Intelligent agent collaborative optimization data center management system based on knowledge graph driving

Abstract

The application discloses an agent collaborative optimization data center management system based on a knowledge graph driver and relates to the technical field of data center management, and specifically comprises the following modules: a modeling and entity management module, a cross-region global scheduling optimization module, an agent game negotiation optimization module, an agent trust management module, a game negotiation conflict identification module, an inference evidence management module and a cross-region consistency verification module; the application establishes a complete knowledge graph model, realizes systematic modeling of resource attributes, constraints, historical decisions and strategy preferences, provides a unified data basis for negotiation among multiple agents, effectively solves suboptimal solution problems caused by information asymmetry, adopts a hierarchical inference mechanism, combines global-region-node three-level inference and multi-round game negotiation, realizes recursive optimization from the global to the local, effectively reduces inference complexity and improves negotiation efficiency.

Description

Technical Field

[0001] This invention relates to the field of data center management technology, and more specifically to a knowledge graph-driven agent-based collaborative optimization data center management system. Background Technology

[0002] With the rapid advancement of information technology, data center management is facing increasingly complex challenges. Traditional data center management systems are mainly based on centralized or simple hierarchical mechanisms, which are ill-suited to handling the complexity and dynamic changes of diverse and heterogeneous data. These systems typically employ fixed resource allocation methods, lacking the ability to dynamically identify and provide real-time feedback on resource demands, resulting in low resource utilization efficiency and difficulty in achieving efficient collaborative scheduling.

[0003] Existing multi-agent collaborative optimization systems lack effective integration and analysis capabilities when processing multi-source heterogeneous data, making it difficult to fully grasp the relationships between different resources and to comprehensively model and optimize complex resource environments. Furthermore, they lack the ability to respond quickly and adapt to dynamic situations such as unforeseen events or resource shortages, making it difficult to guarantee stability and reliability.

[0004] The existing technology has the following shortcomings:

[0005] (1) Suboptimal solutions caused by information asymmetry:

[0006] (2) The agent plays the game based only on local observations, making it difficult to form a globally optimal solution;

[0007] (3) The existing negotiation mechanism suffers from suboptimal solutions and inference complexity due to information asymmetry, which affects the resource scheduling efficiency and system stability of the data center.

[0008] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0009] The purpose of this invention is to provide a knowledge graph-driven intelligent agent collaborative optimization data center management system to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a knowledge graph-driven intelligent agent collaborative optimization data center management system, specifically including the following modules: modeling and entity management module, cross-regional global scheduling optimization module, intelligent agent game negotiation optimization module, intelligent agent trust management module, game negotiation conflict identification module, reasoning evidence management module, and cross-regional consistency verification module;

[0011] Modeling and Entity Management Module: Constructs a knowledge graph by defining resources, constraints, policies, dynamic entities and their temporal modeling, relational reasoning and incremental update mechanisms, to unify the expression, reasoning and evolution of data center resources, monitoring data, scheduling records, energy consumption data and agent collaboration;

[0012] Cross-regional global scheduling optimization module: Divides the data center into layers, performs cross-regional conflict detection, outputs the allocation candidate sorting sequence of the regional layer according to priority, performs optimization and scheduling of the regional layer and global layer, generates an executable instruction set and realizes dynamic closed-loop scheduling;

[0013] Intelligent Agent Game Negotiation Optimization Module: Through multi-agent collaborative multi-round game negotiation, it performs preference extraction, local and global reasoning, evidence chain construction, multi-round negotiation, conflict resolution, and instruction issuance under the guidance of knowledge graph;

[0014] The agent trust management module calculates multi-dimensional single-index scores for each agent and normalizes them into trust scores. It sets upper and lower limit thresholds and trend adaptive mechanisms, and dynamically adjusts the information contribution, evidence chain strength and confidence propagation in reasoning through deweighting, weighting and information access restrictions.

[0015] Game negotiation conflict identification module: Based on the input information package of the multi-round game negotiation process, it extracts conflict features and aligns entities, maps them to the conflict ontology, generates resource, constraint, goal and temporal conflict types and initial strength and weight, constructs and applies a set of conflict resolution strategies, and writes the selected strategies into the conflict resolution record node of the knowledge graph.

[0016] Reasoning Evidence Management Module: By freezing the context of candidate proposals and using subgraph-driven reasoning, it generates reasoning steps in stages, along with evidence citations and changes in trust scores, produces draft candidate decisions and comparison indicators, calculates differences, and marks comparison results.

[0017] Cross-regional consistency verification module: triggers local incremental reasoning and evidence mounting through resource changes, performs parallel reasoning and preliminary negotiation at the cross-regional layer, and converges at the coordination layer to form a global reasoning trajectory and optimization decision.

