Sustainable consumption behavior simulation system fusing large language model and multiple agents

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a sustainable consumption behavior simulation system fusing a large language model and multiple agents, which comprises an external semantic generation module and an internal behavior evolution module, the external semantic generation module constructs a multi-type large language model agent, generates semantic intervention information according to state parameters of the internal behavior evolution module, and converts the semantic intervention information into a standardized numerical signal through a preset mapping function; an internal behavior evolution module constructs a consumer agent with double-layer attributes, cognitive interaction is completed through social similarity calculation, uncertainty interval verification and nonlinear attitude updating by adopting an optimized social influence model process based on interactive network topology, and prediction from cognition to behavior tendency is realized by an embedded neural network. And meanwhile, the new group state is fed back to an external module to form a closed loop. The system has both semantic dynamics and behavior simulation authenticity, and the accuracy and practicability of sustainable consumption behavior evolution simulation are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a sustainable consumer behavior simulation system that integrates a large language model and multiple agents. Background Technology

[0002] Sustainable consumer behavior is not a single, static decision, but a dynamic evolutionary process driven by multiple internal and external factors, including individual cognition, social interaction, media guidance, and brand communication. Clarifying the formation and evolution mechanism of this behavior is a core prerequisite for optimizing brand strategies and achieving low-carbon transformation.

[0003] Traditional research often relies on static statistical methods such as structural equation modeling and logistic regression, which can only characterize cross-sectional features of consumer behavior in specific contexts and cannot represent the dynamic evolution trajectory of behavior and group interaction effects. To overcome this limitation, simulation technologies such as multi-agent systems (MAS) have been introduced into the field of consumer behavior research and have become mainstream modeling tools. This technology simulates the interaction between individuals and the relationship between individuals and the environment by pre-setting fixed internal attributes (such as green consumption preferences and herd mentality) and static external environmental rules (such as fixed policy influence coefficients and brand communication intensity) for consumer agents, thereby deducing the behavioral evolution patterns at the group level.

[0004] In the technical system of consumer cognitive communication simulation, opinion dynamics models are the core supporting technology, with the Deffuant model being a classic solution. Its core mechanism is that when the difference in opinions between two consumer agents is below a fixed confidence threshold, they converge through a linear average to form group consensus. To overcome the shortcomings of traditional opinion dynamics models such as the Deffuant model in characterizing social heterogeneity and the complexity of cognitive evolution, the SIM (Grow & Flache) model was proposed. Based on social identity theory and uncertainty theory, this model, through social similarity calculation, dynamic uncertainty interval verification, and asymmetric influence mechanisms, can more realistically simulate complex socio-psychological processes such as attitude polarization and group differentiation in social networks. Its theoretical framework encompasses core technical modules such as social similarity calculation, uncertainty interval and influence condition definition, nonlinear cognitive update mechanism, and dynamic adjustment of uncertainty level.

[0005] In recent years, single-agent systems based on Large Language Models (LLM) (LLM-Agent) have been gradually applied to the field of social behavior simulation due to their strong semantic generation and environmental awareness capabilities. This technology can generate dynamic text intervention information through prompts and context, adjust the text content based on simulation feedback, and transform semantic information into numerical variables that the simulation system can recognize, demonstrating significant technical advantages in scenarios such as social media dissemination.

[0006] However, the existing technological system still has many shortcomings and is difficult to meet the research needs of simulating sustainable consumer behavior in complex scenarios, as follows:

[0007] First, multi-agent system (MAS) technology suffers from two major drawbacks: semantic gaps in environmental modeling and static nature. On the one hand, it simplifies external information rich in context and persuasive logic, such as news reports, policy documents, and brand narratives, into single, fixed numerical parameters, losing semantic information that is highly relevant to consumer decision-making. On the other hand, it cannot adapt to the dynamic changes in real-world scenarios, such as the evolution of media topics, the refinement of policy provisions, and adjustments to brand strategies, resulting in low model fidelity and significant discrepancies between simulation results and actual consumer behavior.

[0008] Second, traditional viewpoint dynamics models (represented by the Deffuant model) suffer from the core problem of oversimplifying cognitive interaction mechanisms. Firstly, they assume symmetry and uniformity in individual attitude updates, neglecting the uncertainty of psychological cognition and the selective influence of social identity and social circle identification on information absorption. Secondly, the fixed confidence threshold setting cannot adapt to real-world social phenomena such as opinion differentiation and attitude opposition, making it difficult to reproduce complex socio-psychological processes such as group polarization and cognitive convergence.

[0009] Third, while the SIM (Grow & Flache) model theoretically improves the ability to characterize the complexity of social attitude evolution, it suffers from the critical problem of high computational complexity, reaching [amount missing]. Direct application in large-scale multi-agent simulation scenarios will lead to low computational efficiency and a surge in simulation time, making it difficult to meet the real-time and scalability requirements of industrial-grade sustainable consumer behavior simulation.

[0010] Fourth, single-agent systems based on large language models (LLM-Agent) suffer from insufficient integration with multi-agent systems. First, their application scope is limited, failing to develop a systematic technical solution in the fields of socio-economic and consumer behavior simulation. Second, they lack a two-way feedback mechanism; the semantic information generated by the LLM-Agent cannot be quantified to influence group behavior evolution, and the dynamic state of consumers cannot guide the LLM-Agent to optimize intervention information. Third, they suffer from engineering defects such as high computational resource consumption, insufficient output stability, and poor repeatability, making it difficult to meet the requirements of controllability, efficiency, and interpretability of simulation systems, and failing to achieve deep linkage between semantic information and behavioral simulation.

