Planning and decision-making system based on multi-agent interaction

Abstract

The invention discloses a planning and decision-making system based on multi-agent interaction, and the system comprises an agent unit which is used for constructing a plurality of agents, and each agent has independent behavior logic and decision-making capability; the planning unit is used for controlling a plurality of agents to execute an initial planning task, and each agent makes an independent decision according to a preset rule; the identification unit is used for determining an interaction state between the intelligent agents and generating an identification of an initial interaction result, and the identification comprises behavior output of each intelligent agent; the interaction result unit is used for dynamically adjusting interaction rules of the agents under user intervention and determining scores and states of interaction results of the adjusted agents; and the preset condition unit is used for generating a new planning task by the planning unit and repeatedly executing the planning task until a preset termination condition of the system is met when the continuous number of the failed interaction results reaches a preset threshold value or the current agent interaction state is determined to be a final state.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a planning and decision-making system based on multi-agent interaction. Background Art

[0002] In the development of modern digital ecosystems, Multi-Agent Systems (MAS) as a distributed computing model are widely used to simulate complex systems, optimize decision-making processes, and enhance automated control capabilities. Existing multi-agent systems achieve collaboration or competition among agents by defining the behavior rules and interaction methods of the agents to achieve specific system goals. Traditional multi-agent systems usually focus on optimizing the overall system goals and accurately modeling the behavior of agents, while ignoring the participation of human users in the system rules and decision-making processes.

[0003] Although some current interactive multi-agent systems allow users to observe the behavior of agents, their interactivity is limited, often presenting a landscape of "focusing on observation rather than interaction". As an external observer, users can watch how agents interact with each other in the system, but rarely can they directly intervene or modify these interaction rules. In this case, users' understanding and influence of the system remain at the surface, and it is difficult for them to deeply experience the complexity and importance of the system rule setting.

[0004] This limitation brings some obvious drawbacks and challenges:

[0005] In traditional systems, the behavior rules of intelligent agents are often predefined by designers, lacking flexibility and diversity. Users cannot adjust the rules according to actual needs or new discoveries, resulting in insufficient adaptability of the system in complex or dynamic environments, reducing work efficiency. Intelligent agents are also unable to work collaboratively in complex environments.

[0006] Therefore, there is an urgent need for a planning and decision-making system based on multi-agent interaction. Summary of the invention

[0007] The present invention provides a planning and decision-making system based on multi-agent interaction to solve the above-mentioned problems existing in the prior art.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] A planning and decision-making system based on multi-agent interaction, including:

[0010] Agent unit, used to build multiple agents, each of which has independent behavior logic and decision-making capabilities;

[0011] A planning unit, which is used to control multiple agents to perform initial planning tasks, where each agent makes independent decisions based on preset rules;

[0012] An identification unit, used to determine the interaction status between agents and generate an identification of the initial interaction result, which includes the behavior output of each agent;

[0013] An interaction result unit, used to dynamically adjust the interaction rules of the agent under user intervention, and determine the score and status of the adjusted interaction results of the agent, where the interaction results include successful interaction results and failed interaction results;

[0014] The preset condition unit is used to generate a new planning task based on the continuous number of failed interaction results reaching a preset threshold, or when the current agent interaction state is determined to be a final state, and repeatedly execute the planning task until the system preset termination condition is reached.

[0015] The intelligent agent unit includes:

[0016] A building module, used for generating a plurality of initial intelligent agent units, each of which includes an independent behavior logic component and a decision-making ability component;

[0017] A behavior logic parameter module is used to determine the behavior logic parameters of each initial intelligent unit, and the behavior logic parameters are configured based on a preset decision tree algorithm;

[0018] The modification module is used to generate multiple new intelligent units by copying and modifying the behavioral logic parameters of the initial intelligent units to adapt to different environments or task requirements.

[0019] The planning units include:

[0020] An allocation module is used to control multiple agents to perform initial planning tasks and allocate an initial set of rules to each agent from a preset rule library;

[0021] The recording module is used to determine the initial behavior path of each intelligent agent, record the results of independent decision-making of each intelligent agent based on the assigned rule set, and generate an identification of the initial behavior path.

[0022] Among them, determining the initial interaction state between the agents and generating the identification of the initial interaction result includes:

[0023] Obtaining behavioral data of multiple agents in an initial interactive state, the behavioral data including specific behavioral output and reaction time of each agent in the interaction;

[0024] Based on the acquired behavior data, determine the initial interaction state between the agents, record the behavior output of each agent in this state, and generate an identification of the initial interaction result;

[0025] The identification of the initial interaction results includes the behavioral output of each agent and its corresponding reaction time;

[0026] If the behavior output of the agent meets the preset rules, a first score of the agent is generated based on the reaction time and the behavior output;

[0027] If the behavior output of the agent does not conform to the preset rules, the score of the behavior output is determined to be zero;

[0028] Output and record the identification of the initial interaction result.

[0029] Among them, determining the score and status of the adjusted agent interaction result includes:

[0030] Obtain user intervention input and determine the dynamic adjustment content of the current agent interaction rules based on the input;

[0031] Perform dynamic adjustments and generate new agent interaction processes based on the adjusted interaction rules, recording the behavior output and reaction time of each agent;

[0032] According to the new agent interaction process, the adjusted interaction results are determined, and corresponding scores and status identifiers are generated. The interaction results include successful interaction results and failed interaction results.

[0033] Among them, the planning unit generates new planning tasks, including:

[0034] Monitor the behavioral data of the agents during interaction and record the consecutive number of failed interaction results;

[0035] When the number of consecutive failed interaction results reaches a preset threshold, or when the current agent interaction state is determined to be a final state, the planning unit is triggered to generate a new planning task;

[0036] Perform a new planning task, update the agent's behavior strategy and restart the interaction process;

[0037] Repeat the above planning tasks and record new interaction results and behavior data;

[0038] Determine whether the current interaction state meets the termination condition preset by the system. If so, terminate the agent interaction process;

[0039] If the system preset termination condition is not met, the system will continue to monitor and update the planning task according to the new interaction results until the termination condition is met.