[0018] As a preferred embodiment of the knowledge graph-driven intelligent agent collaborative optimization data center management system described in this invention, wherein:

[0019] Define knowledge graph entities, specifically including:

[0020] Identify data center resource domains, establish sets of resource entities, and classify them, specifically including:

[0021] Computing resource entities, storage resource entities, network resource entities, cooling resource entities, energy resource entities, and edge resource entities;

[0022] For each type of resource entity, further subclasses and attribute sets are established;

[0023] Identify the constraint domain, establish a set of constraint entities, and group them, specifically including:

[0024] Time-constrained entities, resource-constrained entities, security-constrained entities, and energy-constrained entities;

[0025] Establish a set of policy entities and group them, specifically including:

[0026] Scheduling strategy entity, resource allocation strategy entity, energy consumption optimization strategy entity, conflict recovery strategy entity, agent cooperation strategy entity;

[0027] Lifecycle modeling of policy entities includes:

[0028] Creation, versioning, evaluation, replacement, and rollback status;

[0029] Create a dynamic entity collection and group it, specifically including:

[0030] Real-time monitoring of indicator entities, log entities, event entities, alarm entities, status change entities, and time information entities;

[0031] Perform time-series modeling on dynamic entities, including timestamps, data source identifiers, data source credibility, and version numbers;

[0032] Establish a knowledge graph relation set, define the semantic scope of the relation, monitoring indicator constraints and reverse relations, automatically derive the initial relation edges based on the attributes of all entity sets, and reflect its strength, timeliness and confidence through a relation weighting mechanism;

[0033] Integrate real-time monitoring data, historical scheduling records, and energy consumption data from the data center into a knowledge graph;

[0034] The timestamp and change set of each data change are recorded through a control mechanism.

[0035] The incremental update strategy rewrites or appends to entities and relationships that have changed, and supports rollback for multi-version queries based on timestamps.

[0036] Define a set of intelligent agents, specifically including:

[0037] Global scheduling agent, local resource agent, energy consumption optimization agent, conflict resolution agent;

[0038] Collaboration protocols between agents support optimized task decomposition, resource coordination, and conflict resolution.

[0039] As a preferred embodiment of the knowledge graph-driven intelligent agent collaborative optimization data center management system described in this invention, wherein:

[0040] The data center is divided into n global layers, regional layers, and node layers, and the regional boundaries, resource density, and topology information are stored in the knowledge graph.

[0041] Extract the global KPI set from the knowledge graph, construct a resource allocation candidate set, perform cross-regional conflict detection, and output a conflict warning set;

[0042] Label the causes and priorities of conflicts, and output the allocation candidate sorting sequence of the region layer according to priority based on the optimization task objectives and corresponding constraints.

[0043] Assign a candidate sorting sequence to each region layer, calculate global reachability, and plot the change trajectory;

[0044] Define the reasoning rules of the knowledge graph, and generate early warning improvement suggestions and potential alternative region layers;

[0045] The conflict warnings, priorities, and allocation candidate sorting sequences between regional layers are entered into a time-series version of the knowledge graph;

[0046] Based on resources, the goal of the optimization task, and the corresponding constraints, the local optimization problem of the region layer is defined. By performing region layer inference, the set of preference candidates for the upper region layer and the preference transmission information for the lower-level agents are output.

[0047] Reorder the conflict detection and priority of the resource allocation candidate set, write the regional layer results back to the knowledge graph, and record the timestamp, version number and preference propagation relationship of the current regional layer;

[0048] The final scheduling input set is formed based on the output of the global layer. Execution verification is performed at the node layer to generate the final scheduling instruction set, which specifically includes:

[0049] Resource allocation, migration / startup / shutdown optimization tasks, and energy consumption optimization actions;

[0050] The final scheduling instructions are issued for execution control, the execution status is monitored in real time, the dynamic entities in the knowledge graph are updated, and the knowledge graph is populated.

[0051] Establish query and reasoning interfaces for the knowledge graph, create information flow pipelines between the global layer, regional layer, and node layer, and record change logs for the information flow process.

[0052] As a preferred embodiment of the knowledge graph-driven intelligent agent collaborative optimization data center management system described in this invention, wherein:

[0053] By reading the preference set of the data center through the intelligent agent, the optimization goal, constraints and resource allocation weights are extracted. By searching the subgraph of the knowledge graph, the constraint set and resource status are extracted.

[0054] The initial proposal for resource allocation strategy is generated based on the reasoning results of the knowledge graph. An evidence chain is constructed, and the initial proposal, evidence chain, and expected optimization are packaged into a proposal package. The package is entered into the knowledge graph and labeled with the version number and timestamp. Based on the preference for the upper-level intelligent agent / region layer, the preference transmission information is output.

[0055] It receives requests from other intelligent agents, extracts the evaluation target subgraph, quickly reasones about the local subgraph of the knowledge graph, and generates supporting / opposing conclusions and modification suggestions based on local reasoning rules;

[0056] Calculate the overall support, opposition, and expected optimization of each agent, and write the evaluation results back to the dynamic entities and relationships in the knowledge graph;

[0057] Entering a multi-round negotiation phase, the upper limit of negotiation rounds, the order of actions within each round, and the rules for information exchange are defined. Based on a trust mechanism, the credibility of the evaluation results is assessed, a trust score is calculated, and the following steps are executed in each round:

[0058] The proposal package for the current game phase of the intelligent agent is released and updated;

[0059] During the evaluation phase, the opposing agent responds to the proposal package, outputting support / opposition, modification suggestions, and weight changes;

[0060] Conduct cross-party comparisons and conflict analysis to trigger conflict resolution strategies;

[0061] Update the relevant entities / relationships in the knowledge graph, and record each step of the game negotiation process, including inputs, outputs, and the evolution of trust scores.