[0011] In summary, existing technologies cannot construct simulation models that integrate multiple internal and external factors, accurately depict the dynamic evolution of sustainable consumption behavior, adapt to complex social psychological mechanisms, and have reasonable computational costs. Building such models has become a key technological requirement for promoting the green transformation of industries and formulating scientific low-carbon policies. Summary of the Invention

[0012] Therefore, this invention aims to address the technical problems of traditional multi-agent sustainable consumption behavior simulation models, such as static external environmental interventions lacking semantic information, simplified cognitive interaction mechanisms making it difficult to simulate complex social phenomena like attitude polarization, and the lack of system-level bidirectional closed-loop feedback, limited application scope, high computational cost, and insufficient output stability in the integration of large language models and multi-agent systems. Thus, this invention provides a sustainable consumption behavior simulation system that integrates large language models and multi-agent systems. This system includes: an external semantic generation module and an internal behavior evolution module; The external semantic generation module is configured to: construct multiple types of large language model intelligent agents; each type of large language model intelligent agent, based on the current state parameters generated by the internal behavior evolution module, calls its corresponding external knowledge base to dynamically generate unstructured semantic intervention information; and converts the semantic intervention information into standardized numerical signals that can be parsed by the internal behavior evolution module through a preset mapping function. The internal behavior evolution module is configured to: construct and deploy multiple consumer agents and regulatory agents in an interactive network; At each simulation step, the standardized numerical signal is received; based on the optimized social impact model, multiple consumer agents are driven to perform cognitive interactions within their network neighbors, and the sustainable consumption cognitive value of each consumer agent is calculated and updated. For each consumer agent, its updated sustainable consumption cognitive value is assembled together with the product feature signal and scenario context signal obtained from external signals into an input vector, which is then input into its embedded behavioral decision neural network. After forward propagation and nonlinear mapping of the behavioral decision neural network, its sustainable consumption behavior tendency value in the next simulation step is output. The regulatory agent monitors the latest status of all consumer agents in real time, calculates macro-level indicators at the group level, encapsulates them into new state vectors, and feeds them back to the external semantic generation module in real time through a preset data interface as input for generating the next round of semantic intervention information, thereby forming a closed-loop simulation circuit.

[0013] In one embodiment of the present invention, based on an optimized social influence model, multiple consumer agents are driven to perform cognitive interactions within their network neighbors, and the sustainable consumption cognitive value of each consumer agent is calculated and updated, specifically including: Based on the Watts-Strogatz small-world network model, an interactive topology environment is constructed for multiple consumer agents. By adjusting the average degree K and the reconnection probability β, it adapts to real social network simulation scenarios with different connection densities and randomness. The average degree K controls the initial number of connections for each agent, adjusting the overall connection density of the network. The reconnection probability β controls the reconstruction probability of long-range connections, introducing random connections while maintaining local clustering characteristics. In this interactive topology environment, the cognitive interactions between the multiple consumer agents are restricted to their direct network neighbors. Based on the aforementioned interactive topology environment, at each simulation step, for any agent... With any of its neighboring intelligent agents Based on the difference in their perceived values ​​of sustainable consumption, the social similarity between the two groups was calculated. : ,in, For intelligent agents Intelligent agents with neighbors The social similarity between them is used to quantify how close they are in their understanding of sustainable consumption; and respectively intelligent agents and Neighbor Intelligent Agent The sustainable consumption perception value is a real number that changes continuously within the interval [0,1]. Based on the aforementioned social similarity and intelligent agent The static attributes are used to perform acceptability checks on the influence of neighboring intelligent agents; Update the agent based on the verified neighbor agents. Sustainable consumption awareness value.

[0014] In one embodiment of the present invention, based on the social similarity and the agent... The static attributes are used to perform acceptability checks on the influence of neighboring agents, specifically including: Based on the preset similarity threshold Preliminary distinction of interaction types: when When, agents i and j are determined to be a cognitively similar group; when If so, they are identified as a group with cognitive differences; Based on the aforementioned cognitively similar groups, for each consumer intelligent agent Set an internal acceptance threshold. By comparing the absolute values ​​of the cognitive differences between the two With the internal acceptance threshold Perform acceptance checks: like Then determine the neighboring intelligent agent Cognition is at the level of the intelligent agent We accept this impact within an acceptable range; Otherwise, determine the neighboring intelligent agent. Cognitive differences exceed those of intelligent agents Within acceptable limits, reject this impact; Wherein, the internal acceptance threshold By intelligent agents It is calculated from static attributes including susceptibility and trust level.

[0015] In one embodiment of the present invention, the updating agent The formula for calculating the perceived value of sustainable consumption is as follows: ,in, For intelligent agents At simulation time Sustainable consumption perception value; The preset learning rate parameter is used to control the overall rate of cognitive updates; For intelligent agents Its neighboring intelligent agents The connection weights between the links are used to characterize the importance of the connection to cognitive impact; For intelligent agents The set of neighbors in the interactive topology environment; The connection weight Based on intelligent agents With any of its neighboring intelligent agents Social similarity calculated from the current perception value of sustainable consumption And / or the inherent properties of the connecting edge pre-assigned in the small-world network topology are dynamically determined.

[0016] In one embodiment of the present invention, the updating agent The formula for calculating the perceived value of sustainable consumption also includes a social stress item. It is used to quantify the conformity pressure an individual experiences due to their perception of the behavioral choices of the majority in their social circle; its intensity is related to the intelligence agent. The proportion of sustainable consumers among the direct neighbor agents is positively correlated, wherein the sustainable consumer is defined as a neighbor agent whose sustainable consumption behavior tendency value exceeds a preset behavior adoption threshold; Introducing the aforementioned social pressure item Then, update the intelligent agent. The formula for calculating the perceived value of sustainable consumption is extended as follows: ,in, This represents the social pressure influence coefficient.

[0017] In one embodiment of the present invention, the behavioral decision neural network is a BP neural network, the structure of which includes: The input layer contains three neurons, which respectively receive: the agent's sustainable consumption cognition value at the current simulation moment, the product sustainability strength signal representing the product's environmental attributes, and the product supply facility adaptability signal representing purchase convenience and facility support. The hidden layer contains 5 to 64 neurons and uses the ReLU activation function for nonlinear transformation. The output layer contains one neuron. The sigmoid activation function is used to map the weighted sum of the neurons to the interval [0,1], and outputs the sustainable consumption behavior tendency value at the next simulation time.

[0018] In one embodiment of the present invention, the supervisory agent monitors the latest state of all consumer agents in real time, calculates macro-level indicators at the group level, encapsulates them into a new state vector, and feeds it back to the external semantic generation module in real time through a preset data interface, specifically including: At each simulation step At the end, the updated evolution states of all consumer agents are collected and aggregated in real time, and based on the updated evolution states, macro indicators at the group level are calculated, including: the proportion of sustainable consumers, the average sustainable consumption awareness of the group, and the average sustainable consumption behavior of the group. The macroscopic indicator is encapsulated into a representation of the current simulation step size. At the end, the actual state vector of the overall population evolution result ; The actual state vector is transmitted through a preset data interface. Feedback is sent to the external semantic generation module to trigger the external semantic generation module in the next simulation step. The semantic intervention generation process.

