0049] Exemplarily, when the transportation unit agent detects a road interruption, it automatically initiates an application to request temporary authority to command other standby transportation units on the same route and calls a path planning tool.

[0050] 3) The second-level decision-making agent is used to organize other agents to cooperatively dispose the sudden event after obtaining the temporary command decision-making authority and updating the role identifier to a temporary first-level decision-making agent.

[0051] Exemplarily, after obtaining the authorization, the intelligent agent role identifier is updated, and the intelligent agent can dispatch detour tasks to other transportation unit intelligent agents in the surrounding area and coordinate loading equipment to jointly solve the traffic obstacles.

[0052] Thus, the application realizes the flatization and self-organizing collaborative response to the sudden situation by defining the role flexibility conversion of the secondary decision intelligent agent from execution to temporary command.

[0053] The following will be combined Figure 1 The above-mentioned trusted man-machine hybrid intelligent decision system based on group intelligent agent is described in detail.

[0054] As Figure 1 The trusted man-machine hybrid intelligent decision system based on group intelligent agent includes a superior primary decision intelligent agent 110, a plurality of subordinate primary decision intelligent agents 120 and a plurality of secondary decision intelligent agents 130.

[0055] The superior primary decision intelligent agent 110 is connected with the plurality of subordinate primary decision intelligent agents 120 respectively, and each subordinate primary decision intelligent agent 120 is connected with the plurality of secondary decision intelligent agents 130 respectively.

[0056] In the embodiment of the application, the superior primary decision intelligent agent 110 includes a superior planning intelligent agent 111, a superior preparation intelligent agent 112, a superior execution monitoring intelligent agent 113 and a superior evaluation intelligent agent 114.

[0057] Exemplarily, the superior planning intelligent agent 111 is used to receive the task documents issued by the commander, confirm the accuracy of the intention understanding and the situation judgment of the corresponding scene (such as emergency support) with the commander, issue pre-orders to the subordinate planning intelligent agent, and generate multiple sets of preplans through rule matching, and check the feasibility of the preplans through simulation deduction means.

[0058] Specifically, when generating the preplan, the superior planning intelligent agent 111 can call the evaluation sub-module (i.e. the first working mode of the evaluation intelligent agent) embedded therein to predict the adaptability of different preplans, assist in generating multiple sets of alternative preplans that meet the task target and time window requirements, and finally submit the checked preplan to the commander for selection, and generate a decision scheme framework with clear task target, task time window and core execution requirements based on the selected preplan.

[0059] Thus, the application ensures the pertinence and feasibility of the decision scheme through the intention confirmation-preplan generation-deduction verification process of the superior planning intelligent agent combined with the adaptability evaluation of the embedded evaluation sub-module.

[0060] In the embodiment of the present application, the upper preparation intelligent agent 112 is used for grouping resources such as transport units, support personnel and path planning models according to the selected plan of the commander, establishing a clear command and organization relationship.

[0061] Illustratively, the upper preparation intelligent agent 112 can call the evaluation submodule embedded therein to evaluate the credibility, controllability and matching degree of resources such as transport units and support personnel with the current task, optimize and adapt the resources based on the evaluation results, assist in generating the detailed plan of the current level, carry out preliminary action preparation (such as resource scheduling, vehicle inspection, etc.) at the same time, and synchronously issue the plan of the current level to the lower preparation intelligent agent.

[0062] In this way, the present application provides resource guarantee for the landing execution of the decision scheme through the resource grouping and optimization mechanism of the preparation intelligent agent combined with resource credibility evaluation.

[0063] In the embodiment of the present application, the upper execution monitoring intelligent agent 113 is used for monitoring the running state of each action group and support resource in real time according to the plan generated by the upper preparation intelligent agent.

[0064] Illustratively, the upper execution monitoring intelligent agent 113 can continuously determine whether there is an abnormal situation such as resource conflict or self-disturbance and mutual disturbance, and actively identify external emergencies (such as road emergency regulation and weather mutation in emergency support), evaluate and give disposal suggestions based on the action monitoring data and external event information.

[0065] Further, for the empowerment application submitted by the lower level, the upper execution monitoring intelligent agent 113 can evaluate the influence degree of the emergency on the expected action scheme effect and the resource allocation satisfaction, judge whether to respond to the application, if responding, synchronously solve the resource conflict related to the empowerment application, and complete the authorized empowerment after being authorized by the command team.

[0066] Still further, the upper execution monitoring intelligent agent 113 can also judge whether the task completion condition reported by the lower level meets the requirements of the action scheme, and give task correction suggestions in time.

[0067] In this way, the present application realizes the closed-loop control of the task execution process through the whole-process monitoring and dynamic response mechanism of the execution monitoring intelligent agent, and provides support for the rapid disposal of emergencies.

[0068] In the embodiment of the present application, the upper evaluation intelligent agent 114 is used for bearing the dual responsibilities of evaluating the event / task execution effect and the scheme completion effect.

[0069] An example, when the above level evaluation agent 114 first working mode (i.e. evaluation sub-module) exists, it is embedded in the superior planning agent 111, which is an evaluation sub-module 111a, and assists in completing the prediction evaluation of the plan adaptability; it is embedded in the superior preparation agent 112, which is an evaluation sub-module 112a, and assists in completing the resource reliability and matching degree evaluation; it is embedded in the superior execution monitoring agent 113, which is an evaluation sub-module 113a, and assists in completing the event / task execution effect evaluation, and outputs the plan adjustment suggestion according to the evaluation result of the event / task execution effect.

[0070] Another example, when the above level evaluation agent 114 second working mode (i.e. independent agent) exists, it collects and summarizes the data of the whole action process, evaluates whether the scheme completion effect meets the preset requirements, generates a task summary report if it meets the requirements, outputs a scheme adjustment suggestion if it does not meet the requirements, and feeds back to the superior planning agent 111 or the superior execution monitoring agent 113.

[0071] In this way, the present application realizes the whole cycle coverage of process evaluation-result evaluation through the dual-mode design of the evaluation agent, and provides data support for decision optimization and process adjustment.

[0072] In the embodiment of the present application, the subordinate first-level decision-making agent 120 includes a subordinate planning agent 121, a subordinate preparation agent 122, a subordinate execution monitoring agent 123 and a subordinate evaluation agent 124.

[0073] Exemplarily, the subordinate planning agent 121 is used to input the pre-order of the superior planning agent 111 as a task, can understand the intention of the superior in combination with the actual situation of the jurisdiction range (such as the regional road network condition, the resource warehouse location, etc.), carry out the planning work in parallel, formulate the plan of the subordinate and the branch plan, and report to the superior planning agent 111, receive the conflict detection result and the deviation correction prompt of the superior, and adjust and optimize the plan of the subordinate and the branch plan.

[0074] In this way, the present application realizes the hierarchical decomposition and local adaptation of the decision scheme through the linkage and cooperation of the superior and subordinate planning agents, and improves the cross-level cooperation efficiency.

[0075] Exemplarily, the subordinate preparation agent 122 is used to receive the plan of the subordinate issued by the superior preparation agent 112, input it as a task, reorganize the resources such as the transportation unit, the support personnel and the model under the jurisdiction of the subordinate, establish the organization relationship of the adaptive branch plan, call the evaluation sub-module embedded therein to evaluate the matching degree of the resources and the branch plan of the subordinate, select the adaptive resources to assist in generating the branch plan of the subordinate, and carry out the pre-action preparation, and at the same time, issue the branch plan to the corresponding second-level decision-making agent 130.

[0076] Thus, the application ensures the operability of the decision scheme at the execution level through the local resource adaptation and plan decomposition of the lower-level preparation agent.

[0077] Illustratively, the lower-level execution monitoring agent 123 is used to monitor the execution state of the branch plan at the current level, and to aggregate and analyze the execution data reported by the secondary decision agent 130.

[0078] Specifically, the lower-level execution monitoring agent 123 can monitor the real-time state of the action formation and support resources at the current level, identify resource conflicts, self-disturbance, mutual disturbance and external sudden situations, and give disposal suggestions; for the empowerment application initiated by the secondary decision agent 130, the lower-level execution monitoring agent 123 can first perform preliminary audit to evaluate the rationality and urgency of the application, and then report to the upper-level execution monitoring agent 113; at the same time, according to the task completion situation of the secondary decision agent 130, it is judged whether it meets the branch plan requirements, and timely gives rectification suggestions, if there is a problem beyond the disposal capacity of the current level, it is reported to the upper-level execution monitoring agent 113.