[0062] The inputs and outputs of the game negotiation process, the results of each round, the evolution of trust scores, and the final consensus are recorded in the knowledge graph;

[0063] If a final consensus is reached, the final proposal set and execution instruction set will be generated and issued, the entity and version information of the knowledge graph will be updated, and the time sequence version number and timestamp will be recorded.

[0064] If a final consensus is not reached and the maximum number of rounds has been reached, a rollback strategy will be implemented, specifically including:

[0065] Trigger a global / partial rollback to the latest version, record the timing information of the rollback point, and generate a degradation execution plan;

[0066] The entire game negotiation process is summarized into a negotiation report and written into the log node of the knowledge graph;

[0067] Transform the resource allocation and scheduling strategies contained in the final proposal set into a set of execution instructions;

[0068] Send the execution instruction set to the execution control module to enable resource allocation, optimize task migration, and start / stop tasks.

[0069] After execution, the execution results are backfilled, the deviation between the actual KPI and the target KPI of the optimization task is calculated, and the time series version and preference propagation relationship are updated.

[0070] If a deviation from the threshold or a new conflict occurs during execution, the workflow of the adaptive adjustment and optimization task is triggered, returning to the proposal stage to regenerate the proposal and enter a new game negotiation process loop.

[0071] As a preferred embodiment of the knowledge graph-driven intelligent agent collaborative optimization data center management system described in this invention, wherein:

[0072] For each agent, a single indicator score is calculated based on timeliness, degree of compliance with constraints, and contribution to game negotiation. The single indicator score is then normalized to the [0,1] interval, and a trust score is calculated.

[0073] Set upper and lower thresholds for trust scores. When the trust score falls below the lower threshold, a weighting reduction strategy is triggered; when the trust score exceeds the upper threshold, a weighting increase strategy is triggered. Based on trend analysis, if the trust score continues to decline, the weights are automatically adjusted to favor more stable indicators. The trust score corresponding to the information source is determined, specifically including:

[0074] If the trust score is less than the lower threshold, the information weight is multiplied by a weighting factor to reduce the influence of the information in local reasoning and a new information contribution is set; if the trust score is less than the strict threshold, information access restrictions are implemented.

[0075] If the trust score is greater than the upper limit threshold, the information weight is increased to β times the original value, and its priority is increased in the reasoning queue. For high-trust information, the evidence chain integrity strength factor is increased to improve the confidence propagation.

[0076] As a preferred embodiment of the knowledge graph-driven intelligent agent collaborative optimization data center management system described in this invention, wherein:

[0077] It receives input information packets from a multi-round game negotiation process, extracts potential conflict feature sets, performs entity recognition and relationship alignment on the feature sets, and maps them to conflict ontology in a knowledge graph.

[0078] Calculate the initial conflict intensity and importance weight for each type of conflict;

[0079] A comprehensive conflict score is obtained by weighting the conflicts based on historical collaboration, the integrity of the evidence chain, and the credibility of the data source.

[0080] Establish a set of conflict resolution strategies;

[0081] For each conflict type, bind and combine the available set of strategies, specifically including:

[0082] Increase the weight of certain conflict stakeholders and advance their entry into the inference priority queue;

[0083] One party makes concessions within the time window, and the timing of task and resource allocation is rearranged.

[0084] One party makes concessions within the time window, and the timing of task and resource allocation is rearranged.

[0085] Write the selected strategy into the conflict resolution record node of the knowledge graph.

[0086] As a preferred embodiment of the knowledge graph-driven intelligent agent collaborative optimization data center management system described in this invention, wherein:

[0087] Receive candidate proposals and freeze the relevant context, perform reasoning based on the subgraph of the knowledge graph and reasoning rules, and generate a sequence of reasoning steps;

[0088] At each step, a stage-specific evidence reference is generated and the evidence set reference is recorded. The knowledge graph subgraph ID and the direction of the evidence link are updated. The change in trust score is calculated and recorded to form the final reasoning trajectory entry, which is then bound to the metadata of the proposal with the corresponding proposal ID.

[0089] Generate initial drafts of candidate decisions and control indicators, calculate control differences based on the control baseline and fill in indicator difference values, assign control results to labels and compare with actual implementation results;

[0090] The monitoring situation triggers anomaly detection, generates anomaly event records, extracts evidence to locate the scope of impact, initiates the recovery process, and updates the recovery process records.

[0091] As a preferred embodiment of the knowledge graph-driven intelligent agent collaborative optimization data center management system described in this invention, wherein:

[0092] Receive resource change events and their impact scope, identify the set of affected subgraphs, perform local reasoning only for each affected subgraph, produce local reasoning steps and local evidence references, combine local reasoning results, update the trust score changes related to the subgraph, save incremental reasoning entries to the database, bind them to the original proposal number, and map incremental evidence references to the version and time window within the subgraph.

[0093] The optimization task across the region layer is broken down into several parallel subtask packages. Independent reasoning and preliminary negotiation are performed in each region / resource type to generate local reasoning steps, local evidence citations, and local trust score changes.

[0094] The coordination layer converges, executes consistency verification and conflict resolution strategies across regional layers, summarizes the results across regional layers, and forms a global inference trajectory and corresponding optimization decisions.

[0095] On the other hand, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements the steps of the knowledge graph-driven intelligent agent collaborative optimization data center management system as described above.

[0096] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements the steps of the knowledge graph-driven intelligent agent collaborative optimization data center management system as described above.