[0040] The behavior logic parameters are configured based on the preset decision tree algorithm, including:

[0041] Obtain the basic data set of the initial agent unit, which includes the agent's behavior records in different environments and the corresponding decision outputs;

[0042] Based on the basic data set, the preset decision tree algorithm is trained. The training process includes feature extraction of different behavior records and division of each node of the decision tree according to different features to form a decision tree model containing multiple decision paths.

[0043] According to the trained decision tree model, the initial behavior logic parameters of each intelligent unit are determined, and the initial behavior logic parameters include the initial behavior decision rules of the intelligent unit in a specific environment;

[0044] By inputting the current state information of a specific environment, the decision tree algorithm is used to traverse each node, determine the optimal behavior path, and generate corresponding behavior logic parameters;

[0045] According to the optimal path output by the decision tree algorithm, the behavior logic parameters of the intelligent agent are adjusted to form the final configured behavior logic parameter set.

[0046] Among them, an initial rule set is assigned to each agent from the preset rule base, including:

[0047] Based on multi-level rule settings, determine the initial state and preset tasks of each agent;

[0048] Extracting a first rule set corresponding to the initial state from a preset rule base;

[0049] Assigning a first set of rules to corresponding agents and generating a preliminary planning task based on the first set of rules;

[0050] During the execution of the preliminary planning task, monitoring the task progress of the agent, dynamically adjusting the first rule set based on the monitoring result, and determining the corresponding second rule set;

[0051] Determine whether the rule upgrade condition is met. If so, extract the second rule set corresponding to the current task stage from the preset rule library, assign the second rule set to the corresponding agent, and generate a new planning task;

[0052] Continue to monitor the agent’s task progress based on the new planning task and make further rule set adjustments.

[0053] The identifier for generating the initial interaction result includes:

[0054] Based on the behavioral data, determine the initial interaction state between each agent and record the behavioral output of each agent in this state;

[0055] Get the reaction time corresponding to each agent's behavior output;

[0056] Associating behavioral outputs with corresponding reaction times to generate initial interaction results;

[0057] Extract key behavioral features and time features from the initial interaction results and construct an initial interaction feature set;

[0058] Encoding the initial interaction feature set to generate an identifier of the initial interaction result;

[0059] In the logo, the behavior output and corresponding reaction time of each agent are represented separately and formatted to ensure the consistency and readability of the logo.

[0060] Among them, determining the adjusted interaction results includes:

[0061] Obtain the preset interaction analysis model, input all interaction behaviors into the interaction analysis model, and calculate the interaction performance between each intelligent agent in real time;

[0062] Determine an adjusted interaction state according to the interaction performance, and extract an adjusted interaction result;

[0063] Classify the adjusted interaction results, and mark successful interaction results and failed interaction results respectively;

[0064] Multiple key features corresponding to each interaction result are obtained, and the interaction score is calculated based on the features.

[0065] Compared with the prior art, the present invention has the following advantages:

[0066] A planning and decision-making system based on multi-agent interaction includes: an agent unit, which is used to construct multiple agents, each of which has independent behavior logic and decision-making ability; a planning unit, which is used to control multiple agents to perform initial planning tasks, wherein each agent makes independent decisions according to preset rules; an identification unit, which is used to determine the interaction status between agents and generate an identification of the initial interaction result, which includes the behavior output of each agent; an interaction result unit, which is used to dynamically adjust the interaction rules of the agents under user intervention, determine the score and status of the adjusted agent interaction results, and the interaction results include successful interaction results and failed interaction results; a preset condition unit, which is used to generate a new planning task based on the number of consecutive failed interaction results reaching a preset threshold, or determine that the current agent interaction state is the final state, and repeatedly execute the planning task until the system preset termination condition is reached. It can autonomously adjust the planning and decision-making process, adapt to the changing environment and task requirements, reduce human intervention, and improve work efficiency; it enables multiple agents to work together in a complex environment, simulates the complexity of multi-agent interaction in the real world, and provides support for research and practical applications.

[0067] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the present invention.

[0068] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0070] Figure 1 A structural diagram of a planning and decision-making system based on multi-agent interaction in an embodiment of the present invention;

[0071] Figure 2 is a structural diagram of an intelligent agent unit in an embodiment of the present invention;

[0072] Figure 3 4 is a structural diagram of a planning unit in an embodiment of the present invention. DETAILED DESCRIPTION

[0073] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0074] The embodiment of the present invention provides a planning and decision-making system based on multi-agent interaction, including:

[0075] Agent unit, used to build multiple agents, each of which has independent behavior logic and decision-making capabilities;

[0076] A planning unit, which is used to control multiple agents to perform initial planning tasks, where each agent makes independent decisions based on preset rules;

[0077] An identification unit, used to determine the interaction status between agents and generate an identification of the initial interaction result, which includes the behavior output of each agent;

[0078] An interaction result unit, used to dynamically adjust the interaction rules of the agent under user intervention, and determine the score and status of the adjusted interaction results of the agent, where the interaction results include successful interaction results and failed interaction results;

[0079] The preset condition unit is used to generate a new planning task based on the continuous number of failed interaction results reaching a preset threshold, or when the current agent interaction state is determined to be a final state, and repeatedly execute the planning task until the system preset termination condition is reached.

[0080] The working principle of the above technical solution is as follows: the system first constructs multiple intelligent agents as intelligent agent units. Each intelligent agent has independent behavioral logic and decision-making capabilities, which means that each intelligent agent can make decisions and take actions independently based on the environmental information it perceives and internal decision-making rules. Through the independent decision-making and actions of multiple intelligent agents, the system can simulate diverse behaviors and reaction patterns in complex environments and provide rich data and results.

[0081] Planning unit, the system formulates initial planning tasks for multiple agents. Each agent makes independent decisions and behaviors according to pre-set rules. For example, one agent is responsible for path planning and another is responsible for resource allocation. By setting the initial planning tasks, the system can ensure that each agent has a clear goal and direction, and can demonstrate its own capabilities in an uncertain environment.

