Multi-agent intelligent design optimization method based on knowledge evolution

Abstract

This invention discloses a multi-agent intelligent design optimization method based on knowledge evolution, belonging to the field of multi-agent optimization technology. The method includes: acquiring information of each agent; classifying each agent according to the agent information to obtain agent classification; performing real-time application analysis on the user's agent classification to obtain application optimization requirements; sending the agent classification information and application optimization requirements to the platform; receiving the agent classification information and application optimization requirements of each user in real time; configuring corresponding agent optimization knowledge data for the user according to the agent classification information and application optimization requirements; and optimizing each agent within the user's agent classification according to the agent optimization knowledge data. This invention, by constructing a multi-agent collaborative system, allows each agent to leverage its strengths and achieves knowledge fusion and complementarity through knowledge evolution, greatly improving the ability to handle complex design problems.

Description

Technical Field

[0001] This invention belongs to the field of multi-agent optimization technology, specifically a multi-agent intelligent design optimization method based on knowledge evolution. Background Technology

[0002] In the field of intelligent design optimization, traditional methods often rely on a single agent or centralized architecture to make design decisions through preset rules or static knowledge bases. These methods have significant limitations when facing complex and dynamic design tasks: on the one hand, a single agent, limited by its own knowledge reserves and computing power, struggles to efficiently handle multi-dimensional and highly complex design problems, especially in scenarios involving cross-domain knowledge fusion, where its performance bottleneck is particularly prominent; on the other hand, centralized architectures lack flexibility and scalability, making it difficult to adapt to dynamic changes in design requirements.

[0003] With the deepening development of artificial intelligence technology, multi-agent systems, with their distributed collaboration and parallel processing capabilities, are gradually becoming a key paradigm for solving complex design problems. Multi-agent systems simulate the collaborative behavior of biological groups, enabling multiple agents to optimize overall performance through division of labor, cooperation, and knowledge sharing, based on shared environmental information. However, existing multi-agent systems still have shortcomings in knowledge utilization: most systems rely on static knowledge bases or preset collaboration strategies, lacking adaptability to dynamic environments and task changes; imperfect knowledge transfer mechanisms lead to low efficiency in knowledge sharing among agents, making it difficult to form effective collaborative optimization capabilities.

[0004] Based on this, in order to solve the optimization problem of multi-agent systems, this invention provides a multi-agent intelligent design optimization method based on knowledge evolution. Summary of the Invention

[0005] To address the problems of the above-mentioned solutions, this invention provides a multi-agent intelligent design optimization method based on knowledge evolution.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A knowledge-evolution-based multi-agent intelligent design optimization method, including:

[0008] Acquire information about each intelligent agent, classify each intelligent agent according to the intelligent agent information, and obtain intelligent agent classification; perform real-time application analysis on the user's intelligent agent classification to obtain application optimization requirements; send the intelligent agent classification information and application optimization requirements to the platform.

[0009] It receives intelligent agent classification information and application optimization requirements from each user in real time, configures corresponding intelligent agent optimization knowledge data for the user based on the intelligent agent classification information and application optimization requirements, and optimizes each intelligent agent within the user's intelligent agent classification based on the intelligent agent optimization knowledge data.

[0010] Furthermore, the agents are classified based on their information, including:

[0011] The platform sets the classification criteria for intelligent agents and establishes a classification evaluation model. The classification evaluation model analyzes the information of each pair of intelligent agents and the classification criteria to obtain the classification evaluation results between each intelligent agent. The classification evaluation results include those of the same type and those of different types.

[0012] Agents that are classified and evaluated as belonging to the same category are grouped together and labeled as agent classifications.

[0013] Furthermore, the expression for the classification evaluation model is:

[0014] ;

[0015] In the formula: (A1, A2, BK) are the input data, A1 and A2 are the information of two agents respectively; BK is the agent classification standard; A1⇔A2 means that A1 and A2 meet the agent classification standard; the output data is the classification evaluation value QA(A1, A2, BK), and the classification evaluation value is 1 or 0.

[0016] When the classification evaluation value is 1, the classification evaluation result is that the class belongs to the same category;

[0017] When the classification evaluation value is 0, the classification evaluation result is outlier.

[0018] Furthermore, real-time application analysis is performed on the user's intelligent agent classification, including:

[0019] The system acquires historical analysis records of each agent within the agent classification in real time, performs real-time optimization analysis on the historical analysis records, and obtains the optimization analysis results of the agent classification. The optimization analysis results are either qualified or unqualified, along with the corresponding optimization indicators.