[0019] In one embodiment of the present invention, the semantic intervention information is converted into a standardized numerical signal that can be parsed by the internal behavior evolution module through a preset mapping function, specifically including an inside-out mapping step and an outside-in mapping step: In the inside-out mapping step, the current simulation step size is received. At the end, the current state vector, which represents the evolutionary state of the consumer agent group, is fed back by the internal behavior evolution module. ; Based on the current state vector It uses a template-based prompt word construction function. Generate semantic cue words with targeted intervention objectives for at least one type of external large language model agent. The formal representation of this process is as follows: ; In the outward-to-inward mapping step, the semantic prompt words are... The input is fed into the corresponding large language model agent, triggering it to generate structured semantic response information. The semantic response information It includes at least one quantifiable semantic impact parameter; Through a pre-set influence mapping matrix The semantic response information At least one semantic influence parameter is mapped to the expected influence on the consumer agent's state vector. : ; Based on the expected impact and the current state vector The expected system size in the next simulation step is calculated. The target state vector achieved The process is formally represented as: The target state vector This is the output standardized numerical signal.

[0020] In one embodiment of the present invention, the influence mapping matrix It is configured to linearly combine the semantic influence parameters of different agents according to preset weights to calculate the direct influence on different dimensions of the state vector; its expression is as follows: ,in, This represents the response value of the brand's intelligent agent. This represents the response value of the rule-based agent. This represents the response value of the media agent; , , , The influence coefficient is a pre-calibrated value.

[0021] In one embodiment of the present invention, each consumer agent includes two independent but related types of attribute information: one type is static attribute information, which is used to characterize the consumer's inherent social identity and underlying traits, and remains stable throughout the simulation period without dynamic changes during the interaction process, including conformity, trust, and willingness to accept sustainable consumption; the other type is dynamic attribute information, which is used to provide target evolution variables for cognitive interaction and consumption decision-making, including sustainable consumption cognitive value and sustainable consumption behavior tendency value.

[0022] Compared with the prior art, the above-described technical solution of the present invention has the following advantages: Firstly, by constructing an external semantic generation layer driven by a large language model, the static and numerical external intervention parameters in traditional simulation are transformed into dynamic and semantically rich intervention information flows. This effectively solves the problems of semantic deficiency and static nature in external environment modeling, significantly improves the realism and ecological validity of the simulation environment, and can more accurately reflect the dynamic evolution process of media, policy, and brand information. Secondly, an optimized SIM (Grow & Flache) model (ESW-SIM) based on small-world network topology is introduced into the group behavior evolution layer. It innovatively adopts an optimized social similarity calculation and weighted average cognitive update mechanism, which significantly improves the computational efficiency and large-scale simulation capability of the model. While preserving the core characteristics of social dynamics such as attitude convergence and group differentiation, it can achieve efficient concurrent simulation of thousands or even tens of thousands of consumer agents, providing a reliable computational tool and explanatory support for the macro trend analysis and policy simulation of sustainable consumption behavior. Third, the semantic-behavioral dual-layer closed-loop simulation framework, with the help of standardized interfaces and semantic-numerical mapping function Φ, realizes bidirectional feedback between the two layers, which not only allows external intervention to be dynamically adjusted according to the evolution of the population, but also ensures the computational efficiency and result stability of large-scale simulation, and strengthens the system's adaptive capability. Fourth, by combining three types of LLM-Agents, an optimized SIM (Grow & Flache) model embedded with a small-world network, and a BP neural network, a complete pathway from semantic intervention to cognitive update and then to behavioral decision-making was established. This accurately simulated the cognitive-behavioral transformation path, reduced the simulation bias of traditional research, significantly improved the predictive ability of sustainable consumption behavior adoption rate, and the predicted data were highly consistent with the empirical data. Fifth, the technical solution has proven its practicality in the context of sustainable consumption of clothing, realistically presenting characteristics such as convergence of group cognition and differentiation of attitudes. It also has excellent scalability and can provide reliable simulation tools and decision support for policy making and brand strategy optimization. Attached Figure Description

[0023] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0024] Figure 1 This is a schematic diagram of a sustainable consumption behavior simulation model structure that integrates a large language model and multiple agents, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of an LLM-Agent operating mechanism provided in an embodiment of the present invention; Figure 3 This is a diagram of a BP neural network structure model for consumer behavior decision evolution provided in an embodiment of the present invention; Figure 4 This is a scenario diagram of a dynamic representation model of sustainable consumption behavior, using the apparel industry as an example. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0026] Reference Figure 1 and Figure 2 As shown, the present invention provides a sustainable consumption behavior simulation system that integrates a large language model and multiple agents. The system includes: an external semantic generation module and an internal behavior evolution module. The external semantic generation module is configured to: construct multiple types of large language model intelligent agents; each type of large language model intelligent agent, based on the current state parameters generated by the internal behavior evolution module, calls its corresponding external knowledge base to dynamically generate unstructured semantic intervention information; and converts the semantic intervention information into standardized numerical signals that can be parsed by the internal behavior evolution module through a preset mapping function. The internal behavior evolution module is configured to: construct and deploy multiple consumer agents and regulatory agents in a small-world network; at each simulation step, iteratively update the internal state of the consumer agents based on the standardized numerical signals; based on the updated agent state, perform multi-dimensional simulation and deduction of the cognitive interaction, social communication and sustainable consumption decision-making process of the consumer group, and feed back the new state parameters of the consumer group to the external semantic generation module to form a closed-loop simulation circuit.

[0027] Furthermore, the external semantic generation module is used to construct three types of functional large language model agents (LLM-Agent): media agent, rule agent, and brand agent. Its core function is to construct a dynamic semantic intervention mechanism to generate targeted semantic information with scene adaptability, so as to accurately simulate external social environmental factors such as media information diffusion, rule constraint intervention, and brand communication guidance, and to target the intervention process of consumers' sustainable consumption cognition formation and decision-making behavior evolution.

[0028] The system configures four core functional sub-modules for the three types of intelligent agents: environmental perception, semantic planning, text generation, and interface transmission. At the same time, it builds a supporting resource system, including a media dissemination knowledge base, an intervention rule database (with built-in reward-type and constraint-type intervention logic), and a brand sustainability attribute information database (covering green feature data from the product and supply sides), to provide basic resource support for the operation of each intelligent agent.

[0029] To address the specific needs of the brand agent, additional sub-agents for product sustainability and facility supply are deployed: the product sustainability sub-agent stores core product information such as environmentally friendly material parameters, carbon footprint accounting data, and green certification qualifications; the facility supply sub-agent stores supply-side data such as the brand's green supply chain architecture, spatial distribution of recycling facilities, and convenience of green purchasing channels, enabling hierarchical representation and precise retrieval of the brand's sustainability attributes.