[0082] Identification unit: During the process of the agents performing tasks, the system determines the interaction status between them through identification units and generates identification of each agent's behavior output, including the execution status of the action and interaction feedback. The role of the identification unit is to provide a clear reference point for subsequent system analysis, ensuring that the behavior and decision-making path of each agent can be clearly tracked in complex interactions.

[0083] In the interaction result unit, users can intervene and dynamically adjust the interaction rules of the intelligent agent, thereby generating new interaction results and determining the scores and status of these results. The system will judge whether the interaction is successful or failed based on these scores. This unit makes the system adaptive and flexible, and can adjust the behavior of the intelligent agent under user intervention to adapt to the changing environment or target requirements.

[0084] Preset condition unit: when the system detects that the number of consecutive failed interaction results reaches a preset threshold, or the current agent interaction state is determined to be the final state, the planning unit will generate a new planning task and the system will enter a repeated execution mode until the termination condition is met; this unit ensures that the system can automatically adjust and optimize itself until the preset goal is achieved or the system believes that there is no more room for optimization. This mechanism enhances the robustness and autonomy of the system.

[0085] The beneficial effects of the above technical solution are: the system can autonomously adjust the planning and decision-making process to adapt to the ever-changing environment and task requirements, reduce human intervention, and improve work efficiency; by generating new planning tasks and repeating them when failure occurs, the system can self-optimize, avoid falling into a cycle of failure, and improve the success rate of tasks; the system design enables multiple agents to work together in a complex environment, simulating the complexity of multi-agent interactions in the real world, and providing support for research and practical applications; the system allows users to intervene and adjust the agent interaction rules during operation, so that the system can adapt to specific needs or changing environments; this system design can be expanded to more application scenarios by adding or modifying agent units and interaction rules, such as drone formations, autonomous driving, intelligent manufacturing, etc.

[0086] In another embodiment, the agent unit includes:

[0087] A building module, used for generating a plurality of initial intelligent agent units, each of which includes an independent behavior logic component and a decision-making ability component;

[0088] A behavior logic parameter module is used to determine the behavior logic parameters of each initial intelligent unit, and the behavior logic parameters are configured based on a preset decision tree algorithm;

[0089] The modification module is used to generate multiple new intelligent units by copying and modifying the behavioral logic parameters of the initial intelligent units to adapt to different environments or task requirements.

[0090] The working principle of the above technical solution is as follows: the building module is used to generate multiple initial intelligent agent units, each of which is composed of two core components: the behavior logic component and the decision-making ability component. The behavior logic component is responsible for defining the behavior pattern and response mechanism of the intelligent agent, that is, how the intelligent agent should act under specific conditions; the decision-making ability component provides the intelligent agent with decision-making capabilities, enabling it to make decisions autonomously based on environmental data and internal rules. By designing the behavior logic and decision-making capabilities separately, the system can customize and adjust the functions of the intelligent agent more flexibly. This modular design allows the intelligent agent to operate independently and adapt to changing environments and task requirements.

[0091] After generating the agent unit, the behavior logic parameter module determines the behavior logic parameters of each agent. These parameters are configured through a preset decision tree algorithm. The decision tree algorithm guides the agent on how to choose appropriate actions through a series of conditions and decision nodes. The use of the decision tree algorithm can effectively decompose the complex decision-making process into a series of simple and manageable steps. In this way, the agent can respond quickly to environmental changes and make the best decision.

[0092] The function of the modification module is to generate multiple new intelligent agent units. Specifically, it copies the initial intelligent agent unit and modifies its behavioral logic parameters so that it can adapt to different environments or task requirements. By copying and modifying the initial intelligent agent unit, the system can quickly expand and adapt to new task requirements. This method not only retains the basic functions of the initial intelligent agent, but also enhances the adaptability of the intelligent agent in different scenarios through parameter adjustment, avoiding the design of each intelligent agent from scratch, saving time and resources.

[0093] The beneficial effects of the above technical solution are as follows: since the behavior logic and decision-making ability of the agent are modular and configurable, the system can easily adapt to different task requirements. By modifying the behavior logic parameters, the system can quickly generate agents that adapt to new environments; the use of decision tree algorithms simplifies complex decision-making processes, allowing agents to respond quickly and take actions. This improves the real-time and effectiveness of the system in a dynamic environment; the modification module copies and adjusts the initial agent unit instead of creating a new agent from scratch, which greatly reduces the consumption of development time and computing resources, especially when large-scale deployment of agents is required. This method is more economical and efficient; by copying and modifying the agent unit, the system can quickly generate a variety of agents with different behavior logics, so that it has higher robustness and adaptability when dealing with different tasks. Even in complex and unpredictable environments, the system can maintain efficient operation by adjusting the behavior logic parameters; by adjusting the behavior logic parameters, the system can generate agents with different behavior characteristics to meet different user needs and task scenarios. This enables the system to not only cope with standardized tasks, but also to adapt to special scenarios and requirements in a customized manner. Through the collaborative work of building modules, behavioral logic parameter modules and modification modules, the system can efficiently and flexibly generate and adjust intelligent agents, ensuring that they can adapt to changing environments and complex task requirements while maintaining high efficiency and economy.

[0094] In another embodiment, the planning unit comprises:

[0095] An allocation module is used to control multiple agents to perform initial planning tasks and allocate an initial set of rules to each agent from a preset rule library;

[0096] The recording module is used to determine the initial behavior path of each intelligent agent, record the results of independent decision-making of each intelligent agent based on the assigned rule set, and generate an identification of the initial behavior path.

[0097] The working principle of the above technical solution is as follows: the allocation module selects and allocates a set of initial rules for each agent from a preset rule base, matches and allocates the rules according to the task requirements and the characteristics of the agent, and ensures that each agent obtains a suitable initial set of rules to start executing the task; analyzes the task requirements, determines the required rule types, evaluates the capabilities and characteristics of each agent, filters out suitable rules from the rule base, and allocates the filtered rules to the corresponding agents, and records the allocation results to ensure traceability.