[0079] Thus, the application realizes the nearby, hierarchical feedback and rapid response of the task execution state through the direct monitoring and management of the lower-level execution monitoring agent on the secondary decision agent under its jurisdiction, reducing the delay caused by multi-level communication.

[0080] In the embodiment of the application, the working mode of the lower-level evaluation agent 124 is consistent with that of the upper-level evaluation agent 114, including the embedded sub-module mode and the independent agent mode.

[0081] In one example, when embedded as a sub-module, it is embedded in the lower-level planning agent 121 as an evaluation sub-module 121a, in the lower-level preparation agent 122 as an evaluation sub-module 122a, and in the lower-level execution monitoring agent 123 as an evaluation sub-module 123a, to evaluate the execution effect of each link at the current level in real time.

[0082] In another example, when acting as an independent agent, the lower-level evaluation agent 124 collects and aggregates the execution process data of the branch plan at the current level, evaluates the completion effect of the branch plan, generates a current-level evaluation report and reports to the upper-level evaluation agent 114, to provide basic data for global scheme evaluation.

[0083] Thus, the application realizes hierarchical aggregation and global analysis of evaluation data through the cooperative evaluation of upper and lower-level evaluation agents, ensuring the comprehensiveness of the evaluation results.

[0084] In the embodiment of the application, the secondary decision agent 130 includes a plurality of agents (such as agents 1 to N).

[0085] Each of the intelligent agents comprises a perception sub-module 131, a judgment sub-module 132, a decision sub-module 133, and an execution sub-module 134.

[0086] Exemplarily, each of the intelligent agents is configured to receive a branch plan issued by the lower-level decision-making intelligent agent 120, analyze the plan and automatically generate instruction codes, execute the instructions, and evaluate the execution effect.

[0087] Specifically, in the process of executing the instructions, the second-level decision-making intelligent agent 130 can identify an emergency event through the perception sub-module 131, determine whether it can independently handle the emergency event through the judgment sub-module 132, and if not, initiate an empowerment application to the upper-level execution monitoring intelligent agent (the lower-level execution monitoring intelligent agent 123 or the upper-level execution monitoring intelligent agent 113) through the decision sub-module 133. After obtaining the authorization and / or empowerment, the second-level decision-making intelligent agent 130 can be dynamically configured with one or more of the following supports: temporary command decision-making authority, additional resources, and capability interfaces.

[0088] For example, if the second-level decision-making intelligent agent 130 obtains temporary command decision-making authority, the role identifier of the second-level decision-making intelligent agent 130 will be updated to a temporary first-level decision-making intelligent agent, which can organize other intelligent agents to cooperatively handle the task. After the task is completed, the authority is released and the second-level decision-making intelligent agent 130 returns to the role of the second-level decision-making intelligent agent.

[0089] For another example, if the second-level decision-making intelligent agent 130 only obtains additional resources and capability interfaces, the second-level decision-making intelligent agent 130 remains in the role of the second-level decision-making intelligent agent and independently completes the task by using the additional resources. After the task is completed, all resources interfaces dynamically configured for the second-level decision-making intelligent agent 130 are recovered.

[0090] In this way, the present application realizes rapid response and efficient handling of emergency events by dynamically jumping the role of the second-level decision-making intelligent agent and flexibly adjusting the authority, thereby improving the dynamic adaptability of the system.

[0091] The present application will be described below in conjunction with Figure 2 The present application provides a trusted man-machine hybrid intelligent decision-making method based on group intelligent agents, which is executed based on the hierarchical cascading architecture comprising first-level decision-making intelligent agents and second-level decision-making intelligent agents as shown in Figure 1

[0092] Figure 2 The present application provides a trusted man-machine hybrid intelligent decision-making method based on group intelligent agents, which is executed based on the hierarchical cascading architecture comprising first-level decision-making intelligent agents and second-level decision-making intelligent agents as shown in Figure 2 S201, generating a decision scheme by the first-level decision-making intelligent agent, and issuing an execution plan obtained by decomposing the decision scheme to the second-level decision-making intelligent agent.

[0093] ​​The first-level decision-making agent comprises a planning agent, a preparation agent, an execution monitoring agent, and an evaluation agent. The first-level decision-making agents of the same level support ring-type cascaded sequential cooperation, and the first-level decision-making agents of different levels support cross-level up-down linkage cooperation.

[0094] In the embodiment of the application, the evaluation agent has a first working mode and a second working mode.

[0095] For example, the first working mode is an evaluation sub-module embedded in the planning agent, the preparation agent, and the execution monitoring agent, respectively; and the second working mode is an independent evaluation agent.

[0096] It should be noted that the above description of the first-level decision-making agent and the second-level decision-making agent, and the components included therein can refer to the corresponding detailed description in the above Figure 1 For brevity, the above description will not be repeated here.

[0097] The second-level decision-making agent is configured to: execute an execution plan issued by the first-level decision-making agent; or initiate an empowerment application when a sudden event that cannot be independently handled is detected; or after obtaining temporary command decision-making authority and updating the role identifier to a temporary first-level decision-making agent, organize other agents to jointly handle the sudden event.

[0098] Optionally, as shown in Figure 1 The planning agent can comprise a superior planning agent and an inferior planning agent.

[0099] In some embodiments, the superior planning agent receives a task and issues a pre-order to the inferior planning agent in combination with the commander's intention.

[0100] For example, the task received by the superior planning agent can be a complex scene task such as emergency response and joint action. For example, the task document received by the superior planning agent clearly indicates the core goal of delivering emergency supplies and professional forces to multiple dispersed key points.

[0101] Specifically, the superior planning agent first confirms the intention of the commander to accurately understand the priority of the key points, optimize the global delivery path, and ensure efficient allocation of supplies, and at the same time judges that the current situation of the task area is true, that is, the road traffic capacity is limited and the weather conditions are variable. Subsequently, the superior planning agent issues a pre-order to the inferior planning agent (such as the corresponding command agent in each protection direction) to start the joint emergency protection plan and develop multi-path collaborative delivery, so as to ensure the consistent understanding of the core requirements of the task by the superior and inferior planning agents.

[0102] Further, the superior planning agent and the subordinate planning agent generate multiple sets of plans in parallel, simulate and deduce each set of plans, and determine a decision scheme according to the commander's order.

[0103] For example, the superior planning agent generates multiple sets of plans such as air direct delivery + ground vehicle transportation, establishment of a forward transfer hub + end distributed distribution, resource centralized scheduling + on-demand dynamic allocation, etc. based on the emergency support task through rule matching; and the subordinate planning agent generates branch plans such as mountain path material delivery special plan and urban multi-point distribution plan according to the characteristics (such as terrain and distance) of the support direction.

[0104] Specifically, the superior planning agent and the subordinate planning agent both verify the plans through simulation and deduction. For example, the superior planning agent simulates the coverage range and time efficiency of different delivery modes, and the subordinate planning agent simulates the feasibility and safety risk of the transportation path under specific terrain. During the deduction process, the execution difficulties, resource requirements and expected effects of the plans are recorded synchronously, and finally all the plans are submitted to the commander. The commander selects a target plan of air and ground combination, establishment of a transfer station, and dynamic adjustment of the distribution scheme according to the real-time situation of the task area.

[0105] In this way, the superior planning agent and the subordinate planning agent generate and optimize multiple sets of decision plans through parallel operation and simulation and deduction, and provide the commander with scientific decision options.

[0106] Optionally, as shown in Figure 1 The preparation agent can include a superior preparation agent and a subordinate preparation agent.

[0107] In some embodiments, the superior preparation agent can group and evaluate resources according to the decision scheme, generate a level execution plan and issue it.

[0108] For example, the superior preparation agent groups the transportation units, loading and unloading teams, support materials and path planning models participating in the task according to the selected target plan, and establishes a command and organization relationship of "air delivery group - ground transportation group - material allocation group".