[0030] A one-way data interaction interface is built between the environmental perception submodule and the internal behavior evolution module for all agents. A standardized data receiving protocol is configured to ensure that the core simulation state parameters output by the internal behavior evolution module can be collected in real time, including the average cognitive value of the group, cognitive difference, behavior adoption rate, consumer price sensitivity, and environmental concern. At the same time, a standardized JSON encapsulation protocol is configured for the interface transmission submodule, defining fixed fields such as agent type, intervention type, specific intervention semantic text, and generation timestamp to ensure the standardization and parsability of semantic information transmission.

[0031] All three types of LLM-Agents follow a unified logic for the collaborative operation of four core functional sub-modules: environment awareness, semantic planning, text generation, and interface transmission. The specific execution steps are as follows: The environmental perception submodule obtains the target state parameters for the current simulation cycle from the internal behavior evolution module through a preset data interaction interface. These parameters include the media agent focusing on the group's cognitive differences and average cognitive level; the rule agent focusing on the group's behavior adoption rate and historical simulation feedback data; and the brand agent focusing on consumers' price sensitivity and environmental concerns, providing accurate decision-making basis for semantic generation. Based on the collected state parameters, the semantic planning submodule calls a dedicated resource library to generate a targeted intervention semantic logic framework: the media agent calls the media communication knowledge base to match an appropriate topic-oriented framework; the rule agent calls the intervention rule database to match a reward-type or constraint-type rule framework; and the brand agent calls the brand sustainability attribute information library and links with the product / supply terminal intelligent agents to generate the corresponding communication content framework. The text generation submodule transforms the semantic planning framework into prompts that can be parsed by the large language model. It then calls the large language model to generate natural language semantic text that conforms to the intervention logic. The text must carry contextual adaptability, emotional guidance, or persuasive features, such as "High-cost-performance environmentally friendly clothing selection guide" and "Green consumption subsidy application specifications", avoiding rigid expressions that are singular and numerical. The interface transmission submodule standardizes and encapsulates the generated natural language semantic text according to a preset JSON structure to generate the final semantic intervention information.

[0032] The semantic intervention information is transformed into a standardized numerical signal that can be parsed by the model algorithm in the internal behavior evolution module through a preset mapping function Φ (i.e., a two-layer mapping semantic intervention closed-loop mechanism, DM-CLoSeM). Specifically, this includes an inside-out mapping step and an outside-in mapping step. In the aforementioned inside-out mapping step (state awareness and prompt generation), the supervisory agent ( At each simulation step Real-time monitoring of consumer intelligent agents ( The evolutionary status of the group was analyzed, and key macroeconomic indicators were calculated, including the proportion of sustainable consumers. ( , For indicator functions, (Adoption threshold for consumer behavior) and average group perception of sustainable consumption. ( , Indicates the current simulation step size , No. (where N is the number of consumer agents) and the average sustainable consumption behavior of the group. ( , Indicates the current simulation step size , No. The sustainable consumption behavior of each consumer agent (where N is the number of consumer agents) is used to construct the current state vector. : ; Based on the current state vector The three types of LLM-Agents—media, rules, and brand—are each constructed using a template-based prompt word function. Developing targeted semantic prompts: Media prompts Rule prompt words Brand prompts The inward-outward mapping process is formalized as follows: ,in This is the vector for the prompt word.

[0033] In the outward-to-inward mapping step (semantic parsing and influence quantification), various LLM-Agents call large language models (such as the DeepSeek API) through encapsulated interfaces to generate structured JSON response information. Among them, the response information of the media agent Includes parameters such as conformity, trend, and propagation; and the response information of the rule agent. Includes parameters such as regulatory intensity, incentive intensity, and penalty intensity; brand agent response information. It includes parameters such as marketing intensity, sustainability, and commitment.

[0034] Specifically, , , This document describes a standardized process for generating quantifiable semantic intervention parameter vectors for media agents, rule agents, and brand agents within the external semantic generation module, based on the aggregated group states of the regulatory agent. The calculation of these three types of response values ​​follows a unified technical workflow: "Regulatory agent state aggregation → inward-outward mapping to generate prompt words → LLM call to generate structured JSON response → parsing and quantifying to write to agent attributes → outward-inward mapping to achieve influence injection." This standardized process transforms the internal group states of the system into quantifiable intervention parameters. The specific calculation process is as follows: The calculation process is as follows: When the system calculates the proportion of sustainable consumers When this indicates a low sustainable adoption rate, the media agent calls the templated prompt word construction function. ,Will Macroeconomic indicators are converted into natural language prompts. This achieves a structured mapping from the system's internal numerical states to semantic prompts, a process formally represented as: .

[0035] The prompt word template clearly includes a status description, parameter requirements, and format constraints. For example, "The current sustainable adoption rate is low, but the trend is rising. Please return JSON format data based on this status, including three parameters: conformity, trend, and viral. The parameter values ​​are all in the range of [0,1]", ensuring that the response generated by the large language model is targeted and parsable.

[0036] The media agent calls the DeepSeek API (model parameter temperature=0.7) through the LLMClient tool to send prompt words. Inputting data into a large language model triggers the generation of a structured semantic response. An example of the structured JSON data returned by the large language model is shown below: {"conformity":0.78,"trend":0.66,"viral":0.82}.

[0037] The system parses the JSON data generated by LLM, extracts the values ​​of three fields: conformity, trend, and propagation, and performs range validation and standardization on them within the [0,1] interval to ensure that the parameter values ​​meet the system intervention requirements. Subsequently, the parsed quantified values ​​are mapped to the internal attributes of the media agent: conformity corresponds to the intensity of conformity influence. Trend-related trend amplification strength , Transmissibility corresponds to the propagation coefficient ,Finish The quantification and implementation of media response vectors. The expression is: Based on the example data above, we can obtain .

[0038] The calculation process is as follows: The rule-based agent calls the templated prompt word construction function. ,Will The macroeconomic indicators are embedded with preset prompt templates to generate targeted prompts with role positioning and format constraints. This achieves a structured transformation from the internal numerical state of the system to semantic intervention requirements, and this process is formally represented as follows: .

[0039] The prompt template explicitly includes a status description, role instructions (rule maker), intervention goals, and output format requirements. For example: "The current sustainable adoption rate is 0.25, the group's cognitive level is low and significantly differentiated. Please provide an intervention plan as the rule maker and return JSON format data, including three parameters: regulation, incentive, and penalty, all with values ​​ranging from [0,1]." This ensures that the response generated by the large language model accurately matches the intervention requirements.