[0097] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0098] 1. By establishing a complete knowledge graph model, we can achieve systematic modeling of resource attributes, constraints, historical decisions and policy preferences, providing a unified data foundation for negotiation among multiple agents, effectively solving the suboptimal solution problem caused by information asymmetry, and significantly improving the global optimality of resource scheduling.

[0099] 2. A hierarchical reasoning mechanism is adopted, which combines global-regional-node three-level reasoning with multi-round game-like negotiation to achieve recursive optimization from global to local, effectively reducing reasoning complexity, improving negotiation efficiency, and significantly reducing communication overhead.

[0100] 3. The reliability of agent information is quantitatively scored based on a trust assessment mechanism. By dynamically adjusting the weight of information influence, negotiation deviation caused by false information can be effectively prevented.

[0101] 4. Design conflict resolution strategies that cover rollback and fault tolerance, and combine priority sorting, nearest common ground, or cost-utility trade-off strategies to ensure that the system can quickly recover and reach a consistent scheduling scheme under complex scenarios such as dynamic load and conflicts.

[0102] 5. Through optimization strategies such as incremental inference and parallel computing, the system's computing resource consumption and response latency have been significantly reduced, improving the system's processing efficiency and real-time performance, and enabling it to better adapt to the dynamic changes in the data center. Attached Figure Description

[0103] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0104] Figure 1 This is a flowchart of the method for collaborative optimization of a data center management system based on knowledge graph-driven intelligent agents according to the present invention.

[0105] Figure 2 This is a schematic diagram of the modules of the knowledge graph-driven intelligent agent collaborative optimization data center management system of the present invention. Detailed Implementation

[0106] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0107] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention. This embodiment provides a knowledge graph-driven intelligent agent collaborative optimization data center management system, which specifically includes the following modules: modeling and entity management module, cross-regional global scheduling optimization module, intelligent agent game negotiation optimization module, intelligent agent trust management module, game negotiation conflict identification module, reasoning evidence management module, and cross-regional consistency verification module.

[0108] Modeling and Entity Management Module: Constructs a knowledge graph by defining resources, constraints, policies, dynamic entities and their temporal modeling, relational reasoning and incremental update mechanisms, to unify the expression, reasoning and evolution of data center resources, monitoring data, scheduling records, energy consumption data and agent collaboration;

[0109] Define knowledge graph entities, specifically including:

[0110] Identify data center resource domains, establish sets of resource entities, and classify them, specifically including:

[0111] Computing resource entities, storage resource entities, network resource entities, cooling resource entities, energy resource entities, and edge resource entities;

[0112] For each type of resource entity, further subclasses and attribute sets are established;

[0113] Identify the constraint domain, establish a set of constraint entities, and group them, specifically including:

[0114] Time-constrained entities, resource-constrained entities, security-constrained entities, and energy-constrained entities;

[0115] Establish a set of policy entities and group them, specifically including:

[0116] Scheduling strategy entity, resource allocation strategy entity, energy consumption optimization strategy entity, conflict recovery strategy entity, agent cooperation strategy entity;

[0117] Lifecycle modeling of policy entities includes:

[0118] Creation, versioning, evaluation, replacement, and rollback status;

[0119] Create a dynamic entity collection and group it, specifically including:

[0120] Real-time monitoring of indicator entities, log entities, event entities, alarm entities, status change entities, and time information entities;

[0121] Perform time-series modeling on dynamic entities, including timestamps, data source identifiers, data source credibility, and version numbers;

[0122] Establish a knowledge graph relation set, define the semantic scope of the relation, monitoring indicator constraints and reverse relations, automatically derive the initial relation edges based on the attributes of all entity sets, and reflect its strength, timeliness and confidence through a relation weighting mechanism;

[0123] Integrate real-time monitoring data, historical scheduling records, and energy consumption data from the data center into a knowledge graph;

[0124] The timestamp and change set of each data change are recorded through a control mechanism.

[0125] The incremental update strategy rewrites or appends to entities and relationships that have changed, avoiding full reconstruction, and supports rollback for multi-version queries based on timestamps.

[0126] Define a set of intelligent agents, specifically including:

[0127] Global scheduling agent, local resource agent, energy consumption optimization agent, conflict resolution agent;

[0128] Collaboration protocols between agents support optimized task decomposition, resource coordination, and conflict resolution.

[0129] Cross-regional global scheduling optimization module: Divides the data center into layers, performs cross-regional conflict detection, outputs the allocation candidate sorting sequence of the regional layer according to priority, performs optimization and scheduling of the regional layer and global layer, generates an executable instruction set and realizes dynamic closed-loop scheduling;

[0130] The data center is divided into n global layers, regional layers, and node layers, and the regional boundaries, resource density, and topology information are stored in the knowledge graph.

[0131] Extract the global KPI set from the knowledge graph, construct a resource allocation candidate set, perform cross-regional conflict detection, and output a conflict warning set;

[0132] Label the causes and priorities of conflicts, and output the allocation candidate sorting sequence of the region layer according to priority based on the optimization task objectives and corresponding constraints.