[0098] The recording module tracks and records the decision-making process of each agent based on the allocation rules, generates the initial behavior path of the agent, creates a unique identifier for it, and saves the decision results and behavior path for subsequent analysis and optimization.

[0099] The beneficial effects of the above technical solution are: providing initial behavioral guidance for intelligent agents and accelerating task initiation; achieving division of labor and cooperation among intelligent agents and improving overall efficiency; avoiding conflicts and ensuring system stability through reasonable allocation of rules, providing traceability of the decision-making process, and facilitating analysis and improvement.

[0100] In another embodiment, determining the initial interaction state between agents and generating an identifier of the initial interaction result includes:

[0101] Obtaining behavioral data of multiple agents in an initial interactive state, the behavioral data including specific behavioral output and reaction time of each agent in the interaction;

[0102] Based on the acquired behavior data, determine the initial interaction state between the agents, record the behavior output of each agent in this state, and generate an identification of the initial interaction result;

[0103] The identification of the initial interaction results includes the behavioral output of each agent and its corresponding reaction time;

[0104] If the behavior output of the agent meets the preset rules, a first score of the agent is generated based on the reaction time and the behavior output;

[0105] If the behavior output of the agent does not conform to the preset rules, the score of the behavior output is determined to be zero;

[0106] Output and record the identification of the initial interaction result.

[0107] The working principle of the above technical solution is: when the intelligent agents start to interact, the system will record the behavioral data of each intelligent agent in real time, including the specific behavioral output and reaction time of each intelligent agent in the interaction.

[0108] Suppose two agents A and B interact in a simulated environment, and the system records the following data:

[0109] Behavior output: A selected the action "move to the target point" and B selected the action "attack".

[0110] Reaction time: A's behavioral decision took 2 seconds, and B's behavioral decision took 3 seconds.

[0111] Recording behavioral output and reaction time can provide basic data for subsequent analysis of the agent's decision-making ability and behavioral patterns. These data reflect the agent's response method and efficiency in a specific environment and are an important basis for evaluating the agent's performance.

[0112] Based on the acquired behavioral data, the system determines the interaction status between agents and generates an identification that includes the behavioral output of each agent and the corresponding reaction time.

[0113] The initial interactive state can be expressed as follows:

[0114] Initial interaction flag = {“Agent A”: {“Behavior output”: “Move to target point”, “Reaction time”: 2 seconds}, “Agent B”: {“Behavior output”: “Attack”, “Reaction time”: 3 seconds}}

[0115] The purpose of generating labels is to abstract the complex interactive process so that the system can more easily perform subsequent analysis and comparison. As a simplified representation, the label can be used for retrieval, classification and further rule judgment.

[0116] Based on the behavioral output and reaction time of each agent, the system evaluates its performance. If the behavior complies with the preset rules, a first score is calculated based on the effect of the behavioral output and the reaction time.

[0117] Assume that the rule setting is: the expected effect score of the behavior is E, and the weight of the reaction time is W. Then the agent score S can be expressed as:

[0118] S = E-(W × reaction time)

[0119] If E=100,W=10:

[0120] A's score S = 100 - (10 × 2) = 80

[0121] B's score S = 100 - (10 × 3) = 70

[0122] By introducing a scoring system, the performance of each agent can be quantified and compared. The score reflects the efficiency and accuracy of the agent, and can unify the standards in multiple interaction scenarios, thereby measuring the pros and cons of the agents.

[0123] If the agent's behavioral output does not conform to the preset rules (for example, it makes a wrong decision or behaves irrationally), its score is zero.

[0124] Assuming that agent A chooses a behavior that does not conform to the preset rules (such as choosing to attack when it should not attack), its score is zero, no matter how fast its reaction time is.

[0125] By setting the score of behaviors that do not conform to the preset rules to zero, the system can effectively screen out poorly performing agents. This mechanism encourages agents to follow the rules and get high scores through reasonable behaviors, thereby driving agents to optimize their decision-making process.

[0126] Finally, the system will output and record the generated identification of the initial interaction results as historical data of the agent interaction and the basis for subsequent analysis.

[0127] Recording this data helps analyze the performance of the agent in different scenarios and helps researchers or system developers understand the decision-making process and behavior patterns of the agent. The accumulation of historical data can also be used for the learning and optimization of the agent to further improve its performance.

[0128] The beneficial effects of the above technical solution are as follows: through the recording and scoring mechanism, the system can quantitatively analyze the performance of the intelligent agent, which helps to objectively evaluate the advantages and disadvantages of different intelligent agents and provide a basis for optimization; the instant feedback of the intelligent agent's score allows the system or developer to quickly discover problems and make corresponding adjustments, thereby continuously improving the performance of the intelligent agent; by setting the score of non-compliant behavior to zero, the system can force the intelligent agent to follow the preset rules and ensure that its behavior is within a reasonable range, thereby improving the reliability and controllability of the intelligent agent; recording and analyzing initial interaction identifiers helps to identify patterns and trends in the behavior of the intelligent agent, and these data can serve as the basis for subsequent machine learning or intelligent agent optimization, promoting the continuous learning and improvement of the intelligent agent in a complex environment; through detailed behavioral data recording and analysis, the system can better understand the weaknesses of the intelligent agent and formulate more effective improvement measures to make the intelligent agent perform more robustly in future tasks.

[0129] In another embodiment, determining the score and status of the adjusted agent interaction result includes:

[0130] Obtain user intervention input and determine the dynamic adjustment content of the current agent interaction rules based on the input;

[0131] Perform dynamic adjustments and generate new agent interaction processes based on the adjusted interaction rules, recording the behavior output and reaction time of each agent;

[0132] According to the new agent interaction process, the adjusted interaction results are determined, and corresponding scores and status identifiers are generated. The interaction results include successful interaction results and failed interaction results.

[0133] The working principle of the above technical solution is: the purpose of user intervention input is to correct or optimize the behavior of the intelligent agent to make it more in line with user needs. In this step, the system will parse the user input and extract key information, such as the requirements for specific behavior adjustments. Suppose the user inputs: "The response of the intelligent agent is a bit slow, can it respond faster?"