[0040] The rule-based agent calls the DeepSeekAPI via the LLMClient tool (model parameter temperature=0.7) to provide prompt words. Inputting data into a large language model triggers the generation of structured semantic responses related to supervision, incentives, and penalties. An example of the structured JSON data returned by the large language model is shown below: {"regulation":0.65,"incentive":0.80,"penalty":0.40}.

[0041] The system parses the JSON data generated by LLM, extracts the values ​​of the three fields: regulation, incentive, and penalty, validates them within the [0,1] range, and maps them to the core quantization attributes of the rule-based intelligent agent, thus completing the process. Quantification.

[0042] The regulation field is mapped to the regulatory intensity parameter. The incentive field is mapped to the incentive intensity parameter. The penalty field is mapped to the penalty intensity parameter. Meanwhile, to ensure field compatibility and system scalability, the rule agent additionally retains synonymous internal fields: regulation_strength (equivalent to regulation_level, serving as an internal storage field for regulatory strength) and subsidy_level (equivalent to incentive_level, serving as a field associated with subsidy strength). The consistency of their values ​​is strictly guaranteed through preset parsing mapping rules.

[0043] At this point, the rule response vector Based on the example data above, we can obtain... .

[0044] The calculation process is as follows: The brand agent calls the templated prompt word construction function. ,Will The macroeconomic indicators are embedded with preset prompt templates to generate targeted prompts with role positioning and format constraints. The prompt template explicitly includes a status description, professional role instructions (brand marketing and sustainability strategy expert), intervention requirements, and output format requirements. For example: "The current sustainable consumer ratio is 0.25%, and the group's awareness level is low. Please develop an intervention plan as a brand marketing and sustainability strategy expert and return JSON format data, including four parameters: marketing (marketing intensity), sustainability (sustainability focus), commitment (sustainability commitment), and price (pricing strategy), with all parameters ranging from [0,1]." This ensures that the response generated by the large language model accurately matches the professional scenario requirements of brand intervention.

[0045] The brand agent uses the LLMClient tool to call the DeepSeekAPI (model parameter temperature=0.7), inputting the prompt word Ptbrand into the large language model, triggering the generation of structured semantic responses related to brand marketing, sustainable value communication, commitment endorsement, and pricing strategy. An example of the structured JSON data returned by the large language model is shown below: {"marketing":0.88,"sustainability":0.75,"commitment":0.90,"price":0.70}.

[0046] The system parses the JSON data generated by LLM, extracts the values ​​of four fields: marketing, sustainability, commitment, and price, and after validation within the range of [0,1], maps them to the core quantitative attributes of the brand agent, thus completing the quantitative implementation of Jtbrand. Specifically, the marketing field is mapped to the marketing intensity parameter. The sustainability field is mapped to a parameter indicating the degree of focus on sustainability information. The commitment field is mapped to a sustainability commitment credibility parameter. The price field is mapped to the price strategy strength parameter. At this point, the brand response vector .

[0047] LLM Influence Manager parses JSON response information And based on a preset influence mapping matrix Transform semantic parameters into expected impact on the consumer agent's state vector. : Influence mapping matrix Defined as: ,in , , , These are pre-defined influence coefficients, corresponding to the influence intensity of brand marketing, regulatory rules, media conformity, and regulatory incentives, respectively.

[0048] Based on the expected impact and the current state vector The expected system size in the next simulation step is calculated. The target state vector achieved The process is formally represented as: , The outward-to-inward mapping process is formalized as follows: ,in For the semantic generation function of a large language model, satisfying .

[0049] Closed-loop mechanism integration: Combining the two mappings above, the complete semantic intervention closed-loop mechanism is represented as a composite function: ,Right now: The new state vector This refers to standardized numerical signals that can be parsed by the model algorithm in the internal behavior evolution module, used to drive the cognitive and behavioral evolution of the consumer agent in the next moment.

[0050] It should be noted that various LLM-Agents are only responsible for the dynamic generation of external semantic information and do not participate in the direct interaction process between consumer agents; the entire process from semantic generation to state update is handled through a mapping function. Achieve end-to-end closed-loop encapsulation, completing the full mapping of "internal state perception → external semantic generation → influence quantification → internal state update".

[0051] The internal behavior evolution module is designed to simulate the cognitive interaction and behavioral decision-making process of consumers in a social network environment. This module includes two types of agents: consumer agents and regulatory agents.

[0052] Construct multiple consumer agents, and configure two independent but related types of attribute information for each consumer agent: One type is static attribute information, which is used to characterize the inherent social identity and underlying characteristics of individual consumers. It remains stable throughout the simulation cycle and does not change dynamically with the interaction process. It includes conformity, trust, and initial willingness to accept sustainable consumption. Another type is dynamic attribute information, which provides target evolution variables for cognitive interaction and consumer behavior decision-making. It is updated in real time with external intervention and group interaction, including Sustainable Consumption Cognition (SCC) value and Sustainable Consumption Behavior Propensity (SCCB) value. For the regulatory agent, it is responsible for monitoring the group's state and pre-collecting macro-indicators such as average cognitive level, cognitive dissimilarity, behavioral adoption rate, and polarization index. Its configuration includes a communication interface with an external semantic generation module, which can update the collected individual sustainable consumption cognitive values. Transform into It is passed to the external semantic generation module.

[0053] The interaction topology of consumer agents is constructed based on the Watts–Strogatz small-world network model. By adjusting the two parameters, the average degree K and the reconnection probability β, the characteristics of high clustering coefficient (simulating the local clustering of acquaintance circles) and short average path length (simulating the global reachability of cross-circle information transmission) that exist simultaneously in real social networks are simulated. The interaction between agents is restricted to the scope of their direct network neighbors. This not only conforms to the locality of real social interaction, but also greatly reduces the computational cost of global matching in large-scale simulations.

[0054] The optimized SIM (Grow & Flache) model is used to achieve dynamic interaction and attitude updates of sustainable consumption perceptions among consumer agents, and is implemented in the following steps: The basic dimensions for initializing cognitive interaction for each consumer agent include: Sustainable Consumption Cognition Value (SCC): Represents the agent's level of cognition on sustainable consumption issues, with a value range of [0,1]. The higher the value, the more positive the cognition. This value is the core variable that dynamically evolves in the simulation. The set of static attributes includes conformity (measures the degree to which an individual is influenced by the opinions of the group), trust (the credibility threshold that influences the acceptance of information), influence intensity (an individual's ability to output opinions to the outside world), and susceptibility (an individual's sensitivity to external influences). These attributes are set during simulation initialization and remain relatively stable over the period, jointly regulating the intensity and direction of the agent's social interaction.