[0020] When the optimization analysis result is qualified, the application optimization requirement is zero.

[0021] When the optimization analysis result is that the analysis is unqualified and the corresponding optimization index is obtained, the individual optimization requirements of the optimization index are generated, and the individual optimization requirements are summarized into application optimization requirements.

[0022] Furthermore, real-time optimization analysis is performed on historical analysis records, including:

[0023] Set up various optimization indicators and their corresponding individual standards; summarize the individual standards corresponding to each optimization indicator into scenario application standards;

[0024] By analyzing the historical analysis records of each intelligent agent through scenario application standards, individual analysis results corresponding to each optimization index are obtained. The individual analysis results include whether the index is qualified or not. Optimization analysis results are generated based on the individual analysis results.

[0025] Furthermore, the historical analysis records of each intelligent agent are analyzed through scenario application standards, including:

[0026] Establish an indicator evaluation model, the expression of which is:

[0027] ;

[0028] In the formula: (s, BZ) i ) represents the input data, and s represents the historical analysis records of the corresponding agent; BZ i This represents the optimization standard for the corresponding optimization index, where i represents the corresponding optimization index, i = 1, 2, ..., n, and n is the number of optimization indexes; s→BZ i This indicates that the corresponding historical analysis records meet the optimization criteria of the corresponding optimization indicators; the output data is the indicator evaluation value ZP(s, BZ). i The indicator evaluation value is 1 or 0;

[0029] By analyzing historical analysis records and scenario application standards through an indicator evaluation model, the indicator evaluation values ​​of the corresponding intelligent agents for each optimization indicator are obtained.

[0030] Summarize and optimize the evaluation values ​​of the indicators for each agent within the agent category;

[0031] When the evaluation value of an indicator is 1, the single-item analysis result of the intelligent agent classification is qualified.

[0032] When no indicator has an evaluation value of 1, the single-item analysis result of the agent classification is that the indicator is unqualified.

[0033] Furthermore, when the single-item analysis result of the agent classification is qualified and the agent has an indicator evaluation value of 0, the agent with an indicator evaluation value of 1 is used to optimize and adjust the agent with an indicator evaluation value of 0.

[0034] Furthermore, methods for setting individual criteria for optimization indicators include:

[0035] The platform acquires the classification of each intelligent agent in real time; it also performs real-time feature collection on each intelligent agent classification according to each optimization index to obtain the feature data corresponding to each intelligent agent classification on each optimization index.

[0036] The feature data of the intelligent agent classification on various optimization indicators are compared in real time to obtain the individual standard of the intelligent agent classification on the optimization indicators.

[0037] Identify the user's agent classification information and configure individual standards for each optimization indicator based on the agent classification information.

[0038] Furthermore, methods for setting individual standards for optimization metrics include: manual setting by users according to their own needs.

[0039] Furthermore, based on the agent classification information and application optimization needs, corresponding agent optimization knowledge data is configured for the user, including:

[0040] Identify agent classification information and application optimization needs, and determine corresponding scenario application standards based on agent classification information and application optimization needs;

[0041] Based on the agent classification information, various candidate agents that meet the application scenario standards are identified from various data sources on the platform.

[0042] Prioritize and filter each candidate intelligent agent to obtain the target intelligent agent. Generate intelligent agent optimization knowledge data based on application optimization requirements and the target intelligent agent, and send the intelligent agent optimization knowledge data to the corresponding user.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] This invention constructs a multi-agent collaborative system, allowing each agent to leverage its strengths and achieve knowledge fusion and complementarity through knowledge evolution. It enables the analysis and optimization of design problems from multiple perspectives, greatly enhancing the ability to handle complex design problems. At the same time, it optimizes through multiple channels based on the platform's rich resource data, ensuring timely and efficient optimization of the multi-agent system and improving its performance. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0047] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0048] like Figure 1 As shown, a multi-agent intelligent design optimization method based on knowledge evolution is proposed, which includes:

[0049] Step 1: Obtain information about each agent, such as capability attributes, knowledge representation, communication mechanisms, and task scenarios, to analyze whether they can be optimized using methods such as knowledge transfer; classify each agent based on the agent information to obtain agent classification; perform real-time application analysis on the user's agent classification to obtain application optimization requirements; send the agent classification information and application optimization requirements to the platform; the agent classification information is composed of information about each agent.