[0109] Specifically, the superior preparation agent calls the evaluation sub-module embedded therein to evaluate the carrying capacity (reliability), response scheduling degree (controllability) and matching degree with the planned route of each transportation unit. For example, whether the off-road performance of heavy transport vehicles meets the demand of mountainous road conditions, whether the range and load of unmanned aerial vehicles are suitable for air delivery scenarios, and based on the evaluation results, the resources are optimized to generate a level execution plan of "air group is responsible for A and B point material delivery, ground group travels along C and D routes, and allocation group carries out material sorting and redistribution at transfer station H".

[0110] Meanwhile, pre-action preparations (such as vehicle maintenance, material loading, and personnel standby) are carried out, and the execution plan is issued to the subordinate preparation agent.

[0111] Further, the subordinate preparation agent can receive the current execution plan, generate an adapted branch plan, and issue it to the corresponding secondary decision-making agent.

[0112] For example, after receiving the superior plan, the subordinate preparation agent reorganizes the resources under its jurisdiction, generates branch plans such as the "North Line Ground Transportation Plan" and the "South Line Air Support Plan", and issues them to the corresponding secondary decision-making agent.

[0113] Specifically, the subordinate preparation agent refines the "North Line Ground Transportation Plan" into "No. 1 vehicle team is responsible for the front section of fast traffic, and No. 2 vehicle team is responsible for the rear section of heavy transportation" based on the road conditions and vehicle team composition of the North Line, and specifies the respective departure timing, traffic section, communication liaison, and convergence point to form detailed instructions that can be immediately executed.

[0114] In this way, the present application converts the target plan into an executable specific action scheme by resource organization and plan refinement of the preparation agent, combined with resource credibility and matching degree evaluation, ensuring the rationality of resource allocation and the operability of the plan.

[0115] In some embodiments, the process of generating a decision scheme includes a first quantitative evaluation.

[0116] For example, the first quantitative evaluation includes at least a predictive evaluation of the adaptability of the plan and a matching evaluation of the credibility and controllability of the resources involved in the decision scheme.

[0117] The evaluation process is based on a pre-set quantitative index system (such as adaptability score, credibility score, and matching degree score), and the value range of each score is 0-100 points.

[0118] Specifically, the predictive evaluation of the adaptability of the plan is performed by the evaluation sub-module embedded in the planning agent, which gives a quantitative score by analyzing the degree of fit between the scheme and the task target, the degree of compatibility with scene constraints, and other indicators. For example, the adaptability score of a certain plan is 85 points, which meets the threshold requirement of "adaptability score ≥ 80 points".

[0119] Specifically, the matching evaluation of the credibility and controllability of the resource is performed by an evaluation submodule embedded in the preparation agent, a credibility and controllability score is given by analyzing historical execution records (such as the past task punctuality rate of the transportation unit) and current state (such as the intact rate of the vehicle) of the resource, and a task execution controllability score is given by analyzing the degree of fit between the resource function and the task demand. For example, the credibility score of a certain heavy transport vehicle team is 90 points, the controllability score is 88 points, and the task execution controllability score is 92 points, all of which meet the threshold requirements of the corresponding indicators.

[0120] It should be noted that the inter-agent state monitoring and interaction verification used in the first quantitative evaluation is based on the inter-agent communication protocol (A2A, ANP, ACP) with added credibility and controllability information fields, and the protocol message carries the quantitative evaluation results through the new fields to ensure reliable transmission of the evaluation data.

[0121] In this way, the present application performs double screening on the decision scheme and the resource through the first quantitative evaluation, avoiding the entry of schemes and resources with insufficient adaptability or low credibility into the execution process from the source, and improving the reliability of the decision scheme.

[0122] S202, receiving the empowerment application initiated by the secondary decision agent after a sudden event that cannot be independently handled in the execution of the execution plan, and performing a second quantitative evaluation.

[0123] For example, the secondary decision agent monitors the surrounding environment and task execution state in real time through its own perception submodule during the execution of the branch plan, and automatically triggers the empowerment application process when a sudden event that exceeds its disposal capacity is detected.

[0124] Specifically, taking the secondary decision agent executing the north line ground transportation plan as an example, it detects a sudden event that "the main road ahead cannot be passed due to a sudden situation, and the original route is blocked" during transportation, analyzes its own ability through the judgment submodule: only has the execution authority of the planned route transportation, no path re-planning decision right and related regional other transportation force dispatching right, cannot independently solve the problem, and then quantitatively evaluates (second quantitative evaluation) the impact of the disposal of the sudden event through the evaluation submodule.

[0125] For example, the impact of the route interruption on the overall support progress is evaluated, and the impact score is given as 75 points (the value range is 0-100 points, and the higher the score, the greater the impact), which meets the condition of "impact score ≥ 70 points to initiate empowerment application", and then initiates the empowerment application to the superior execution monitoring agent.

[0126] The application content can include an emergency event description, a self-disposal capability assessment, required decision-making authority (such as backup path planning authority, cooperative unit scheduling authority) and resource interface requirements.

[0127] In this way, the application ensures the pertinence and necessity of empowerment application through the emergency event autonomous identification and capability assessment of the secondary decision-making agent, combined with the influence degree judgment of the second quantitative assessment.

[0128] In the embodiment of the application, the second quantitative assessment at least includes an execution effect assessment of the disposal influence of the emergency event and its influence on the plan execution effect.

[0129] Exemplarily, the second quantitative assessment is performed by an evaluation sub-module embedded in the execution monitoring agent, and the evaluation indexes include the influence of the emergency event on the original plan execution progress, the influence on the task target achievement, the influence on the surrounding environment and other actions, and the like, and the comprehensive influence score is obtained through weighted calculation.

[0130] Specifically, after receiving the empowerment application, the superior execution monitoring agent calls the evaluation sub-module embedded therein, and further refines the evaluation in combination with the application content and the original plan requirements: the road interruption causes the transportation delay to be expected for 2 hours, which influences the material receiving of two key points (progress influence score 80 points); the delay may cause the subsequent operation of the point to be stagnant, which influences the global task rhythm (target influence score 78 points); there is no other equivalent route that can be immediately replaced, and other transportation units nearby need to be coordinated for detour support, which may affect their original tasks (associated influence score 72 points), and the weighted calculation comprehensive influence score is 77 points, which meets the threshold requirement of “comprehensive influence score ≥ 75 points to respond to the empowerment application”.

[0131] It should be noted that in the second quantitative assessment process, the interaction verification between the agents is also based on the A2A, ANP and ACP protocols with added trusted value and controllable value fields, to ensure the authenticity of the evaluation data and the transmission security.

[0132] In this way, the application objectively judges the influence of the emergency event through the second quantitative assessment, and provides a quantitative basis for whether to respond to the empowerment application, avoiding the blindness of the authority adjustment.

[0133] S203, in response to the empowerment application and authorized by the commander, dynamically configuring temporary command decision-making authority and / or supplemented resources and capability interfaces for the secondary decision-making agent that proposes the empowerment application, and updating the role identifier after configuring the temporary command decision-making authority.

[0134] In some embodiments, the execution monitoring agent can evaluate the empowerment application to generate an authorization proposal.

[0135] Exemplarily, the execution monitoring agent (superior or subordinate) audits the rationality of the empowerment application, the necessity of the required permissions, and the availability of the resource interface based on the second quantitative evaluation result, the current resource configuration state, and the task priority.

[0136] Specifically, the execution monitoring agent analyzes the backup path planning right and the cooperative unit scheduling right required in the empowerment application, confirms that the permissions belong to the command type of the first decision-making agent, and that there are currently other transport unit resources that can be scheduled, and the interface can be normally called; combined with the task priority (emergency support is high priority), it is judged that granting the permission will not affect the execution of other core tasks, and then generates the authorization proposal of "agree to empower, grant backup path planning right and cooperative unit scheduling right, open related transport unit resource interface, and empower effective period until the alternative route is opened and the goods are delivered to the target point".

[0137] Further, after receiving the confirmation instruction of the commander on the authorization proposal, the temporary command decision-making right and / or the supplemented resource and capability interface are dynamically configured for the second decision-making agent who proposed the empowerment application.

[0138] Exemplarily, the execution monitoring agent submits the authorization proposal to the commander, and the commander views the application details, evaluation results and authorization proposal through the command terminal, and confirms that there is no error and issues the "agree to authorize" instruction.