[0134] Based on user feedback, the system will dynamically adjust its rules. For example, in response to the above feedback, the system can speed up the response by reducing complex calculations or adopting a simpler algorithm. This adjustment may involve adjusting the system's priorities or optimizing the internal data processing process. Dynamic adjustment means that the agent's behavior rules are not fixed, but can be changed in real time based on user input. This process involves modifying the threshold of the response time or optimizing the model reasoning path. For example, reducing some redundant calculation steps.

[0135] The system reduces its default response time from 2 seconds to 1 second by streamlining the computational process, prioritizing core issues and reducing unnecessary background processing tasks.

[0136] The agent interacts with the user according to the new rules. Based on the adjusted rules, the system regenerates the dialogue flow. The system monitors each step of the interaction and records the agent's behavioral output and reaction time. The core of this stage is that the system uses new rules to generate interactive content and evaluates the effectiveness of this content in real time. This involves using machine learning models to re-predict user needs or applying new algorithms to generate content. When the adjusted system receives a new user question, it can respond within 1 second and provide answers that meet user needs.

[0137] The system evaluates the success or failure based on the results of the interaction and generates a corresponding score and status identification. This usually involves a set of predefined indicators, such as response time, user satisfaction, task completion, etc. The system quantifies the effect of the interaction through a set of evaluation indicators and assigns a score to it. For example, if the agent can provide an accurate answer within the specified time, the system will mark the interaction as "successful" and give it a high score. On the contrary, if the interaction is not ideal, it is marked as "failed" or "partially successful". After the interaction is completed, the system evaluation found that the response time was reduced to less than 1 second, and the user was satisfied with the accuracy of the answer, so the system generated a high score and marked it as "successful".

[0138] The beneficial effects of the above technical solution are: dynamic adjustment enables the system to be closer to user needs, thereby providing more personalized services; by optimizing algorithms and adjusting rules, the system's response time and processing capabilities are significantly improved; by recording and evaluating each interaction, the system can learn which adjustments are effective, thereby continuously improving future interactive performance; this dynamic adjustment process based on user input can ensure that the system continues to optimize its performance under the guidance of user feedback, thereby improving the overall user experience.

[0139] In another embodiment, the planning unit generates a new planning task, including:

[0140] Monitor the behavioral data of the agents during interaction and record the consecutive number of failed interaction results;

[0141] When the number of consecutive failed interaction results reaches a preset threshold, or when the current agent interaction state is determined to be a final state, the planning unit is triggered to generate a new planning task;

[0142] Perform a new planning task, update the agent's behavior strategy and restart the interaction process;

[0143] Repeat the above planning tasks and record new interaction results and behavior data;

[0144] Determine whether the current interaction state meets the termination condition preset by the system. If so, terminate the agent interaction process;

[0145] If the system preset termination condition is not met, the system will continue to monitor and update the planning task according to the new interaction results until the termination condition is met.

[0146] The working principle of the above technical solution is: during the interaction of agents, the system continuously monitors and records the behavior data of each agent, paying special attention to failed interaction results. Failed interaction results refer to the failure of the agent to complete the preset task or make an incorrect decision. The system counts the number of consecutive failure results. Suppose in a task, the goal of agent A is to move to a specific location, but due to path planning errors, it fails to reach the target location three times in a row. The system records these failed results and updates its consecutive failure count to 3.

[0147] Continuous failures may indicate that there is a problem with the current behavior strategy or task planning and that adjustments are needed. Recording the number of consecutive failures helps identify the moment when intervention is needed, thereby preventing the agent from falling into an ineffective loop.

[0148] When the system detects that the number of consecutive failures reaches a preset threshold, or the current interaction state is determined to be an invalid or unsolvable final state, the system triggers the planning unit to generate a new planning task to update the agent's behavior strategy. Assuming that the system's preset threshold is 3 consecutive failures, when agent A reaches this threshold, the system will start the planning unit, re-evaluate the task, and adjust the path planning algorithm or strategy to help the agent complete the task. Triggering a new planning task can break the existing deadlock and find a more effective solution through re-planning. This mechanism ensures the system's dynamic adjustment ability, enabling the agent to adapt and optimize in a constantly changing environment.

[0149] The system updates the agent's behavior strategy based on the new planning task and allows the agent to restart the interaction process. The planning unit will give a new path planning solution, and agent A will adjust its action strategy accordingly and try to reach the target location again. Restarting the interaction process allows the agent to try to solve the problem under the guidance of the new strategy, which increases the possibility of success and avoids repeated mistakes.

[0150] If the new planning task fails to successfully solve the problem, the system will monitor the new interaction results again, record the relevant behavior data, and continue planning and adjustment until it succeeds or meets certain termination conditions. If the new path planning still fails, the system will continue to adjust and introduce more complex planning algorithms, such as machine learning models, to adapt to more complex environments. Through continuous adjustment and optimization, the system can gradually approach the optimal solution, enhance the performance of the intelligent agent, and enable it to gradually succeed in complex environments.

[0151] After each planning and execution, the system will check whether the current interaction status meets the preset termination conditions, which include successful completion of the task, reaching the time limit, or multiple consecutive failures. The system sets Agent A to reach the target position after 10 attempts, otherwise the task will be terminated. If the task is successful or the upper limit of the number of attempts is reached, the system will decide whether to terminate. Setting termination conditions can prevent the system from falling into an infinite loop or wasting resources, thereby ensuring the efficiency and rationality of the system operation.

[0152] If the termination condition is not met, the system will continue to monitor the agent's interaction process and adjust and optimize according to the new interaction results until the termination condition is met. If agent A successfully reaches the target position on the 7th attempt, the system will consider the task completed and terminate the subsequent interaction process. Continuous monitoring and dynamic adjustment enable the system to adapt to changes and ultimately achieve the task goals. This cyclic process ensures that the agent gradually optimizes its strategy and ultimately meets the task requirements.