[0133] Assign a candidate sorting sequence to each region layer, calculate global reachability, and plot the change trajectory;

[0134] Define the reasoning rules of the knowledge graph, and generate early warning improvement suggestions and potential alternative region layers;

[0135] The conflict warnings, priorities, and allocation candidate sorting sequences between regional layers are entered into a time-series version of the knowledge graph;

[0136] Based on resources, the goal of the optimization task, and the corresponding constraints, the local optimization problem of the region layer is defined. By performing region layer inference, the set of preference candidates for the upper region layer and the preference transmission information for the lower-level agents are output.

[0137] Reorder the conflict detection and priority of the resource allocation candidate set, write the regional layer results back to the knowledge graph, record the timestamp, version number and preference propagation relationship of the current regional layer, and if the current regional layer detects resource shortage or constraint overrun, trigger the regional layer adaptive strategy to update the constraints and resource allocation weights.

[0138] The final scheduling input set is formed based on the output of the global layer. Execution verification is performed at the node layer to generate the final scheduling instruction set, which specifically includes:

[0139] Resource allocation, migration / startup / shutdown optimization tasks, and energy consumption optimization actions;

[0140] The final scheduling instructions are issued for execution control, the execution status is monitored in real time, the dynamic entities in the knowledge graph are updated, and the knowledge graph is populated.

[0141] Establish query and reasoning interfaces for the knowledge graph, create information flow pipelines between the global layer, regional layer, and node layer, and record change logs for the information flow process.

[0142] Intelligent Agent Game Negotiation Optimization Module: Through multi-agent collaborative multi-round game negotiation, it performs preference extraction, local and global reasoning, evidence chain construction, multi-round negotiation, conflict resolution, and instruction issuance under the guidance of knowledge graph;

[0143] By reading the preference set of the data center through the intelligent agent, the optimization goal, constraints and resource allocation weights are extracted. By searching the subgraph of the knowledge graph, the constraint set and resource status are extracted.

[0144] The initial proposal for resource allocation strategy is generated based on the reasoning results of the knowledge graph. An evidence chain is constructed, including optimizing the proposition premise, reasoning rules, data source, confidence level, and evidence source identifier. The initial proposal, evidence chain, and expected optimization are packaged into a proposal package, entered into the knowledge graph and labeled with version number and timestamp. Based on the preference for the upper-level intelligent agent / region layer, the preference transmission information is output.

[0145] It receives requests from other intelligent agents, extracts the evaluation target subgraph, quickly reasones about the local subgraph of the knowledge graph, and generates supporting / opposing conclusions and modification suggestions based on local reasoning rules;

[0146] Calculate the overall support, opposition, and expected optimization of each agent, and write the evaluation results back to the dynamic entities and relationships in the knowledge graph;

[0147] Entering a multi-round negotiation phase, the upper limit of negotiation rounds, the order of actions within each round, and the rules for information exchange are defined. Based on a trust mechanism, the credibility of the evaluation results is assessed, a trust score is calculated, and the following steps are executed in each round:

[0148] The proposal package for the current game phase of the intelligent agent is released and updated;

[0149] During the evaluation phase, the opposing agent responds to the proposal package, outputting support / opposition, modification suggestions, and weight changes;

[0150] Conduct cross-party comparisons and conflict analysis to trigger conflict resolution strategies;

[0151] Update the relevant entities / relationships in the knowledge graph, and record each step of the game negotiation process, including inputs, outputs, and the evolution of trust scores.

[0152] If an irreconcilable conflict arises during the game negotiation process, a conflict alarm will be output and a rollback / degradation strategy will be executed.

[0153] The inputs and outputs of the game negotiation process, the results of each round, the evolution of trust scores, and the final consensus are recorded in the knowledge graph;

[0154] If a final consensus is reached, the final proposal set and execution instruction set will be generated and issued, the entity and version information of the knowledge graph will be updated, and the time sequence version number and timestamp will be recorded.

[0155] If a final consensus is not reached and the maximum number of rounds has been reached, a rollback strategy will be implemented, specifically including:

[0156] Trigger a global / partial rollback to the latest version, record the timing information of the rollback point, and generate a degradation execution plan;

[0157] The entire game negotiation process is summarized into a negotiation report and written into the log node of the knowledge graph;

[0158] Transform the resource allocation and scheduling strategies contained in the final proposal set into a set of execution instructions;

[0159] Send the execution instruction set to the execution control module to enable resource allocation, optimize task migration, and start / stop tasks.

[0160] After execution, the execution results are backfilled, the deviation between the actual KPI and the target KPI of the optimization task is calculated, and the time series version and preference propagation relationship are updated.

[0161] If a deviation from the threshold or a new conflict occurs during execution, the workflow of the adaptive adjustment and optimization task is triggered, returning to the proposal stage to regenerate the proposal and enter a new game negotiation process loop.

[0162] The agent trust management module calculates multi-dimensional single-index scores for each agent and normalizes them into trust scores. It sets upper and lower limit thresholds and trend adaptive mechanisms, and dynamically adjusts the information contribution, evidence chain strength and confidence propagation in reasoning through deweighting, weighting and information access restrictions.

[0163] For each agent, a single indicator score is calculated based on timeliness, degree of compliance with constraints, and contribution to game negotiation. The single indicator score is then normalized to the [0,1] interval, and a trust score is calculated.