[0139] Specifically, after receiving the commander's instruction, the execution monitoring agent dynamically configures the command decision-making right for the second decision-making agent, updates its role identifier to "temporary first decision-making agent", and enables it to have the core functions of the command type of agent; at the same time, the corresponding resource interface is opened, so that it can call external resources such as geographic information system (used for planning backup path) and surrounding transport unit scheduling system.

[0140] It should be noted that the calling process of external resources is based on the resource calling protocol (such as MCP) with added identity authentication and calling audit mechanism. When the second decision-making agent calls the resource, it needs to pass the identity authentication (based on the pre-set trusted identifier), and the system automatically records the audit information such as calling time, calling content, and execution result to ensure the traceability of resource calling.

[0141] For example, the empowered second decision-making agent queries the surrounding road network conditions through the geographic information system interface, plans an alternative route of "detouring X number of backup roads"; through the transport unit scheduling interface, coordinates a nearby reserve vehicle team, issues a scheduling instruction of "cooperating to execute the transportation task of the detour section", and adjusts the travel timing of itself and other related units to ensure that the goods are delivered as soon as possible.

[0142] Further, after configuring the temporary command decision-making right, the role identifier is updated.

[0143] Exemplarily, the update of the role identity is completed in synchronization with the system authority management hub, the state of the secondary decision-making agent in the system changes from "executor" to "temporary commander", and the collaborative network accessed by the secondary decision-making agent, the type of instructions receivable and the instruction authority deliverable are all adjusted accordingly.

[0144] Specifically, after the system updates its identity, other secondary decision-making agents at the same level can identify its new role and receive instructions for organizational collaboration; the superior agent can also see the change of its role state in the monitoring interface, facilitating global control.

[0145] In this way, the present application realizes precise and safe configuration of authority and resources through the closed-loop process of audit-authorization-configuration combined with the trusted calling mechanism of the MCP protocol, enabling the secondary decision-making agent to quickly acquire the ability to handle emergencies.

[0146] S204, after the completion of the emergency handling, the temporary command decision-making authority and the supplemented resources and capability interfaces dynamically configured for the secondary decision-making agent are recovered, and the role identity thereof is restored to the original state.

[0147] In some embodiments, after the secondary decision-making agent (temporary primary decision-making agent) completes the emergency handling, it automatically reports the task completion situation to the execution monitoring agent that authorized its empowerment.

[0148] The task completion situation can include emergency handling results, authority usage, resource calling records, and task summary abstracts.

[0149] Exemplarily, after completing the tasks of "alternative route planning", "collaborative unit scheduling", and "material transportation", the summary abstract reported by the empowered secondary decision-making agent clearly states that "the alternative route is passable, the materials have arrived at the target point, the overall progress delay is controlled within 1 hour, and the authority and resources are not used out of scope".

[0150] Specifically, after receiving the reported information, the execution monitoring agent calls the evaluation submodule embedded therein to evaluate the handling effect, confirms that the emergency has been properly solved and the task completion effect meets the requirements, and then sends a "authority recovery" instruction to automatically revoke the command decision-making authority configured for the agent, closes the corresponding resource interface, restores its role identity to a secondary decision-making agent, and updates the authority change log and resource calling audit record to form a complete closed loop.

[0151] In this way, the present application realizes precise and safe configuration of authority and resources through the closed-loop process of audit-authorization-configuration combined with the trusted calling mechanism of the MCP protocol, enabling the secondary decision-making agent to quickly acquire the ability to handle emergencies.

[0152] S205, monitor the evaluation result of the first quantitative evaluation or the second quantitative evaluation, when the evaluation result is lower than the corresponding threshold value, interrupt or suspend the current process, and request the commander to intervene.

[0153] In the embodiments of the present application, the threshold value of the first quantitative evaluation can include a scheme adaptability threshold value (such as 80 points), a resource reliability threshold value (such as 85 points), and a resource matching degree threshold value (such as 80 points); and the threshold value of the second quantitative evaluation includes a comprehensive influence degree threshold value (such as 75 points).

[0154] The threshold value can be preset or dynamically adjusted by the commander according to the task type and scene requirements.

[0155] For example, during the generation of a decision scheme, the adaptability quantitative evaluation score of a certain plan is 75 points, which is lower than the preset threshold value of 80 points. At this time, the decision scheme generation process is automatically interrupted, and the evaluation report, score details, and improvement suggestions of the plan are submitted to the commander, requesting the commander to intervene to judge whether to continue optimizing the plan or to generate a new plan.

[0156] Specifically, the commander checks the evaluation report and finds that the lack of adaptability of the plan is mainly due to "not fully considering the constraints of the task area night traffic capacity", and then issues the instruction "supplement the night action plan and re-deduce the evaluation". The superior planning agent optimizes the plan in combination with the instruction, re-performs simulation deduction and quantitative evaluation, and continues until the score meets the threshold requirement.

[0157] In another exemplary scenario, the comprehensive influence score of the second quantitative evaluation is 72 points, which is lower than the preset threshold value of 75 points. The enablement application response process is suspended, and the evaluation result and application details are submitted to the commander. The commander judges that the impact of the emergency event is limited, and can be solved by the secondary decision-making agent through existing resources, without granting command-type permissions. Then, the commander issues the instruction "reject the enablement application and guide the secondary decision-making agent to use local resources to try to clear or wait for a short time", and feeds back the instruction to the agent.

[0158] It should be noted that when critical operations such as decision-making permission adjustment and high-risk resource allocation are involved, even if the quantitative evaluation result meets the threshold requirement, an intervention reminder can be sent to the commander, and the subsequent process is executed after the commander confirms it, to ensure the safety of human-machine cooperation.

[0159] In this way, the present application builds a dual control mode of automatic process + manual intervention through quantitative evaluation and threshold intervention mechanism, which not only guarantees the process efficiency in regular scenarios, but also introduces manual judgment in critical nodes or abnormal situations, ensuring the controllability of the whole decision-making process.

[0160] In the method, through the combination of hierarchical architecture and quantitative evaluation, full-link enhancement from static planning to dynamic execution is realized; in the scheme generation stage, the first quantitative evaluation is performed synchronously, which can exclude schemes with poor adaptability or unreliable resources in advance, thereby guaranteeing the reliability of the decision basis from the source; when a sudden event occurs during execution, the second quantitative evaluation is triggered, which can quantitatively evaluate the event impact in real time, thereby providing accurate and objective decision basis for subsequent dynamic adjustment of authority; on this basis, the system supports the temporary authority and interface configuration of the execution body and post-recovery through personnel authorization based on the evaluation results, so that the authority adjustment is no longer a rigid preset, but becomes a data-driven and safely controlled flexible response capability; finally, the continuous monitoring of the two evaluation results and the setting of intervention thresholds constitute a double insurance, which ensures that any decrease in reliability can be captured in time and handed over to personnel for takeover, thereby realizing efficient cooperation across levels, dynamic and safe adjustment of decision-making authority, and reliable and controllable decision-making process.

[0161] Optionally, as described in S201 above, the evaluation agent has a first working mode and a second working mode.

[0162] In an optional implementation, in the first working mode, an evaluation submodule embedded in the planning agent is configured to perform the prediction evaluation of the adaptability of the preplan in the first quantitative evaluation.

[0163] For example, the evaluation submodule embedded in the planning agent is configured to preset multi-dimensional adaptability evaluation indexes, including task target fitness, scene constraint compatibility, resource demand feasibility, and execution difficulty controllability.

[0164] Specifically, in the evaluation of the emergency support preplan, the task target fitness index evaluates the coverage degree of the preplan on the core target of “timely and sufficient delivery of materials and efficient allocation of forces”; the scene constraint compatibility index evaluates the adaptability of the preplan to the task area constraints such as “road traffic conditions and weather changes”; the resource demand feasibility index evaluates the availability of the resources required by the preplan; and the execution difficulty controllability index evaluates the controllability of the risks in the execution process of the preplan, and each index is weighted and calculated according to the preset weight to obtain the adaptability quantitative score.