[0153] The beneficial effects of the above technical solution are: through continuous monitoring and adjustment, the system can help the intelligent agent gradually optimize its behavior strategy and enhance its adaptability in complex environments; setting termination conditions and trigger mechanisms can effectively prevent the system from falling into an invalid cycle, thereby preventing waste of resources and improving the efficiency of task execution; through continuous adjustment and updating of planning units, the system can cope with the ever-changing environment and task requirements, ensuring that the intelligent agent always moves towards the goal; the system can intervene in time and adjust the strategy when the intelligent agent fails continuously, which improves the robustness of the intelligent agent in the face of various challenges; recording failure data and new interaction results provides a valuable data basis for the continuous learning and improvement of the intelligent agent, which helps it perform better in future tasks.

[0154] In another embodiment, the behavior logic parameters are configured based on a preset decision tree algorithm, including:

[0155] Obtain the basic data set of the initial agent unit, which includes the agent's behavior records in different environments and the corresponding decision outputs;

[0156] Based on the basic data set, the preset decision tree algorithm is trained. The training process includes feature extraction of different behavior records and division of each node of the decision tree according to different features to form a decision tree model containing multiple decision paths.

[0157] According to the trained decision tree model, the initial behavior logic parameters of each intelligent unit are determined, and the initial behavior logic parameters include the initial behavior decision rules of the intelligent unit in a specific environment;

[0158] By inputting the current state information of a specific environment, the decision tree algorithm is used to traverse each node, determine the optimal behavior path, and generate corresponding behavior logic parameters;

[0159] According to the optimal path output by the decision tree algorithm, the behavior logic parameters of the intelligent agent are adjusted to form the final configured behavior logic parameter set.

[0160] The working principle of the above technical solution is: obtain the basic data set of the initial intelligent agent unit, the basic data set includes the behavior records of the intelligent agent in different environments and the corresponding decision outputs;

[0161] Based on the basic data set, the preset decision tree algorithm is trained. The training process includes feature extraction of different behavior records and division of each node of the decision tree according to different features to form a decision tree model containing multiple decision paths.

[0162] According to the trained decision tree model, the initial behavior logic parameters of each intelligent unit are determined, and the initial behavior logic parameters include the initial behavior decision rules of the intelligent unit in a specific environment;

[0163] By inputting the current state information of a specific environment, the decision tree algorithm is used to traverse each node, determine the optimal behavior path, and generate corresponding behavior logic parameters;

[0164] According to the optimal path output by the decision tree algorithm, the behavior logic parameters of the intelligent agent are adjusted to form the final configured behavior logic parameter set.

[0165] Use the decision tree algorithm to traverse each node and determine the optimal behavior path, including:

[0166] Obtain current state information of a specific environment, the current state information including multiple environment state parameters;

[0167] Preprocess the current state information according to the environmental state parameters to obtain state characteristic data that can be used for decision-making;

[0168] Using state feature data as input, a decision tree algorithm model is constructed, where each node of the decision tree represents a behavior decision point, and the branch path represents the behavior choice;

[0169] Traverse each node of the decision tree and obtain the behavior selection parameters corresponding to each node;

[0170] Compare each behavior selection parameter, determine the behavior results under different paths, and generate the behavior logic parameters of each path;

[0171] According to the preset optimization goal, the behavior logic parameters generated during the traversal process are sorted, and the optimal path is selected to obtain the final optimal behavior path;

[0172] According to the optimal behavior path, the corresponding behavior logic parameters are output to complete the generation of the optimal behavior path.

[0173] For example, in an intelligent agricultural system, it is necessary to select the best irrigation strategy for a certain crop. The decision tree algorithm is used to determine the optimal irrigation strategy (such as drip irrigation, sprinkler irrigation, or no irrigation) based on the current environmental conditions (such as soil moisture, temperature, rainfall forecast, etc.).

[0174] Specific steps:

[0175] Get the current environment status information:

[0176] Current soil moisture: 45%

[0177] Current temperature: 28℃

[0178] Rainfall forecast in the next 12 hours: 20%

[0179] Preprocessing status information:

[0180] Soil moisture, temperature, rainfall forecast, etc. are converted into standardized state feature data. For example, soil moisture of 45% is classified as "moderate humidity", temperature of 28°C is classified as "suitable temperature", and rainfall forecast of 20% is classified as "low rainfall probability".

[0181] Build a decision tree model:

[0182] The root node of the decision tree is "Soil Moisture". Depending on the moisture level, the tree will branch into different paths, for example:

[0183] If humidity is above 60%, do not irrigate;

[0184] If the humidity is between 40% and 60%, consider the air temperature;

[0185] If humidity is below 40%, irrigate.

[0186] For the temperature node, possible paths are:

[0187] If the temperature is above 30°C, sprinkler irrigation is preferred;

[0188] If the temperature is between 20℃ and 30℃, drip irrigation or no irrigation is selected according to the rainfall forecast.

[0189] Traverse the decision tree nodes:

[0190] Based on the current environmental conditions (45% humidity, 28°C temperature, 20% rainfall forecast), the decision tree first branches at the "soil moisture" node and enters the "humidity between 40% and 60%" path.

[0191] Then branch on the "Temperature" node and select the "Temperature between 20℃ and 30℃" path.

[0192] Finally, based on rainfall forecasts, drip irrigation strategies are further selected.

[0193] Generate behavior logic parameters:

[0194] In each path branch, corresponding behavior logic parameters are generated. For example, for the drip irrigation strategy, the logic parameters include the amount of drip irrigation water, time, frequency, etc.

[0195] Assume that the logical parameters generated by the drip irrigation strategy are: 500 ml of drip irrigation per day, each lasting 30 minutes, at 6 am.

[0196] Sort paths by optimization goal:

[0197] Assuming the optimization goal is to maximize the water use efficiency of crops while minimizing water consumption, the system will score and rank the generated behavioral logic parameters.

[0198] If another path (such as sprinkler irrigation) scores higher (for example, sprinkler irrigation may waste more water), the drip irrigation path will be preferred.

[0199] Select the optimal path and output the behavior logic parameters:

[0200] Based on the scoring results, the system finally selects the drip irrigation strategy and outputs the corresponding behavioral logic parameters: 500 ml of drip irrigation per day, each lasting 30 minutes, at 6 am.