[0055] Based on the aforementioned interactive topology environment, at each simulation step, for any agent... With any of its neighboring intelligent agents Based on the difference in their perceived values ​​of sustainable consumption, the following optimized formula is used to calculate their social similarity in real time: ,in, For intelligent agents Intelligent agents with neighbors The social similarity between them is used to quantify how close they are in their understanding of sustainable consumption; and respectively intelligent agents and Neighbor Intelligent Agent The sustainable consumption perception value is a real number that varies continuously within the interval [0,1]. This formula uses the absolute difference between the two perception values ​​to characterize social distance; the smaller the difference, the higher the social similarity. The closer the value is to 1, the more similar the cognitive states of the two individuals are, and theoretically, the stronger their willingness to interact.

[0056] Based on the calculated social similarity Determine interaction preferences: Set a similarity threshold (For example ), used to initially distinguish interaction types: when If so, then agents i and j are determined to be a cognitively similar group, which is likely to have positive interactions that promote consensus; when When this happens, the group is identified as having cognitive differences, and their tendency to interact weakens, and they may even reject viewpoints, which provides a basis for simulating group differentiation.

[0057] During the interaction process, not all neighborly influences are accepted. Even positive interactions that pass the initial assessment still require internal verification by the receiver to take effect. The system calculates an internal acceptance threshold based on the static attributes of the consumer agent i (especially susceptibility and trust level). The core of this verification is comparing cognitive differences with individual tolerance: like Then determine the neighboring intelligent agent Cognition is at the level of the intelligent agent We accept this impact within an acceptable range; Otherwise, determine the neighboring intelligent agent. Cognitive differences exceed those of intelligent agents Within acceptable limits, we reject this impact.

[0058] This optimization mechanism drives the threshold with clear and stable attributes. It replaces the complex dynamic uncertainty interval verification in the original model, significantly improving computational efficiency while ensuring the rationality of model behavior.

[0059] For neighboring agents that pass the above verification, the agents are updated using a weighted averaging mechanism based on the local structure of the network. Cognitive state: ,in, For intelligent agents At simulation time Sustainable consumption perception value For intelligent agents At the next simulation moment Updated awareness of sustainable consumption; For Neighbor Intelligent Agent At simulation time Sustainable consumption perception value; For intelligent agents The set of neighbors in a small-world network; For connection weights, social similarity can be used. Alternatively, the strength of network connections can be dynamically calculated based on preset parameters to give different neighbors a differentiated influence. This is the global learning rate parameter, used to control the overall update speed and prevent iterative oscillations.

[0060] In the process of updating consumer intelligence agent i's sustainable consumption cognition, to more realistically simulate social effects such as conformity, a social pressure term can be added to the above basic update formula. This social pressure item The simulated herd mentality pressure experienced by an individual due to perceived behavioral norms within their social circle is typically positively correlated with the proportion of neighbors directly connected to agent i who have been identified as sustainable consumers. Specifically, neighbors whose sustainable consumption propensity exceeds a predefined behavioral decision threshold are defined as sustainable consumers.

[0061] Introducing social pressure items Subsequently, the formula for updating the perceived value of sustainable consumption can be expanded to: ,in, This is the social pressure influence coefficient. This extended formula, by coupling the influence of differences in neighboring viewpoints (the main body of the formula) with the pressure of consensus on neighboring behavior (the social pressure term), jointly drives the evolution of individual cognition, thereby achieving a comprehensive simulation of the dual mechanisms of "viewpoint exchange" and "behavioral demonstration" in the process of social influence.

[0062] After the cognitive update is completed, the consumer agent's sustainable consumption behavior tendency value is processed by a pre-trained BP neural network embedded within it (such as...). Figure 3 (As shown) Decision prediction is performed. Specifically, the structure of the BP neural network is as follows:

[0063] The input layer, consisting of three neurons, receives the agent's sustainable consumption cognitive value at the current simulation moment. The product feature signal (i.e., the product sustainability strength signal PS, which characterizes the product's environmental attributes) and the scenario context signal (i.e., the product supply facility adaptability signal PFSC, which characterizes the ease of purchase and facility support) are obtained from external signals. The external product sustainability strength signal PS is used to characterize the strength of attributes such as brand environmental certification and carbon footprint. The external product supply facility adaptability signal PFSC is used to characterize external contexts such as the ease of recycling facilities and price subsidies for purchasing green products.

[0064] The hidden layer, configured with 5 to 64 neurons, performs a weighted summation of the aforementioned three-dimensional feature vectors using learnable weight parameters, followed by a ReLU function. By performing nonlinear mapping, the original signal is transformed into a high-order discriminative feature, solving the gradient vanishing problem while learning complex interaction patterns between features.

[0065] The output layer, containing one neuron, weights and summarizes the higher-order discriminative features, then maps the summarized result to the [0,1] value range using a Sigmoid activation function, and outputs the sustainable consumption behavior tendency value for the next simulation step. The predicted value, when When this indicates a very strong tendency towards sustainable consumption behavior; when At that time, it indicates that the tendency towards sustainable consumption behavior is extremely weak, and the tendency is towards traditional consumption.

[0066] At each simulation step At the end, the regulatory agent collects and aggregates the updated evolutionary states of all consumer agents in real time, and calculates macro-level indicators at the group level based on the updated evolutionary states, including: the group's average sustainable consumption awareness value, the proportion of sustainable consumers, and the behavior adoption speed; these macro-level indicators are then encapsulated into a representation of the current simulation step size. At the end, the actual state vector of the overall population evolution result The actual state vector is transmitted through a preset data interface. Feedback is sent to the external semantic generation module to trigger the external semantic generation module in the next simulation step. The semantic intervention generation process forms a closed-loop regulation of "cognitive evolution → state feedback → semantic intervention".

[0067] The following section uses the apparel industry as a typical application scenario to demonstrate the implementation and verification of the dynamic simulation scheme for sustainable consumption behavior based on a large language model and a multi-agent system proposed in this invention. At the same time, the specific implementation of the system's software and hardware architecture is described in detail to fully disclose the technical solution of this invention and ensure that those skilled in the art can reproduce the solution based on this embodiment.