[0165] In another optional implementation, in the first working mode, an evaluation submodule embedded in the preparation agent is configured to perform the matching evaluation of the reliability, controllability and task execution controllability of the resources involved in the decision scheme in the first quantitative evaluation.

[0166] Exemplarily, the evaluation sub-module embedded in the preparation intelligent agent sets differentiated evaluation indexes for different types of resources such as transportation units, support personnel, path planning models, etc. The credibility index of the transportation unit includes the past task punctuality rate and the equipment intact rate, and the controllability index includes the remote scheduling success rate and the fault response speed. The credibility index of the support personnel includes the professional skill certification level and the historical task evaluation, and the controllability index includes the instruction response speed and the coordination degree. The credibility index of the model resource includes the path prediction accuracy rate and the historical error rate, and the controllability index includes the parameter adjustment flexibility and the abnormal processing capability.

[0167] Specifically, when evaluating a heavy transport vehicle team (transportation unit resource), the past long-distance transportation task punctuality rate is 95% (credibility index score 95), the response time of accepting remote path re-planning is 30 minutes (controllability index score 90), and the off-road transportation function is completely matched with the current task demand (matching degree score 100). The comprehensive evaluation result of the resource meets the threshold requirement.

[0168] In another alternative implementation, in the first working mode, the evaluation sub-module embedded in the execution monitoring intelligent agent is used to perform the evaluation of the impact of the handling of the emergency on the execution effect of the plan in the second quantitative evaluation.

[0169] Exemplarily, the evaluation indexes of the evaluation sub-module include progress impact, target impact, correlation impact, risk diffusion impact, etc. Each index is set with a weight according to the priority of the task scene. For example, in the emergency support scene, the target impact (the impact on the overall task rhythm and the point position support effect) has the highest weight.

[0170] Specifically, when evaluating the “road interruption” emergency, the progress impact index evaluates the delay time of the transportation plan, the target impact index evaluates the influence degree on the key point position material receiving target, the correlation impact index evaluates the linkage influence on other parallel transportation tasks, and the risk diffusion impact index evaluates whether it may cause traffic congestion or secondary accidents. The comprehensive impact score is calculated by weighted calculation.

[0171] In another alternative implementation, in the second working mode, an independent evaluation intelligent agent is used to perform the third quantitative evaluation of the overall completion effect of the decision scheme.

[0172] Exemplarily, after the task is completely finished, the independent evaluation intelligent agent is started. It collects all the data generated in the whole process from planning, preparation, execution to monitoring, including the original scores of the quantitative evaluation in each stage, the commander's instruction record, the resource actual call log, the final achievement state of the task, etc.

[0173] Specifically, the third quantitative evaluation sets multi-dimensional effect evaluation indexes from a global perspective, such as a task target achievement rate, a scheme execution efficiency ratio, a resource use efficiency, and a system coordination reliability, etc. For example, it calculates the actual delivery amount / plan delivery amount to obtain the target achievement rate, and analyzes the execution efficiency by comparing the actual time consumption / planned deduction time consumption, thereby generating an objective and comprehensive task summary report for iterative optimization of future decision schemes and evaluation models.

[0174] In this way, the application realizes closed-loop inspection and feedback of task effects through the third quantitative evaluation, and continuously improves the decision and execution capabilities of the system.

[0175] Optionally, in addition to the first quantitative evaluation and / or the second quantitative evaluation described in S201 and S202, the inter-agent communication protocol with the added trusted value and controllable value information fields can be used to monitor and interactively verify the agent cooperation state.

[0176] In some embodiments, the trusted value and controllable value fields are added to the message structure of the inter-agent communication protocol (A2A, ANP, ACP), the field length is 1 byte, the value range is 0-100 (corresponding to percentage), and the quantitative evaluation results of the agent are carried in real time.

[0177] For example, when the superior planning agent sends a pre-order to the subordinate planning agent, the A2A protocol message adds fields to carry information such as “superior planning agent trusted value: 98 points, controllable value: 99 points” and “pre-plan adaptability trusted value: 85 points”. After receiving the information, the subordinate planning agent verifies the field value to determine the credibility of the information. If the trusted value is less than 80 points, the subordinate planning agent feeds back to the superior agent “information verification exception, please resend”.

[0178] Specifically, the protocol also supports encrypted transmission of the field value, and uses a symmetric encryption algorithm to encrypt the trusted value and controllable value fields to prevent data tampering and ensure the security and reliability of inter-agent interaction.

[0179] In this way, the application realizes real-time transmission and security verification of evaluation data by integrating the quantitative evaluation results into the communication process through the extended inter-agent communication protocol, thereby providing technical support for the trustworthiness of cross-agent cooperation.

[0180] Optionally, in addition to the first quantitative evaluation and / or the second quantitative evaluation described in S201 and S202, the resource calling protocol with the added identity authentication and calling audit mechanism can be used to verify the credibility of external resources.

[0181] The external resources include external models, external databases, and external tools.

[0182] Exemplarily, the external model refers to a path planning model, a traffic flow prediction model and the like deployed on a third-party server; the external database is an external data storage system for storing regional geographic information, real-time state of a road network, resource inventory data; and the external tool refers to a geographic information query tool, a resource scheduling tool and the like provided by a third party.

[0183] In some embodiments, the MCP extends the identity authentication process and the calling audit field, the identity authentication adopts a triple authentication mechanism of an agent identity, a key and a trusted certificate, and the calling audit field records information such as calling time, calling agent identity, resource use duration and execution result.

[0184] Exemplarily, when the secondary decision agent calls an external geographic information system (database resource), the secondary decision agent first sends an identity authentication request to a resource server, carries the agent identity, the preset key and the trusted certificate issued by the system, the resource server verifies and passes, opens a query interface, and records calling audit information (such as calling time: XXXX year XX month XX day XX hour XX minute, calling agent: secondary decision agent-north line transportation No. 1, query content: real-time traffic state of standby road X, and execution result: query success).

[0185] Specifically, during the calling process, the protocol monitors the data transmission state in real time, and if data loss, tampering or the like occurs, the calling is immediately interrupted and abnormal log is recorded and reported to the execution monitoring agent; after the calling is completed, the audit information is synchronously uploaded to the system log server for storage for future reference, so that the whole process of resource calling is traceable.

[0186] In this way, the application constructs a trusted link for external resource calling by extending the security mechanism of the resource calling protocol, avoids the risk of untrusted resource access and malicious calling, and ensures the security of system and external resource interaction.

[0187] The complete process of the trusted human-machine hybrid intelligent decision-making method based on group agents provided by the application will be described in detail below. Figure 3 The complete process of the trusted human-machine hybrid intelligent decision-making method based on group agents provided by the application will be described in detail below.

[0188] Exemplarily, as shown in the figure, Figure 3 the complete process can be divided into the following four main stages: Stage one: task initiation and parallel planning.

[0189] S301, the commander initiates a task and issues the task to a superior first-level decision agent.

[0190] For example, the commander defines the core mission objectives based on the actual business scenario requirements (such as emergency support and complex task scheduling), and pushes the mission document to the superior first-level decision-making intelligent agent through a dedicated command terminal as the trigger condition for starting the entire decision-making process.

[0191] S302, The superior first-level decision-making intelligent agent understands the execution intention of the planning intelligent agent and confirms it with the commander.

[0192] For example, after receiving the task document, the superior planning agent analyzes the core objectives, constraints and key requirements of the task, and simultaneously feeds back the intent understanding results to the commander's terminal. After the commander confirms that the intent understanding is accurate, the process proceeds to the next stage.

[0193] S303, The superior authority assesses the execution status of the intelligent agent's plan and confirms it with the commander.

[0194] For example, the superior planning intelligent agent combines the scenario information corresponding to the task (such as the regional road network conditions and resource distribution of the emergency support scenario) to conduct situation analysis, clarify the current task execution difficulties, resource gaps and other information, and submit the judgment results to the commander for confirmation before entering the plan generation stage.

[0195] S304. The superior planning agent generates multiple contingency plans. During this process, the embedded evaluation submodule synchronously performs the prediction evaluation of the contingency plan's adaptability in the first quantitative evaluation. Subsequently, the contingency plans are simulated and deduced, and finally the commander selects and determines the target contingency plan.