[0050] In one embodiment, the agent information and application optimization requirements corresponding to the agent classification are sent to the platform provider, typically to a platform server established by the platform provider. The platform provider then provides optimization services to each user, and the platform server can be established based on local servers, the cloud, etc.

[0051] In one embodiment, the agents are classified according to their information. This can be done directly using existing methods, such as classifying the same agents in a group across many scenarios.

[0052] In one embodiment, classifying agents based on agent information includes:

[0053] The platform sets the classification criteria for intelligent agents, which are based on the fundamental criteria that enable knowledge transfer and optimization training. These criteria can be applied directly or adjusted according to needs. A classification evaluation model is established to analyze whether different intelligent agents belong to the same classification. The training set is labeled using existing historical classification data of intelligent agents.

[0054] By analyzing the information of each pair of agents and the classification criteria of agents through a classification evaluation model, the classification evaluation results between each agent are obtained. The classification evaluation results include those of the same class and those of different classes.

[0055] Agents that are classified and evaluated as belonging to the same category are grouped together and labeled as agent classifications.

[0056] In one embodiment, the classification evaluation model is built based on existing intelligent algorithms, such as machine learning and deep learning algorithms.

[0057] In one embodiment, the expression for the classification evaluation model is:

[0058] ;

[0059] In the formula: (A1, A2, BK) are the input data, A1 and A2 are the information of two agents respectively; BK is the agent classification standard; A1⇔A2 means that A1 and A2 meet the agent classification standard; the output data is the classification evaluation value QA(A1, A2, BK), and the classification evaluation value is 1 or 0.

[0060] When the classification evaluation value is 1, the classification evaluation result is that the class belongs to the same category;

[0061] When the classification evaluation value is 0, the classification evaluation result is outlier.

[0062] In one embodiment, real-time application analysis of user agent classification includes:

[0063] Real-time acquisition of historical analysis records for each agent within the agent category refers to historical analysis records that can represent the current agent. Real-time optimization analysis is performed on these historical analysis records to determine whether indicators such as task completion rate, response time / latency, accuracy / precision, computing resource utilization, energy consumption, storage utilization, environmental change adaptability, task drift detection, failure rate, and robustness against attacks meet the application requirements of the application scenario. The optimization analysis results are then obtained, which are either qualified or unqualified, along with the corresponding optimization indicators.

[0064] Application optimization needs are determined in real time based on the results of the optimization analysis.

[0065] For example, when the optimization analysis result is qualified, the application optimization requirement is zero;

[0066] When the optimization analysis results in non-compliance and corresponding optimization indicators, individual optimization requirements for each optimization indicator are formed based on the degree of non-compliance. These individual optimization requirements are then summarized into application optimization requirements, which aim to overcome the non-compliance in the analysis. Alternatively, other methods can be used to generate application optimization requirements based on the optimization analysis results.

[0067] In one embodiment, real-time optimization analysis of historical analysis records includes:

[0068] Users can set various optimization metrics and their corresponding individual standards; that is, users can dynamically adjust them according to changes in needs; and the individual standards corresponding to each optimization metric are summarized into scenario application standards.

[0069] By analyzing the historical analysis records of each intelligent agent through scenario application standards, individual analysis results corresponding to each optimization index are obtained. The individual analysis results include whether the index is qualified or not. Optimization analysis results are generated based on the individual analysis results.

[0070] In one embodiment, by analyzing the historical analysis records of each agent through scenario application standards, it is possible to determine whether the agent meets each individual standard based on existing evaluation methods. For example, the analysis features corresponding to meeting and not meeting the individual standards can be pre-statistically counted, and then a matching judgment can be made.

[0071] In one embodiment, the historical analysis records of each agent are analyzed using scenario application standards, including:

[0072] Establish an indicator evaluation model, the expression of which is:

[0073] ;

[0074] In the formula: (s, BZ) i ) represents the input data, and s represents the historical analysis records of the corresponding agent; BZ i This represents the optimization standard for the corresponding optimization index, where i represents the corresponding optimization index, i = 1, 2, ..., n, and n is the number of optimization indexes; s→BZ i This indicates that the corresponding historical analysis records meet the optimization criteria of the corresponding optimization indicators; the output data is the indicator evaluation value ZP(s, BZ). i The indicator evaluation value is 1 or 0; the corresponding training set is marked using the corresponding historical analysis records for training;

[0075] By analyzing historical analysis records and scenario application standards through an indicator evaluation model, the indicator evaluation values ​​of the corresponding intelligent agents for each optimization indicator are obtained.