[0164] Set upper and lower thresholds for trust scores. When the trust score falls below the lower threshold, a weighting reduction strategy is triggered; when the trust score exceeds the upper threshold, a weighting increase strategy is triggered. Based on trend analysis, if the trust score continues to decline, the weights are automatically adjusted to favor more stable indicators. The trust score corresponding to the information source is determined, specifically including:

[0165] If the trust score is less than the lower threshold, the information weight is multiplied by a weighting factor to reduce the influence of the information in local reasoning and a new information contribution is set.

[0166] If the trust score is less than the strict threshold, information access restrictions are implemented; if the trust score is greater than the upper limit threshold, the information weight is increased to β times the original value, and its priority is increased in the inference queue. For high-trust information, the evidence chain integrity strength factor is increased to improve confidence propagation.

[0167] Game negotiation conflict identification module: Based on the input information package of the multi-round game negotiation process, it extracts conflict features and aligns entities, maps them to the conflict ontology, generates resource, constraint, goal and temporal conflict types and initial strength and weight, constructs and applies a set of conflict resolution strategies, and writes the selected strategies into the conflict resolution record node of the knowledge graph.

[0168] It receives input information packets from a multi-round game negotiation process, extracts potential conflict feature sets, performs entity recognition and relationship alignment on the feature sets, and maps them to conflict ontology in a knowledge graph.

[0169] When multiple parties request the same resource at the same time and the total demand exceeds the resource capacity, a conflict type identifier "resource conflict" and a list of conflicting resources are generated.

[0170] When the constraint sets of the proposals of all parties contain contradictory or incompatible constraint combinations, a conflict type identifier "constraint conflict" and a conflict constraint set are generated.

[0171] When the optimization directions of the objective functions of different parties conflict with each other and cannot be satisfied simultaneously in the current time series, a conflict type identifier "objective conflict" and an objective conflict matrix are generated.

[0172] When scheduling constraints of the same resource / task conflict with each other in different time windows, making the sequence infeasible, a conflict type identifier "time conflict" and a time constraint graph are generated.

[0173] Calculate the initial conflict intensity and importance weight for each type of conflict;

[0174] A comprehensive conflict score is obtained by weighting the conflicts based on historical collaboration, the integrity of the evidence chain, and the credibility of the data source.

[0175] Establish a set of conflict resolution strategies;

[0176] For each conflict type, bind and combine the available set of strategies, specifically including:

[0177] Increase the weight of certain conflict stakeholders and advance their entry into the inference priority queue;

[0178] One party makes concessions within the time window, and the timing of task and resource allocation is rearranged.

[0179] One party makes concessions within the time window, and the timing of task and resource allocation is rearranged.

[0180] Write the selected strategy into the conflict resolution record node of the knowledge graph.

[0181] Reasoning Evidence Management Module: By freezing the context of candidate proposals and using subgraph-driven reasoning, it generates reasoning steps in stages, along with evidence citations and changes in trust scores, produces draft candidate decisions and comparison indicators, calculates differences, and marks comparison results.

[0182] Receive candidate proposals and freeze the relevant context, perform reasoning based on the subgraph of the knowledge graph and reasoning rules, and generate a sequence of reasoning steps;

[0183] At each step, a stage-specific evidence reference is generated and the evidence set reference is recorded. The knowledge graph subgraph ID and the direction of the evidence link are updated. The change in trust score is calculated and recorded to form the final reasoning trajectory entry, which is then bound to the metadata of the proposal with the corresponding proposal ID.

[0184] Generate initial drafts of candidate decisions and control indicators, calculate control differences based on the control baseline and fill in indicator difference values, assign control results to labels and compare with actual implementation results;

[0185] The monitoring situation triggers anomaly detection, generates anomaly event records, extracts evidence to locate the scope of impact, initiates the recovery process, and updates the recovery process records.

[0186] Cross-regional consistency verification module: triggers local incremental reasoning and evidence mounting through resource changes, performs parallel reasoning and preliminary negotiation at the cross-regional layer, and converges at the coordination layer to form a global reasoning trajectory and optimization decision;

[0187] Receive resource change events and their impact scope, identify the set of affected subgraphs, perform local reasoning only for each affected subgraph, produce local reasoning steps and local evidence references, combine local reasoning results, update the trust score changes related to the subgraph, save incremental reasoning entries to the database, bind them to the original proposal number, and map incremental evidence references to the version and time window within the subgraph.

[0188] The optimization task across the region layer is broken down into several parallel subtask packages. Independent reasoning and preliminary negotiation are performed in each region / resource type to generate local reasoning steps, local evidence citations, and local trust score changes.

[0189] The coordination layer converges, executes consistency verification and conflict resolution strategies across regional layers, summarizes the results across regional layers, and forms a global inference trajectory and corresponding optimization decisions.

[0190] By establishing a complete knowledge graph model, we can systematically model resource attributes, constraints, historical decisions, and strategy preferences, providing a unified data foundation for negotiation among multiple agents. This effectively solves the suboptimal solution problem caused by information asymmetry and significantly improves the global optimality of resource scheduling.

[0191] A hierarchical reasoning mechanism is adopted, which combines global-regional-node three-level reasoning with multi-round game-like negotiation to achieve recursive optimization from global to local, effectively reducing reasoning complexity, improving negotiation efficiency, and significantly reducing communication overhead.

[0192] The reliability of agent information is quantitatively scored based on a trust assessment mechanism. By dynamically adjusting the weight of information influence, negotiation deviations caused by false information can be effectively prevented.