[0196] For example, the superior planning agent generates multiple alternative plans based on the confirmed intentions and circumstances; first, the embedded evaluation submodule evaluates the suitability of each plan (such as its fit with the task objectives), and then verifies the feasibility of the plan (such as resource scheduling efficiency) through simulation and deduction. The evaluation and deduction results are then submitted to the commander, who selects the target plan that meets the task requirements.

[0197] S305. The superior planning agent issues a pre-command to the planning agent of the next lower-level decision-making agent.

[0198] For example, the superior planning agent, based on the selected target plan, clarifies the task division and core collaboration requirements of the subordinate agents, and issues pre-commands to the subordinate planning agents.

[0199] S306. The superior planning agent issues its self-generated target plan to the planning agent of the next lower-level decision-making agent.

[0200] S307. Lower-level planning agents, based on pre-issued commands, conduct their own planning work in parallel with higher-level planning agents. The lower-level planning agents understand the execution intent, which is then confirmed by their own commander.

[0201] Exemplarily, after receiving the pre-order, the lower-level planning agent analyzes the task intention and division requirement of the superior, feeds back the understanding result of its own intention to the corresponding commander, and confirms the understanding with the commander.

[0202] S308, the subordinate planning agent executes the situation judgment and confirms with the commander of the level.

[0203] Exemplarily, the subordinate planning agent independently conducts situation research and judgment in combination with the actual situation of its jurisdiction area (such as the resource allocation of the partition, the geographical environment), and determines the execution constraints and key links of the task of the level, and submits the judgment result to the corresponding commander for confirmation.

[0204] S309, the subordinate planning agent generates an adaptive branch plan and performs simulation deduction; in this process, the embedded evaluation submodule also performs plan adaptability evaluation; finally, the commander of the level selects and determines the target plan of the level.

[0205] Exemplarily, the subordinate planning agent generates an adaptive branch plan based on the confirmed intention of the superior and the situation of the level; through the embedded evaluation submodule, the plan is evaluated, and after the feasibility is verified through simulation deduction, the result is submitted to the corresponding commander to select the target plan of the level.

[0206] S310, the subordinate planning agent reports the target plan of the level.

[0207] Phase two: resource preparation and plan decomposition.

[0208] S311, the preparation agent of the superior first-level decision-making agent generates a plan according to the target plan. In this process, the embedded evaluation submodule synchronously performs the matching evaluation of the matching degree and credibility of resources and tasks in the first quantitative evaluation.

[0209] Exemplarily, the superior preparation agent classifies and groups the resources such as transportation units, support personnel, and models participating in the task according to the selected target plan, establishes a clear command level and organizational cooperation relationship. In this process, the embedded evaluation submodule is called to evaluate the credibility (such as equipment intact rate), controllability (such as response speed) and matching degree with the task of the resources.

[0210] S312, the superior preparation agent groups resources, establishes command / organization relationship, and performs pre-preparation according to the target plan.

[0211] Exemplarily, the superior preparation agent conducts resource scheduling (such as equipment maintenance, material loading), personnel gathering and other pre-actions in advance based on the generated plan.

[0212] S313, the superior preparation agent issues the generated overall plan to the subordinate preparation agent.

[0213] Exemplarily, the superior preparation agent will issue the overall plan including task flow, resource allocation, time node to the subordinate preparation agent, and explicitly the execution requirements and coordination rhythm of the subordinate.

[0214] S314, the preparation agent of the subordinate first-level decision-making agent generates the current-level plan according to the overall plan. The matching evaluation of the matching degree and credibility of the resources and tasks is carried out synchronously.

[0215] Exemplarily, the subordinate preparation agent reorganizes the resources and personnel under its jurisdiction according to the overall plan of the superior, establishes the command and cooperation relationship suitable for the current-level task, and evaluates the matching degree of the resources and the current-level task.

[0216] S315, the subordinate preparation agent executes the pre-preparation of the current level.

[0217] Exemplarily, the subordinate preparation agent carries out the pre-preparation of resource scheduling, personnel preparation, etc. of the current level based on the branch plan generated by itself, to ensure the matching of the execution rhythm with the superior plan.

[0218] S316, the subordinate preparation agent reports the plan generated by itself.

[0219] S317, the subordinate preparation agent issues the specific branch execution plan to the corresponding second-level decision-making agent.

[0220] Exemplarily, the subordinate preparation agent issues the branch plan refined to the execution level (such as specific transportation route, operation instruction) to the corresponding second-level decision-making agent.

[0221] Stage three: dynamic execution and monitoring response.

[0222] In this stage, the execution monitoring agent at each level continuously monitors the task execution state, and its activities include the following parallel or sequential steps: S318, the execution monitoring agent of the superior first-level decision-making agent monitors the conflict (such as internal organization of resources).

[0223] Exemplarily, the superior execution monitoring agent tracks the resource usage and agent cooperation state in the global task execution process in real time, identifies whether there is a resource scheduling conflict (such as the same device is repeatedly allocated) or agent action interference, etc. using data analysis, and records the conflict details.

[0224] S319, the superior execution monitoring agent identifies the emergency event (such as external emergency event).

[0225] Exemplarily, the superior execution monitoring intelligent agent obtains abnormal information in the task scene (such as interruption of the main transportation route in emergency support) through a connected external perception interface or a subordinate report, and immediately performs real-time evaluation on the influence range, emergency degree and preliminary disposal direction of the emergency.

[0226] S320, the superior execution monitoring intelligent agent evaluates the event / task execution effect, and executes after being confirmed by the commander.

[0227] Exemplarily, the superior execution monitoring intelligent agent quantitatively evaluates the execution progress and stage effect of the current overall task or specific event, feeds back the evaluation result and suggestion to the commander, and continues to promote or adjust the process after the commander confirms that the execution meets the expectation or adjusts the scheme.

[0228] Further, after S320, the superior execution monitoring intelligent agent receives the application and immediately calls the embedded evaluation submodule to perform a second quantitative evaluation to determine whether the emergency event affects / does not achieve the planned execution effect.

[0229] Exemplarily, the execution monitoring intelligent agent obtains emergency event information through a perception interface, and calls the embedded evaluation submodule to perform real-time evaluation on the influence range, disposal difficulty and effect influence on the overall plan of the event.

[0230] S321, if it is determined that the influence / does not achieve the planned execution effect, the plan is adjusted.

[0231] S322, the execution monitoring intelligent agent of the subordinate first-level decision-making intelligent agent monitors conflicts (such as internal organization of resources).

[0232] Exemplarily, the subordinate execution monitoring intelligent agent focuses on the execution of the branch plan at this level, monitors the resource use and intelligent agent cooperation state at this level, identifies and records resource conflicts or operation conflicts in the jurisdiction, and synchronously reports important conflict information to the superior execution monitoring intelligent agent.

[0233] S323, the subordinate execution monitoring intelligent agent identifies an emergency event (such as an event facing external emergencies).

[0234] Exemplarily, the subordinate execution monitoring intelligent agent perceives an abnormal situation (such as a designated transport vehicle fleet unable to pass according to the plan) in the area it is responsible for, immediately performs a rapid evaluation on the influence and disposal difficulty of the local emergency event, and pushes the event information and preliminary evaluation result to the superior execution monitoring intelligent agent.

[0235] S324, the subordinate execution monitoring intelligent agent evaluates the event / task execution effect, and executes after being confirmed by the commander.

[0236] For example, the lower-level execution monitoring agent continuously evaluates the execution effect of events or tasks at its level, and feeds back the evaluation results to the commander at its level or the commander at the higher level. After confirmation by the commander, the agent guides the execution or adjustment of subsequent actions at its level.

[0237] Furthermore, after S324 above, after the lower-level execution monitoring agent receives the application, it immediately calls the embedded evaluation submodule to perform a second quantitative evaluation to determine whether the sudden event affects / fails to achieve the planned execution effect.

[0238] For example, the monitoring agent obtains information about emergencies through the perception interface and calls the embedded evaluation submodule to conduct real-time evaluation of the scope of the event's impact, the difficulty of handling it, and its effect on the overall plan.

[0239] S325. If it is determined that the plan has been affected or has not achieved its intended effect, the plan shall be adjusted.

[0240] In the context of this continuous monitoring, emergencies may be detected by the secondary decision-making agent and trigger the following dynamic response closed loop.