[0076] Summarize and optimize the evaluation values ​​of the indicators for each agent within the agent category;

[0077] When the evaluation value of an indicator is 1, the single analysis result of the agent classification is qualified. Since there are qualified agents in the agent classification, there is no need to use the resources on the platform for transfer learning. The qualified agents can be directly used to transfer learn other unqualified agents to achieve internal optimization.

[0078] When no indicator has an evaluation value of 1, the single-item analysis result of the agent classification is that the indicator is unqualified.

[0079] In one embodiment, the individual criteria corresponding to each optimization indicator are not set by the user, but by the platform using comprehensive information resources to determine the best analytical effect for the current situation, and then setting the individual criteria; the method for setting the individual criteria for optimization indicators includes:

[0080] The platform determines the various intelligent agent categories based on the existing intelligent agents; and collects features of each intelligent agent category in real time according to each optimization index, that is, collects feature data related to the optimization index to obtain the feature data corresponding to each optimization index.

[0081] The feature data of the intelligent agent classification on various optimization indicators are compared in real time to determine the feature data with the best effect and mark it as the single standard of the intelligent agent classification on the optimization indicator.

[0082] Identify the user's agent classification information and configure individual standards for each optimization indicator based on the agent classification information.

[0083] In one embodiment, feature data with the highest pre-defined proportion of effectiveness can be selected as a single standard, taking into account the impracticality caused by the cost of applying the best effect; therefore, a single standard can be dynamically generated for the user based on the user's situation, i.e., which effect feature data is selected as the single standard.

[0084] Step 2: Receive agent classification information and application optimization requirements from each user in real time, configure corresponding agent optimization knowledge data for the user based on the agent classification information and application optimization requirements, and optimize each agent within the user's agent classification based on the agent optimization knowledge data.

[0085] In one embodiment, corresponding agent optimization knowledge data is configured for the user based on agent classification information and application optimization requirements, including:

[0086] Identify the intelligent agent classification information and application optimization requirements, and determine the corresponding scenario application standards based on the intelligent agent classification information and application optimization requirements. These standards can be set by the platform (the user's scenario application standards determined by the platform in the above embodiments) or set by the user, and can be collected from the user.

[0087] Based on the intelligent agent classification information, intelligent agents that meet the application standards of the scenario are obtained from various data sources and data channels of the platform and marked as candidate intelligent agents, such as intelligent agents of other users with sharing permissions, and other shared intelligent agent channels, etc.

[0088] Prioritize and filter each candidate intelligent agent to obtain the target intelligent agent. Generate intelligent agent optimization knowledge data based on application optimization requirements and the target intelligent agent in a way that will not lead to the leakage of the corresponding information of the target intelligent agent. Send the intelligent agent optimization knowledge data to the corresponding user.

[0089] In one embodiment, intelligent agent optimization knowledge data that will not lead to the leakage of the corresponding information of the target intelligent agent is generated based on the application optimization requirements and the target intelligent agent. This is implemented based on existing technologies, such as removing hidden information or replacing information, or it can be processed based on federated learning.

[0090] In one embodiment, priority screening is performed on each candidate intelligent agent based on existing priority algorithms, such as prioritizing based on optimization effect, cost, or efficiency, or based on a combination of multiple parameters; and intelligent sorting can also be performed directly based on machine learning, deep learning algorithms, etc.

[0091] The above formulas are all numerical calculations after removing dimensions. The formulas are obtained by software simulation based on a large amount of data and are closest to the real situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by simulation based on a large amount of data.

[0092] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A multi-agent intelligent design optimization method based on knowledge evolution, characterized in that, The methods include: Acquire information about each agent, classify each agent based on the information, and obtain agent classification. Real-time application analysis of user's intelligent agent classification to obtain application optimization requirements; Send the intelligent agent classification information and application optimization requirements to the platform. It receives intelligent agent classification information and application optimization requirements from each user in real time, configures corresponding intelligent agent optimization knowledge data for the user based on the intelligent agent classification information and application optimization requirements, and optimizes each intelligent agent within the user's intelligent agent classification based on the intelligent agent optimization knowledge data.

2. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 1, characterized in that, The agents are classified based on their information, including: The platform sets the classification criteria for intelligent agents and establishes a classification evaluation model. The classification evaluation model analyzes the information of each pair of intelligent agents and the classification criteria to obtain the classification evaluation results between each intelligent agent. The classification evaluation results include those of the same type and those of different types. Agents that are classified and evaluated as belonging to the same category are grouped together and labeled as agent classifications.

3. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 2, characterized in that, The expression for the classification evaluation model is: ； In the formula: (A1, A2, BK) are the input data, A1 and A2 are the information of two agents respectively; BK is the agent classification standard; A1⇔A2 means that A1 and A2 meet the agent classification standard; the output data is the classification evaluation value QA(A1, A2, BK), and the classification evaluation value is 1 or 0. When the classification evaluation value is 1, the classification evaluation result is that the class belongs to the same category; When the classification evaluation value is 0, the classification evaluation result is outlier.

4. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 1, characterized in that, Real-time application analysis of user agent classification, including: The system acquires historical analysis records of each agent within the agent classification in real time, performs real-time optimization analysis on the historical analysis records, and obtains the optimization analysis results of the agent classification. The optimization analysis results are either qualified or unqualified, along with the corresponding optimization indicators. When the optimization analysis result is qualified, the application optimization requirement is zero. When the optimization analysis result is that the analysis is unqualified and the corresponding optimization index is obtained, the individual optimization requirements of the optimization index are generated, and the individual optimization requirements are summarized into application optimization requirements.

5. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 4, characterized in that, Real-time optimization analysis of historical records, including: Set up various optimization indicators and their corresponding individual standards; summarize the individual standards corresponding to each optimization indicator into scenario application standards; By analyzing the historical analysis records of each intelligent agent through scenario application standards, individual analysis results corresponding to each optimization index are obtained. The individual analysis results include whether the index is qualified or not. Optimization analysis results are generated based on the individual analysis results.

6. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 5, characterized in that, The historical analysis records of each intelligent agent are analyzed using scenario application standards, including: Establish an indicator evaluation model, the expression of which is: ； In the formula: (s, BZ) i ) represents the input data, and s represents the historical analysis records of the corresponding agent; BZ i This represents the optimization standard for the corresponding optimization index, where i represents the corresponding optimization index, i = 1, 2, ..., n, and n is the number of optimization indexes; s→BZ i This indicates that the corresponding historical analysis records meet the optimization criteria of the corresponding optimization indicators; the output data is the indicator evaluation value ZP(s, BZ). i The indicator evaluation value is 1 or 0; By analyzing historical analysis records and scenario application standards through an indicator evaluation model, the indicator evaluation values ​​of the corresponding intelligent agents for each optimization indicator are obtained. Summarize and optimize the evaluation values ​​of the indicators for each agent within the agent category; When the evaluation value of an indicator is 1, the single-item analysis result of the intelligent agent classification is qualified. When no indicator has an evaluation value of 1, the single-item analysis result of the agent classification is that the indicator is unqualified.

7. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 6, characterized in that, When the single-item analysis result of the agent classification is qualified and the agent's indicator evaluation value is 0, the agent with the indicator evaluation value of 1 is used to optimize and adjust the agent with the indicator evaluation value of 0.

8. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 5, characterized in that, Methods for setting individual criteria for optimization indicators include: The platform acquires the classification of each intelligent agent in real time; it also performs real-time feature collection on each intelligent agent classification according to each optimization index to obtain the feature data corresponding to each intelligent agent classification on each optimization index. The feature data of the intelligent agent classification on various optimization indicators are compared in real time to obtain the individual standard of the intelligent agent classification on the optimization indicators. Identify the user's agent classification information and configure individual standards for each optimization indicator based on the agent classification information.

9. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 5, characterized in that, Methods for setting individual standards for optimization metrics include: manual setting by users according to their own needs.

10. The knowledge evolution-based multi-agent intelligent design optimization method according to claim 5, characterized in that, Based on the agent classification information and application optimization needs, the system configures corresponding agent optimization knowledge data for the user, including: Identify agent classification information and application optimization needs, and determine corresponding scenario application standards based on agent classification information and application optimization needs; Based on the agent classification information, various candidate agents that meet the application scenario standards are identified from various data sources on the platform. Prioritize and filter each candidate intelligent agent to obtain the target intelligent agent. Generate intelligent agent optimization knowledge data based on application optimization requirements and the target intelligent agent, and send the intelligent agent optimization knowledge data to the corresponding user.

Images