[0193] The design covers conflict resolution strategies that cover rollback and fault tolerance, and combines priority ranking, nearest common ground, or cost-utility trade-off strategies to ensure that the system can quickly recover and reach a consistent scheduling scheme under complex scenarios such as dynamic load and conflicts.

[0194] Through optimization strategies such as incremental inference and parallel computing, the system's computing resource consumption and response latency have been significantly reduced, while its processing efficiency and real-time performance have been improved, enabling it to better adapt to the dynamic changes in the data center.

[0195] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An agent collaborative optimization data center management system based on knowledge graph driving, characterized in that, Specifically include the following modules: modeling and entity management module, cross-regional global scheduling optimization module, agent game negotiation optimization module, agent trust management module, game negotiation conflict identification module, reasoning evidence management module and cross-regional consistency verification module; The modeling and entity management module: construct a knowledge graph, define resources, constraints, strategies, dynamic entities and their time sequence modeling, relationship reasoning and incremental update mechanism, unify the expression, reasoning and evolution of data center resources, monitoring data, scheduling records, energy consumption data and agent collaboration; The cross-regional global scheduling optimization module: divides the data center into layers, performs cross-regional layer conflict detection, outputs the allocation candidate sorting sequence of the regional layer according to the priority, performs optimization and scheduling of the regional layer and the global layer, generates executable instruction set and realizes dynamic closed-loop scheduling; The agent game negotiation optimization module: through multi-agent collaboration multi-round game negotiation, preference extraction, local and global reasoning, evidence chain construction, multi-round negotiation, conflict resolution and execution instruction issuing are carried out under the driving of the knowledge graph; The agent trust management module: calculates the multi-dimensional single index score of each agent and normalizes it into trust score, sets upper and lower threshold and trend adaptive mechanism, dynamically adjusts the information contribution, evidence chain strength and confidence propagation in reasoning through weight reduction, weight increase and information access restriction; The game negotiation conflict identification module: according to the input information package of multi-round game negotiation process, the conflict feature extraction and entity alignment are carried out, which is mapped to the conflict ontology, the resource, constraint, target and time sequence conflict type and initial strength and weight are generated, the conflict resolution strategy set is constructed and applied, and the selected strategy is written into the knowledge graph conflict resolution record node; The reasoning evidence management module: through context freezing and subgraph driven reasoning on the candidate proposal, reasoning steps are generated according to the stage with evidence reference and trust score change, candidate decisions and initial draft of contrast indicators are output and difference is calculated, contrast results are marked; The cross-regional consistency verification module: through resource change triggering local incremental reasoning and evidence mounting, cross-regional layer parallel reasoning and preliminary negotiation, coordination layer convergence and global reasoning track and optimization decision formation, specifically including: Only local reasoning is performed on each affected subgraph, local reasoning steps and local evidence references are output, local reasoning results are combined, trust score changes related to the subgraph are updated, incremental reasoning entries are stored in the database, and are bound to the original proposal number, and incremental evidence references are mapped to the version and time window in the subgraph; The optimization task of cross-regional layer is divided into several parallel subtask packages, independent reasoning and preliminary negotiation are performed in each region / resource type, local reasoning steps, local evidence references and local trust score changes are generated; The coordination layer converges, performs cross-regional layer consistency verification and conflict resolution strategy, summarizes the cross-regional layer results, forms the global reasoning track and the corresponding optimization decision. 2.The knowledge graph driven intelligent agent collaborative optimization data center management system of claim 1, wherein: The modeling and entity management module, specifically includes: Time sequence modeling of dynamic entities; Establish a knowledge graph relationship set, define the semantic range of the relationship, monitor the index constraints and reverse relationships, automatically derive the initial relationship edge based on the attributes of all entity sets, and reflect the strength, timeliness and confidence through the relationship weighting mechanism; Access the knowledge graph with real-time monitoring data, historical scheduling records, and energy consumption data from the data center; Record the timestamp and change set of each data change through the control mechanism; Through the incremental update strategy, rewrite or append the changed entities and relationships to avoid full reconstruction, and support rollback based on the timestamp multi-version query; Optimize task decomposition, resource collaboration, and conflict resolution through collaboration protocols between agents. 3.The knowledge graph driven intelligent agent collaborative optimization data center management system of claim 2, wherein: The cross-regional global scheduling optimization module specifically includes: Divide the data center into n global layers, regional layers, and node layers; Extract the global KPI set from the knowledge graph, construct the resource allocation candidate set, and perform cross-regional layer conflict detection to output the conflict warning set; Label the conflict reasons and priorities, and based on the target and corresponding constraints of the optimization task, output the allocation candidate sorting sequence of the regional layer according to the priority; For each regional layer allocation candidate sorting sequence, calculate the global reachability and draw the change trajectory; Record the conflict warning, priority, and allocation candidate sorting sequence between regional layers in the time sequence version of the knowledge graph; Define the local optimization problem of the regional layer based on the resources, optimization task target, and corresponding constraints, and output the preferred candidate set for the upper regional layer and the preferred transmission information for the lower agent by executing regional layer reasoning; Reorder the conflict detection and priority of the resource allocation candidate set, write the regional layer results back to the knowledge graph, record the timestamp, version number, and preference transmission relationship of the current regional layer, and if the current regional layer detects resource shortage or constraint overrun, trigger the regional layer adaptive strategy to update the constraint and resource allocation weight; Based on the global layer output, form the final scheduling input set, perform executability verification at the node layer, and generate the final scheduling instruction set; Execute the final scheduling instructions, monitor the execution status in real time, update the dynamic entities of the knowledge graph, and backfill the knowledge graph; Establish a query interface and reasoning interface for the knowledge graph, and establish an information flow pipeline between the global layer, regional layer, and node layer to record change logs during the information flow process. 