[0241] S326. The secondary decision-making agent identifies unexpected events during the execution of the plan.

[0242] For example, during the execution of a branch plan, the secondary decision-making agent obtains abnormal information in the task scenario (such as road collapse during material transportation) through its own perception module and identifies sudden events that exceed the normal execution scope.

[0243] S327. The secondary decision-making agent assesses the event and determines that it cannot be handled independently.

[0244] For example, a secondary decision-making agent assesses the difficulty of handling an emergency based on its own execution authority, resource allocation, and handling capabilities, and determines whether it has the conditions to handle the situation independently.

[0245] S328. If it is determined that the task cannot be completed independently, the secondary decision-making agent initiates an empowerment request to the lower-level execution monitoring agent, which then forwards the request to the higher-level execution monitoring agent.

[0246] For example, when the secondary decision-making agent determines that it cannot handle the situation independently, it sends an empowerment request to the superior execution monitoring agent.

[0247] S329. After authorization by the commander, the secondary decision-making agent that submitted the application is authorized and its role identifier is updated and / or it is empowered.

[0248] Exemplarily, after being authorized by the commander, the secondary decision-making agent obtains temporary command decision-making authority and / or supplemental resource and capability interface, and updates the role identification to a temporary primary decision-making agent after obtaining the temporary command decision-making authority.

[0249] S330, the secondary decision-making agent analyzes and generates treatment instructions.

[0250] And based on the analysis of the emergency scene, the treatment plan is generated, and specific execution instructions (such as dispatching standby resources and adjusting transportation routes) are generated.

[0251] S331, the secondary decision-making agent executes the treatment instructions.

[0252] Exemplarily, the temporary primary decision-making agent executes the treatment operation of the emergency according to the generated treatment instructions by calling corresponding collaborative agents and resources.

[0253] S332, the secondary decision-making agent judges the task completion condition.

[0254] Exemplarily, the temporary primary decision-making agent evaluates the execution progress and effect of the treatment operation.

[0255] S333, if the planned requirements are not met, continue to execute.

[0256] Exemplarily, if the evaluation result shows that the treatment target has not been reached, the temporary primary decision-making agent adjusts the execution strategy and continues to execute the treatment instructions.

[0257] Further, if the planned requirements are met, the judgment step after S324 is executed.

[0258] Further, after the temporary primary decision-making agent completes the treatment, the authority is released and the role is restored to the secondary decision-making agent.

[0259] Exemplarily, when the emergency treatment is completed and the effect meets the requirements, the system automatically revokes the temporary authority and interface configured for the agent, and restores its role identification to the original secondary decision-making agent.

[0260] Phase four: global effect evaluation and summary.

[0261] S334, if the emergency does not affect / achieve the planned execution effect after S320, the independent evaluation agent (second working mode) of the superior primary decision-making agent collects and summarizes the data.

[0262] Exemplarily, the superior evaluation agent collects data of the whole task execution process (such as preplan execution data, resource use data, and emergency treatment data), and classifies, summarizes and structures the data.

[0263] S335, the superior evaluation agent performs a third quantitative evaluation on the execution effect of the decision scheme, and the commander confirms.

[0264] Illustratively, the superior evaluation agent performs a third quantitative evaluation on the execution effect of the overall scheme and the achievement of the target based on the summary data, and submits the evaluation result to the commander for confirmation.

[0265] S336, if the execution effect is achieved, a final summary report is generated.

[0266] Illustratively, if the evaluation result shows that the execution effect of the scheme meets the standard, the superior evaluation agent generates a task summary report, which forms a final document after being confirmed by the commander.

[0267] S337, if the execution effect is not achieved, the scheme is adjusted and re-executed.

[0268] S338, if the sudden event does not affect / achieve the planned execution effect after S324, the independent evaluation agent of the subordinate decision agent collects summary data.

[0269] Illustratively, the subordinate evaluation agent collects various data in the execution process of the task at this level, and synchronously submits the classified and summarized data to the superior evaluation agent.

[0270] S339, the subordinate evaluation agent evaluates the overall completion effect of the scheme at this level, and the commander confirms.

[0271] Illustratively, the subordinate evaluation agent evaluates the execution effect of the branch plan based on the summary data at this level, and submits the result to the corresponding commander for confirmation.

[0272] S340, if the effect is achieved, a summary report at this level is generated and synchronized to the superior.

[0273] Illustratively, if the execution effect of the plan at this level meets the standard, the subordinate evaluation agent generates a task summary report at this level, which is synchronized to the superior evaluation agent after being confirmed by the corresponding commander.

[0274] S341, if the effect is not achieved, the scheme is adjusted and re-executed.

[0275] Illustratively, if the evaluation result shows that the planned effect is not achieved, the subordinate evaluation agent adjusts the branch scheme in combination with the problem causes, and re-issues the execution after being confirmed by the corresponding commander.

[0276] In this way, the present application realizes efficient collaboration, dynamic adaptation and reliable control of group intelligent agent decision-making through the hierarchical and parallel, dynamic response intelligent agent collaboration process and the quantitative evaluation mechanism throughout the process, effectively improving the flexibility, reliability and traceability of task execution in complex scenarios.

[0277] Figure 4 An example of a schematic diagram of a physical structure of an electronic device is shown in FIG. 4A, which can include a processor 410, a communications interface 420, a memory 430, and a communications bus 440, wherein the processor 410, the communications interface 420, and the memory 430 can communicate with each other through the communications bus 440. Figure 4

[0278] The processor 410 can invoke a logical instruction in the memory 430 to execute a trusted human-machine hybrid intelligent decision-making method based on swarm agents, which includes generating a decision scheme by a primary decision-making agent and issuing an execution plan based on the decision scheme to a secondary decision-making agent; wherein the process of generating a decision scheme includes first quantitative evaluation, which at least includes predicting and evaluating the adaptability of the pre-plan, and matching the trustworthiness and controllability of the resources involved in the decision scheme; receiving the empowerment application initiated by the secondary decision-making agent after the sudden event that cannot be handled independently during the execution of the execution plan is autonomously judged, and performing second quantitative evaluation, which at least includes evaluating the execution effect of the disposal impact of the sudden event and its impact on the plan execution effect; in response to the empowerment application and the authorization of the commander, dynamically configuring temporary command decision-making authority and / or supplemental resources and capability interfaces for the secondary decision-making agent that proposes the empowerment application, and updating its role identifier after configuring the temporary command decision-making authority; after the sudden event is handled, recycling the temporary command decision-making authority and the supplemental resources and capability interfaces dynamically configured for the secondary decision-making agent, and restoring its role identifier to the original state; monitoring the evaluation results of the first quantitative evaluation or the second quantitative evaluation, and when the evaluation results are lower than the corresponding threshold, interrupting or suspending the current process, and requesting the commander to intervene.

[0279] ​Moreover, the logic instructions in the memory 430 described above can be implemented in the form of software functional units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0280] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the trusted human-machine hybrid intelligent decision-making method based on swarm agents provided by the above-mentioned methods, the method comprising: generating a decision scheme by a primary decision-making agent, and issuing an execution plan obtained by decomposing the decision scheme to a secondary decision-making agent; wherein the process of generating the decision scheme comprises a first quantitative evaluation, the first quantitative evaluation at least includes predicting and evaluating the adaptability of the preplan, and matching the trustworthiness and controllability of the resources involved in the decision scheme; receiving the empowerment application initiated by the secondary decision-making agent after the secondary decision-making agent autonomously judges that the unexpected event cannot be handled independently during the execution of the execution plan, and performing a second quantitative evaluation, the second quantitative evaluation at least includes evaluating the execution effect of the unexpected event and its influence on the plan execution effect; in response to the empowerment application and the authorization of the commander, dynamically configuring temporary command decision-making authority and / or supplemental resources and capability interfaces for the secondary decision-making agent that proposes the empowerment application, and updating its role identifier after configuring the temporary command decision-making authority; after the unexpected event is handled, recycling the temporary command decision-making authority and the supplemental resources and capability interfaces dynamically configured for the secondary decision-making agent, and restoring its role identifier to the original state; monitoring the evaluation results of the first quantitative evaluation or the second quantitative evaluation, and when the evaluation results are lower than the corresponding threshold, interrupting or suspending the current process, and requesting the commander to intervene.