[0201] These parameters will be used to control agricultural equipment and perform optimal irrigation behavior.

[0202] The beneficial effects of the above technical solution are: based on a large amount of real data training, the decision-making is more in line with the actual situation; the behavior parameters can be continuously adjusted according to the new environmental status to improve adaptability; the decision tree model has good interpretability, which is convenient for understanding and debugging the decision-making process of the intelligent agent; the decision tree algorithm has fast calculation speed and is suitable for real-time decision-making.

[0203] In another embodiment, an initial set of rules is assigned to each agent from a preset rule base, including:

[0204] Based on multi-level rule settings, determine the initial state and preset tasks of each agent;

[0205] Extracting a first rule set corresponding to the initial state from a preset rule base;

[0206] Assigning a first set of rules to corresponding agents and generating a preliminary planning task based on the first set of rules;

[0207] During the execution of the preliminary planning task, monitoring the task progress of the agent, dynamically adjusting the first rule set based on the monitoring result, and determining the corresponding second rule set;

[0208] Determine whether the rule upgrade condition is met. If so, extract the second rule set corresponding to the current task stage from the preset rule library, assign the second rule set to the corresponding agent, and generate a new planning task;

[0209] Continue to monitor the agent’s task progress based on the new planning task and make further rule set adjustments.

[0210] The working principle of the above technical solution is: set initial conditions and goals for each agent; for example, in a strategy game, an agent is set as a "farmer", with the initial position in the center of the village and the preset task being "collecting resources".

[0211] According to the initial state of the agent, appropriate rules are selected from the preset rule base; for the "farmer" agent, the first rule set includes: if there are trees nearby, cut them down; if the backpack is full, return to the center of the village; if encountering an enemy, run away.

[0212] Apply the selected rules to the agent and develop a preliminary action plan based on these rules.

[0213] The beneficial effects of the above technical solution are: the intelligent agent can dynamically adjust its behavior according to changes in the environment and tasks, thereby improving the flexibility and robustness of the system; through multi-level rule setting, the intelligent agent can gradually master more complex tasks, simulating the process of learning and growth; through real-time monitoring and adjustment, the behavior of the intelligent agent can be continuously optimized to improve the efficiency of task execution; different intelligent agents can obtain different rule sets based on their respective initial states and task progress to achieve personalized behavior patterns; the preset rule library can be continuously expanded, enabling the system to cope with more diverse scenarios and tasks.

[0214] In another embodiment, generating an identifier of an initial interaction result includes:

[0215] Based on the behavioral data, determine the initial interaction state between each agent and record the behavioral output of each agent in this state;

[0216] Get the reaction time corresponding to each agent's behavior output;

[0217] Associating behavioral outputs with corresponding reaction times to generate initial interaction results;

[0218] Extract key behavioral features and time features from the initial interaction results and construct an initial interaction feature set;

[0219] Encoding the initial interaction feature set to generate an identifier of the initial interaction result;

[0220] In the logo, the behavior output and corresponding reaction time of each agent are represented separately and formatted to ensure the consistency and readability of the logo.

[0221] The working principle of the above technical solution is as follows: Initial state determination: According to the existing behavior data, set the initial interaction state for each agent and record its behavior output, which establishes the starting point of the system. Record the time required for each agent to produce behavior output, which provides information in the time dimension. Associate the behavior output and reaction time to form the initial interaction result, which integrates the behavior and time data. Extract key behavior and time features from the interaction results and construct a feature set, which compresses the original data and retains key information. Encode the feature set and generate an identifier. This further compresses the data for easy storage and processing. Format the identifier to ensure consistency and readability.

[0222] The beneficial effects of the above technical solution are: through feature extraction and encoding, the amount of data to be processed is greatly reduced, and the system efficiency is improved; formatting ensures the data, which is convenient for subsequent analysis and comparison. At the same time, the behavioral and time dimensions are considered, providing more comprehensive interaction information. It provides a clear starting point and reference benchmark for subsequent interaction analysis. This method can be applied to different types and numbers of intelligent agents and has good scalability. Through identification, the behavior changes of each intelligent agent can be easily tracked.

[0223] In another embodiment, determining the adjusted interaction result includes:

[0224] Obtain the preset interaction analysis model, input all interaction behaviors into the interaction analysis model, and calculate the interaction performance between each intelligent agent in real time;

[0225] Determine an adjusted interaction state according to the interaction performance, and extract an adjusted interaction result;

[0226] Classify the adjusted interaction results, and mark successful interaction results and failed interaction results respectively;

[0227] Multiple key features corresponding to each interaction result are obtained, and the interaction score is calculated based on the features.

[0228] The working principle of the above technical solution is: use the preset interaction analysis model to process all interactive behaviors and calculate the interaction performance between agents in real time. This is a dynamic evaluation process. Based on the interaction performance, the system adjusts the interaction status of the agent and extracts new interaction results, which reflects the adaptability of the system. The adjusted interaction results are divided into two categories: success and failure, and marked, which provides a clear result judgment. Key features are extracted from each interaction result, and the interaction score is calculated based on these features, which realizes quantitative evaluation.

[0229] The beneficial effects of the above technical solution are: through real-time analysis and adjustment, the system can respond to changes quickly and improve interaction efficiency; adjust the state according to the interaction performance so that the system can adapt to different situations and improve flexibility; classify the results into success and failure, provide clear evaluation criteria, and facilitate subsequent decision-making; through feature extraction and score calculation, the quantification of interaction results is achieved, which is convenient for comparison and sorting.

[0230] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A planning and decision-making system based on multi-agent interaction, characterized in that: include: Agent unit, used to build multiple agents, each of which has independent behavior logic and decision-making capabilities; A planning unit, which is used to control multiple agents to perform initial planning tasks, where each agent makes independent decisions based on preset rules; An identification unit, used to determine the interaction status between agents and generate an identification of the initial interaction result, which includes the behavior output of each agent; An interaction result unit, used to dynamically adjust the interaction rules of the agent under user intervention, and determine the score and status of the adjusted interaction results of the agent, where the interaction results include successful interaction results and failed interaction results; The preset condition unit is used to generate a new planning task based on the continuous number of failed interaction results reaching a preset threshold, or when the current agent interaction state is determined to be a final state, and repeatedly execute the planning task until the system preset termination condition is reached.