4.The knowledge graph driven intelligent agent collaborative optimization data center management system of claim 3, wherein: The agent game negotiation optimization module specifically includes: Agents read the preference set of the data center, extract the optimization target, constraint, and resource allocation weight, retrieve the subgraph of the knowledge graph, and extract the constraint set and resource situation; Based on the reasoning results of the knowledge graph, generate an initial proposal for the resource allocation strategy, build an evidence chain, package the initial proposal, evidence chain, and expected optimization into a proposal package, and record the version number and timestamp in the knowledge graph, and output the preference transmission information based on the preference weighting of the upper agent / region layer; Receive requests from other agents, extract evaluation target subgraphs, quickly reason the local subgraph of the knowledge graph, and based on the local reasoning rules, produce support / opposition conclusions and modification suggestions; Calculate the comprehensive support, opposition, and expected optimization of each agent, and write the evaluation results back to the dynamic entities and relationships of the knowledge graph. Enter the multi-round game negotiation, define the upper limit of negotiation rounds, the order of action within the round and the information exchange rules, based on the trust mechanism, the reliability of the evaluation results is evaluated, and the trust score is calculated. 5.The knowledge graph driven intelligent agent collaborative optimization data center management system of claim 4, wherein: The following steps are performed in each round of the multi-round game negotiation: Propose an update of the proposal package of the current agent's game stage; The evaluation stage of the opposite agent responds to the proposal package, outputs support / opposition, modification suggestions and weight changes; Perform cross-side comparison and conflict analysis, trigger conflict resolution strategies; Update the relevant entities / relationships of the knowledge graph, record the input and output of each step of the game negotiation process and the evolution of the trust score; Record the input and output of the game negotiation process, the results of each round, the evolution of the trust score and the final consensus reached in the knowledge graph; Summarize the entire game negotiation process into a negotiation report and write it into the log node of the knowledge graph; Convert the resource allocation and scheduling strategy contained in the final proposal set into an execution instruction set; Issue the execution instruction set to the execution control module to start resource allocation, optimization task migration, start / stop; After execution is completed, generate an execution result backfill, calculate the deviation between the actual KPI and the target KPI of the optimization task, update the time sequence version and preference transmission relationship. 6.The knowledge graph driven intelligent agent collaborative optimization data center management system of claim 1, wherein: The agent trust management module specifically includes: For each agent, calculate the single-index score based on the timeliness, constraint compliance degree, and game negotiation contribution degree indicators, normalize the single-index score to the [0, 1] interval, and calculate the trust score; Set upper and lower thresholds for the trust score, trigger the weight reduction strategy when the trust score is less than the lower threshold, trigger the weight increase strategy when the trust score is greater than the upper threshold, and based on trend analysis, if the trust score is continuously decreasing, automatically adjust the weight to tilt towards more stable indicators, and determine the trust score corresponding to the information source, which specifically includes: If the trust score is less than the lower threshold, multiply the information weight by the weight reduction coefficient to reduce the influence of the information in local reasoning, and set a new information contribution degree; If the trust score is less than the strict threshold, execute information access restriction; if the trust score is greater than the upper threshold, increase the information weight to β times the original value, and prioritize the position in the reasoning queue, for high-trust information, increase the evidence chain integrity strength factor to improve confidence propagation. 7.The knowledge graph driven intelligent agent collaborative optimization data center management system of claim 1, wherein: The game negotiation conflict identification module specifically includes: Receive the input information package of the multi-round game negotiation process, extract the potential conflict feature set, perform entity recognition and relationship alignment on the feature set, and map it to the conflict ontology in the knowledge graph; Calculate the initial conflict intensity and importance weight for each type of conflict; Based on historical collaboration, evidence chain integrity, and data source credibility, weight the conflict to obtain a comprehensive conflict score; Establish a conflict resolution strategy set; Bind and combine available strategies for each conflict type. 8.The knowledge graph driven intelligent agent collaborative optimization data center management system of claim 1, wherein: The reasoning evidence management module specifically includes: Receive the candidate proposal and freeze the related context, perform reasoning based on the subgraph of the knowledge graph and the reasoning rules, and generate a reasoning step sequence; In each step, generate phase evidence reference and record evidence set reference, update the knowledge graph subgraph ID of the reference and the direction of the evidence link, calculate and record the trust score change, form the final reasoning track entry, and bind it to the corresponding proposal ID in the metadata of the proposal; Generate candidate decision and initial draft of control indicators, calculate control difference based on control baseline and fill in indicator difference value, give control result label and compare with actual execution result; Monitor the situation to trigger anomaly detection, generate anomaly event record, extract evidence reference to locate the impact range, start the recovery process, and update the recovery process record. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the modules of the knowledge graph driven intelligent agent collaborative optimization data center management system in any one of claims 1-8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the modules of the knowledge graph driven intelligent agent collaborative optimization data center management system in any one of claims 1-8.

Images