[0281] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements a trusted human-machine hybrid intelligent decision-making method based on swarm agents provided by the above method, the method comprising: generating, by a primary decision-making agent, a decision scheme, and issuing an execution plan based on the decision scheme to a secondary decision-making agent; wherein the process of generating the decision scheme comprises a first quantitative assessment, which at least includes predicting and evaluating the adaptability of the preplan, and matching the trustworthiness and controllability of the resources involved in the decision scheme; receiving an empowerment application initiated by the secondary decision-making agent after a sudden event that cannot be handled independently during the execution of the execution plan is autonomously judged, and performing a second quantitative assessment, which at least includes evaluating the execution effect of the sudden event and its impact on the plan execution effect; in response to the empowerment application and the authorization of the commander, dynamically configuring temporary command decision-making authority and / or supplemental resource and capability interface for the secondary decision-making agent that proposes the empowerment application, and updating its role identifier after configuring the temporary command decision-making authority; after the sudden event is handled, recycling the temporary command decision-making authority and the supplemental resource and capability interface dynamically configured for the secondary decision-making agent, and restoring its role identifier to the original state; monitoring the evaluation results of the first quantitative assessment or the second quantitative assessment, and when the evaluation results are lower than the corresponding threshold, interrupting or suspending the current process, and requesting the intervention of the commander.

[0282] The system embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place or distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments. Those skilled in the art can understand and implement without creative labor.

[0283] From the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0284] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents, characterized in that, The method is executed based on a hierarchical cascaded architecture comprising a first-level decision agent and a second-level decision agent; the method includes: The first-level decision-making agent generates a decision scheme and distributes the execution plan obtained from the decomposition of the decision scheme to the second-level decision-making agent; wherein, the process of generating the decision scheme includes a first quantitative evaluation, which includes at least a predictive evaluation of the suitability of the plan and a matching evaluation of the credibility and controllability of the resources involved in the decision scheme. The system receives an empowerment request initiated by the secondary decision-making agent after it autonomously determines that the emergency is one that cannot be handled independently during the execution of the execution plan, and performs a second quantitative assessment. The second quantitative assessment includes at least an execution effect assessment of the impact of the handling of the emergency and its impact on the execution effect of the plan. In response to the empowerment request and with the authorization of the commander, the secondary decision-making agent that made the empowerment request is dynamically configured with temporary command and decision-making permissions and / or supplementary resource and capability interfaces, and its role identifier is updated after configuring the temporary command and decision-making permissions. After the emergency is handled, the temporary command and decision-making authority and the supplementary resource and capability interfaces that were dynamically configured for the secondary decision-making agent are revoked, and its role identifier is restored to its original state. Monitor the evaluation results of the first or second quantitative assessment. When the evaluation result is lower than the corresponding threshold, interrupt or suspend the current process and request the commander to intervene.

2. The trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents according to claim 1, wherein the first-level decision-making intelligence agent includes a planning intelligence agent, a preparation intelligence agent, an execution monitoring intelligence agent, and an evaluation intelligence agent; in, The first-level decision-making agents at the same level support sequential collaboration in a ring-shaped cascade, and the first-level decision-making agents at different levels support cross-level vertical linkage collaboration. The secondary decision-making agent is used for: Execute the execution plan issued by the first-level decision-making agent; or, When an emergency that cannot be handled independently is detected, an empowerment request is initiated to obtain the temporary command and decision-making authority and / or the supplementary resource and capability interfaces required for parallel and collaborative execution with other secondary decision-making agents at the same level. or, After obtaining the temporary command and decision-making authority and updating the role identifier to a temporary level-one decision-making intelligent agent, organize other intelligent agents to coordinate the handling of the emergency.

3. The trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents according to claim 2, characterized in that, The evaluation agent has a first working mode and a second working mode; The first working mode is an evaluation submodule, which is embedded in the planning agent, the preparation agent and the execution monitoring agent respectively; The second working mode is an independent evaluation agent.

4. The trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents according to claim 3, characterized in that, An evaluation submodule embedded in the planning agent is used to perform a predictive evaluation of the suitability of the plan in the first quantitative evaluation. An evaluation submodule embedded in the prepared agent is used to perform a matching evaluation of the credibility, controllability, and task execution controllability of the resources involved in the decision-making scheme in the first quantitative evaluation. An evaluation submodule embedded in the execution monitoring agent is used to perform the second quantitative evaluation of the impact of the handling of the emergency and its impact on the execution effect of the plan; The independent evaluation agent is used to perform a third quantitative evaluation of the overall effectiveness of the decision-making scheme.

5. The trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents according to claim 2, characterized in that, The planning agent includes a superior planning agent and subordinate planning agents. The steps for generating the decision scheme include: The higher-level planning agent receives the task and issues advance orders to the lower-level planning agent based on the commander's intentions; Multiple contingency plans are generated in parallel by the superior planning agent and the subordinate planning agent. Each contingency plan is simulated and analyzed, and the decision-making scheme is determined according to the commander's instructions.

6. The trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents according to claim 2, characterized in that, The preparation agent includes a superior preparation agent and a subordinate preparation agent. The steps for generating the execution plan include: The superior preparation agent organizes and evaluates resources according to the decision-making scheme, generates a local execution plan, and issues it. The lower-level preparation agent receives the execution plan of this level, generates an adapted branch plan, and sends it to the corresponding secondary decision-making agent.

7. The trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents according to claim 2, characterized in that, The step of dynamically configuring the secondary decision agent that submits the empowerment application includes: The execution monitoring agent evaluates the empowerment application and generates an authorization proposal. Upon receiving confirmation from the commander regarding the authorization proposal, the temporary command and decision-making authority and / or the supplementary resource and capability interfaces are dynamically configured for the secondary decision-making agent that submitted the empowerment application. After configuring the temporary command and decision-making permissions, update its role identifier.

8. The trusted human-machine hybrid intelligent decision-making method based on swarm intelligence agents according to claim 1, characterized in that, The method further includes: Based on an inter-agent communication protocol that incorporates trusted and controllable value information fields, the collaborative state of the agents is monitored and interactively verified. Based on a resource call protocol with added identity authentication and call auditing mechanisms, the trustworthiness of external resources is verified; The external resources include external models, external databases, and external tools.

9. A trusted human-machine hybrid intelligent decision-making system based on swarm intelligence, characterized in that, The system includes: A first-level decision agent is configured to perform the operations performed by the first-level decision agent in the method according to any one of claims 1-8; A secondary decision agent is configured to perform the operations executed by the secondary decision agent in the method according to any one of claims 1-8; The Trusted and Controllable Module is deployed among various intelligent agents and is configured to monitor and interactively verify the collaborative state of intelligent agents based on an inter-agent communication protocol with added trusted and controllable value information fields. The resource management module, deployed between the intelligent agent and external resources, is configured to verify the trustworthiness of external resources based on a resource call protocol with added identity authentication and call auditing mechanisms. The external resources include external models, external databases, and external tools.

10. The trusted human-machine hybrid intelligent decision-making system based on swarm intelligence as described in claim 9, characterized in that, The first-level decision-making agent includes a planning agent, a preparation agent, an execution monitoring agent, and an evaluation agent. The evaluation agent has a first working mode and a second working mode; The first working mode is an evaluation submodule, which is embedded in the planning agent, the preparation agent and the execution monitoring agent respectively; The second working mode is an independent evaluation agent; Among them, sequential collaboration in a ring-shaped cascade is supported between the first-level decision-making agents at the same level, and cross-level vertical linkage collaboration is supported between the first-level decision-making agents at different levels. The secondary decision-making agent is used for: Execute the execution plan issued by the first-level decision-making agent; or, When an emergency that cannot be handled independently is detected, an empowerment request is initiated to obtain the temporary command and decision-making authority and / or the supplementary resource and capability interfaces required for parallel and collaborative execution with other secondary decision-making agents at the same level. or, After obtaining the temporary command and decision-making authority and updating the role identifier to a temporary level-one decision-making intelligent agent, organize other intelligent agents to coordinate the handling of the emergency.

Images