2. A planning and decision-making system based on multi-agent interaction according to claim 1, characterized in that: The agent unit includes: A building module, used for generating a plurality of initial intelligent agent units, each of which includes an independent behavior logic component and a decision-making ability component; A behavior logic parameter module is used to determine the behavior logic parameters of each initial intelligent unit, and the behavior logic parameters are configured based on a preset decision tree algorithm; The modification module is used to generate multiple new intelligent units by copying and modifying the behavioral logic parameters of the initial intelligent units to adapt to different environments or task requirements.

3. A planning and decision-making system based on multi-agent interaction according to claim 1, characterized in that: The planning units include: An allocation module is used to control multiple agents to perform initial planning tasks and allocate an initial set of rules to each agent from a preset rule library; The recording module is used to determine the initial behavior path of each intelligent agent, record the results of independent decision-making of each intelligent agent based on the assigned rule set, and generate an identification of the initial behavior path.

4. A planning and decision-making system based on multi-agent interaction according to claim 1, characterized in that: Determine the initial interaction state between agents and generate an identification of the initial interaction result, including: Obtaining behavioral data of multiple agents in an initial interactive state, the behavioral data including specific behavioral output and reaction time of each agent in the interaction; Based on the acquired behavior data, determine the initial interaction state between the agents, record the behavior output of each agent in this state, and generate an identification of the initial interaction result; The identification of the initial interaction results includes the behavioral output of each agent and its corresponding reaction time; If the behavior output of the agent meets the preset rules, a first score of the agent is generated based on the reaction time and the behavior output; If the behavior output of the agent does not conform to the preset rules, the score of the behavior output is determined to be zero; Output and record the identification of the initial interaction result.

5. A planning and decision-making system based on multi-agent interaction according to claim 1, characterized in that: Determine the score and status of the adjusted agent interaction results, including: Obtain user intervention input and determine the dynamic adjustment content of the current agent interaction rules based on the input; Perform dynamic adjustments and generate new agent interaction processes based on the adjusted interaction rules, recording the behavior output and reaction time of each agent; According to the new agent interaction process, the adjusted interaction results are determined, and corresponding scores and status identifiers are generated. The interaction results include successful interaction results and failed interaction results.

6. A planning and decision-making system based on multi-agent interaction according to claim 1, characterized in that: The planning unit generates new planning tasks, including: Monitor the behavioral data of the agents during interaction and record the consecutive number of failed interaction results; When the number of consecutive failed interaction results reaches a preset threshold, or when the current agent interaction state is determined to be a final state, the planning unit is triggered to generate a new planning task; Perform a new planning task, update the agent's behavior strategy and restart the interaction process; Repeat the above planning tasks and record new interaction results and behavior data; Determine whether the current interaction state meets the termination condition preset by the system. If so, terminate the agent interaction process; If the system preset termination condition is not met, the system will continue to monitor and update the planning task according to the new interaction results until the termination condition is met.

7. A planning and decision-making system based on multi-agent interaction according to claim 2, characterized in that: Behavior logic parameters are configured based on a preset decision tree algorithm, including: Obtain the basic data set of the initial agent unit, which includes the agent's behavior records in different environments and the corresponding decision outputs; Based on the basic data set, the preset decision tree algorithm is trained. The training process includes feature extraction of different behavior records and division of each node of the decision tree according to different features to form a decision tree model containing multiple decision paths. According to the trained decision tree model, the initial behavior logic parameters of each intelligent unit are determined, and the initial behavior logic parameters include the initial behavior decision rules of the intelligent unit in a specific environment; By inputting the current state information of a specific environment, the decision tree algorithm is used to traverse each node, determine the optimal behavior path, and generate corresponding behavior logic parameters; According to the optimal path output by the decision tree algorithm, the behavior logic parameters of the intelligent agent are adjusted to form the final configured behavior logic parameter set.

8. A planning and decision-making system based on multi-agent interaction according to claim 3, characterized in that: Assign an initial set of rules to each agent from a preset rule base, including: Based on multi-level rule settings, determine the initial state and preset tasks of each agent; Extracting a first rule set corresponding to the initial state from a preset rule base; Assigning a first set of rules to corresponding agents and generating a preliminary planning task based on the first set of rules; During the execution of the preliminary planning task, monitoring the task progress of the agent, dynamically adjusting the first rule set based on the monitoring result, and determining the corresponding second rule set; Determine whether the rule upgrade condition is met. If so, extract the second rule set corresponding to the current task stage from the preset rule library, assign the second rule set to the corresponding agent, and generate a new planning task; Continue to monitor the agent’s task progress based on the new planning task and make further rule set adjustments.

9. A planning and decision-making system based on multi-agent interaction according to claim 4, characterized in that: Generates the initial interactive result identifier, including: Based on the behavioral data, determine the initial interaction state between each agent and record the behavioral output of each agent in this state; Get the reaction time corresponding to each agent's behavior output; Associating behavioral outputs with corresponding reaction times to generate initial interaction results; Extract key behavioral features and time features from the initial interaction results and construct an initial interaction feature set; Encoding the initial interaction feature set to generate an identifier of the initial interaction result; In the logo, the behavior output and corresponding reaction time of each agent are represented separately and formatted to ensure the consistency and readability of the logo.

10. A planning and decision-making system based on multi-agent interaction according to claim 5, characterized in that: Determine adjusted interaction outcomes, including: Obtain the preset interaction analysis model, input all interaction behaviors into the interaction analysis model, and calculate the interaction performance between each intelligent agent in real time; Determine an adjusted interaction state according to the interaction performance, and extract an adjusted interaction result; Classify the adjusted interaction results, and mark successful interaction results and failed interaction results respectively; Multiple key features corresponding to each interaction result are obtained, and the interaction score is calculated based on the features.

Images