[0068] The simulation data in this embodiment comes from two sources, both of which have undergone standardized preprocessing: The first is basic consumer shopping behavior data, derived from the Consumer Behavior and Shopping Habits Dataset. This dataset includes 3900 consumption records, which are expanded threefold after being shuffled to 11700 records. It covers 20 dimensions of behavioral characteristics, including demographics, purchase history, brand preference, and price sensitivity. After data cleaning, outlier removal, and standardization, this data is used to initialize the static attributes and consumption preferences of the consumer agent. The second is consumer sustainable consumption awareness and attitude questionnaire data. 2000 valid questionnaires were collected through a professional online survey platform. The questionnaire covers core psychological indicators such as awareness of sustainable clothing, environmental attitudes, purchase intentions, and value recognition. After reliability and validity testing, this data is used to set the initial cognitive state and decision parameters of the consumer agent.

[0069] To balance computational efficiency and the observability of simulation results, this embodiment configures 2000 consumer agents. The social interaction topology of the consumer agents adopts a Watts-Strogatz small-world network structure, with an average network degree of 6 and a reconnection probability of 0.1, to reproduce the high clustering coefficient and short average path length characteristics of real-world social networks. Based on the actual statistical results of the questionnaire survey, the initial sustainable consumption cognition value of the consumer agents is set to follow a normal distribution with a mean of 0.25 and a standard deviation of 0.18. The simulation cycle is set to 50 time steps to fully simulate the long-term evolution of consumer cognition and behavior under external intervention.

[0070] like Figure 4 As shown, the simulation process in this embodiment is divided into four core stages, each executed in a closed-loop sequence:

[0071] (1) System initialization stage: Based on the preprocessed questionnaire data and consumer behavior data, a complete digital profile of the consumer agent is constructed, including static attributes (conformity, trust, etc.), initial cognitive state (cognitive value of sustainable consumption issues) and network topology connection relationship, laying the foundation for subsequent simulation.

[0072] (2) Cognitive interaction stage: At each time step, the optimized SIM (Grow & Flache) model (ESW-SIM) embedded small world network is executed. The consumer agent sequentially completes operations such as social similarity calculation, influence reception condition verification, and weighted average cognitive update to realize cognitive propagation and attitude evolution within the group.

[0073] (3) External semantic intervention stage: The intervention logic of the LLM agent is triggered every 5 time steps. The media agent dynamically generates controversial or incentive-based sustainable clothing communication content based on the difference in group cognition and the average level of cognition. The rule agent outputs differentiated policy intervention schemes such as subsidies and certification based on reinforcement learning algorithms. The brand agent (including product sustainability sub-agent and facility supply sub-agent) adaptively generates “economic” or “environmentally friendly” brand narrative content based on consumers’ price sensitivity and environmental concerns. All intervention content is converted into quantitative signals and input into the simulation system through the semantic-numerical mapping interface.

[0074] (4) Behavioral decision-making stage: The consumer agent will input the iteratively updated sustainable consumption perception value, combined with external environmental signals such as product sustainability and supply facility conditions, into the embedded BP neural network to complete the decision mapping; the BP neural network has a three-layer structure with an input dimension of 3, a hidden layer node number of 32, and an output layer using the Sigmoid activation function, and finally outputs a sustainable clothing purchase behavior tendency value in the range of 0-1.

[0075] After 50 time steps of simulation, the system reached a stable evolutionary state, with core indicators showing significant optimization: the adoption rate of sustainable clothing increased to 67.3%, an increase of 42.3 percentage points from the initial baseline; the average consumer awareness of sustainable consumption increased from 0.31 to 0.72; the effects of different types of external interventions were quantifiable: price subsidies increased purchase conversion rates by 18.5%, media campaigns increased awareness dissemination speed by 32.7%, and brand sustainability narratives increased product trust by 27.2%. Based on the above simulation results, targeted optimization suggestions are provided for the apparel industry: on the product side, focus on improving clothing durability and optimizing pricing rationality; on the communication side, convey the long-term use value of sustainable clothing through empirical data; on the policy side, design a tiered subsidy scheme to improve the efficiency of fund utilization; on the channel side, optimize the shopping experience to enhance the efficiency of converting awareness into behavior.

[0076] This system can employ a high-performance computing architecture to meet the computational demands of large-scale social simulation. This high-performance computing architecture typically includes multi-core processors, large-capacity memory, high-performance GPU acceleration units, a hierarchical storage system, and high-speed network interconnection. In one specific embodiment, the computing node is configured with a 48-core AMD EPYC 9654 CPU and 128GB of memory. The acceleration computing unit uses an NVIDIA RTX 6000 GPU (96GB of video memory), the storage system employs a hierarchical scheme combining high-speed solid-state drives and large-capacity hard disk drives, and the network uses a high-speed Ethernet network. This configuration can support the concurrent operation of more than 10,000 agents, meeting the computational requirements of neural network inference and LLM service calls.

[0077] The software system adopts a modular Python architecture, and each functional module achieves data interaction and collaborative work through standard Python interfaces and function calls.

[0078] The core of the system consists of the following functional modules: The agent module contains independent Python class implementations of five types of agents: consumers, social media, policy, brand, and regulation. Each type of agent encapsulates its specific behavioral logic and state management functions. The simulation model module contains multiple independent simulation model Python files. One main simulation model file coordinates the global process and calls other models; several other model files implement network evolution and cognitive interaction logic based on different theories (such as SIM and Deffuant). The tools module contains several independent Python tool files, which respectively implement auxiliary functions such as calling the AI ​​large language model API (via HTTP request), managing configuration files (reading JSON and other formats), and exporting data (generating CSV and other files); User interface module: contains a graphical interface Python file written based on the Qt framework, providing windows for parameter input, simulation control, and result display; The experimentation and visualization module includes scripts for running comparative experiments and parameter tuning experiments, as well as Python code for generating graphs. The system implements key workflows in the following ways: Data and Configuration: Initial data and parameters are obtained by reading external CSV, JSON, and other files.

[0079] Simulation execution: The main program calls the update methods and model calculation functions of various intelligent agents step by step in a loop; External semantic integration: When needed, it obtains text responses by sending HTTP requests to the API endpoints of large language models such as DeepSeek, and parses them into signals usable by the system; Result output: The state data during the simulation process is recorded in memory and finally written to files such as CSV or displayed through charts.

[0080] The system runs in a standard Python environment and relies on scientific computing and network analysis libraries such as NumPy and PyTorch. Its architecture allows for adjustments such as adding agent types, agent behaviors, changing network models, or integrating new AI large language model services by modifying or replacing specific Python class and function files.

[0081] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application; it should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions; these computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0082] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0084] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation methods; those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively list all possible implementation methods here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A sustainable consumption behavior simulation system fusing large language models with multi-agent, comprising: An external semantic generation module and an internal behavior evolution module; The external semantic generation module is configured to: construct a plurality of large language model agents of multiple types, each large language model agent calls a corresponding external knowledge base based on a current state parameter generated by the internal behavior evolution module, dynamically generates unstructured semantic intervention information, and converts the semantic intervention information into standardized numerical signals that can be parsed by the internal behavior evolution module through a preset mapping function; The internal behavior evolution module is configured to: construct a plurality of consumer agents and a supervision agent deployed in an interaction network; At each simulation step, the standardized numerical signals are received; based on an optimized social influence model, the plurality of consumer agents are driven to perform cognitive interaction within a network neighbor range thereof, and a sustainable consumption cognitive value of each consumer agent is calculated and updated; For each consumer agent, the updated sustainable consumption cognitive value, product feature signals and scene context signals parsed from external signals are assembled into an input vector, which is input into a behavior decision neural network embedded therein, and a sustainable consumption behavior tendency value of the consumer agent at a next simulation step is output through forward propagation and nonlinear mapping of the behavior decision neural network. The supervision agent monitors the latest states of all consumer agents in real time, calculates macro indicators at a group level, encapsulates the macro indicators into a new state vector, and feeds back the new state vector to the external semantic generation module in real time through a preset data interface as an input for generating next-round semantic intervention information, thereby forming a closed-loop simulation loop.

2. The sustainable consumption behavior simulation system of claim 1, wherein, Based on the optimized social influence model, the plurality of consumer agents are driven to perform cognitive interaction within a network neighbor range thereof, and a sustainable consumption cognitive value of each consumer agent is calculated and updated, specifically including: Based on a Watts-Strogatz small-world network model, an interaction topological environment is constructed for the plurality of consumer agents, a real social network simulation scene of different connection density and randomness is adapted by adjusting an average degree K and a reconnection probability β; the average degree K controls an initial connection number of each agent and is used to adjust an overall connection density of the network; the reconnection probability β controls a reconstruction probability of a long-range connection of connection reconstruction randomness and is used to introduce random connection while maintaining local clustering characteristics; in the interaction topological environment, cognitive interaction among the plurality of consumer agents is limited within a direct network neighbor range thereof; Based on the interaction topological environment, at each simulation step, for any agent with any of its neighbor agents , based on the difference of their sustainable consumption awareness values, compute the social similarity between them : ， wherein, is the sustainable consumption awareness value of the agent and the neighbor agent is the social similarity between the agent and is the sustainable consumption awareness value of the agent and the neighbor agent , respectively, and is a real number continuously varying in the interval [0, 1]; based on the social similarity and static attributes of the agent , and the acceptability of the influence on the neighbor agent is verified. updating the agent based on the neighbor agent that passed the check sustainable consumption awareness value.

3. The sustainable consumption behavior simulation system of claim 2, wherein, based on the social similarity and static attributes of the agent performing acceptability checking on the influence of the neighbor agent, specifically comprising: According to the preset similarity threshold , the interaction type is preliminarily distinguished: when , it is determined that the agents i and j are a cognitive similar group; and when , it is determined as a cognitive difference group. Based on the cognitive similar group, for each consumer agent Set an internal acceptance threshold By comparing the absolute value of the cognitive difference between the two With the internal acceptance threshold Perform acceptance verification: If , then the neighbor agent's cognition is determined to be within the acceptable range of the agent's cognition, and the influence is accepted. Otherwise, it is determined that the cognitive difference of the neighbor agent exceeds the acceptable range of the agent , and the influence is rejected. wherein the internal acceptance threshold by the agent The static properties of susceptibility and trustworthiness are calculated by the agent itself.

4. The sustainable consumption behavior simulation system of claim 2, wherein, The updating agent The formula for calculating the sustainable consumption awareness value of the user is as follows: wherein, is an agent at a simulation time a sustainable consumption awareness value, is an agent at a next simulation time an updated sustainable consumption awareness value; is a preset learning rate parameter, used to control the overall rate of awareness update; is an agent a connection weight between the agent and its neighbor agent , used to represent the importance of the connection to the awareness influence; is an agent a neighbor set of the agent in the interaction topology environment; The connection weight Based on intelligent agents With any of its neighboring intelligent agents Social similarity calculated from the current perception value of sustainable consumption And / or the inherent properties of the connecting edge pre-assigned in the small-world network topology are dynamically determined.

5. The sustainable consumption behavior simulation system of claim 4, wherein, The updated agent The formula for calculating the perceived value of sustainable consumption also includes a social stress item. It is used to quantify the conformity pressure an individual experiences due to their perception of the behavioral choices of the majority in their social circle; its intensity is related to the intelligence agent. The proportion of sustainable consumers among the direct neighbor agents is positively correlated, wherein the sustainable consumer is defined as a neighbor agent whose sustainable consumption behavior tendency value exceeds a preset behavior adoption threshold; Introducing the social pressure term Afterwards, the agent is updated The formula for the calculation of the sustainable consumption awareness value is extended to: wherein, is the social pressure influence coefficient.

6. The sustainable consumption behavior simulation system of claim 1, wherein, The behavior decision neural network is a BP neural network, and a structure of the BP neural network includes: An input layer including three neurons receiving: a sustainable consumption cognitive value of the agent at a current simulation time, a product sustainability intensity signal representing an environmental property of a product, and a product supply facility adaptability signal representing purchase convenience and facility support; A hidden layer configured with 5 to 64 neurons and adopting a ReLU activation function for nonlinear transformation; An output layer including one neuron, which adopts a Sigmoid activation function to map a weighted sum of the neuron to a [0, 1] interval and output a sustainable consumption behavior tendency value at a next simulation time.

7. The sustainable consumption behavior simulation system of claim 1, wherein, The supervisory agent monitors the latest state of all consumer agents in real time, calculates macro indicators at the group level, encapsulates them as a new state vector, and feeds them back to the external semantic generation module in real time through a preset data interface, specifically including: At each simulation step At the end of the simulation, the updated evolution states of all consumer agents are collected and aggregated in real time, and based on the updated evolution states, macro indicators at the group level are calculated, including: the proportion of sustainable consumers, the average sustainable consumption awareness of the group, and the average sustainable consumption behavior of the group. encapsulating the macroscopic indicators into a table representing the current simulation step at the end, the actual state vector of the population as a whole evolution result ; The actual state vector is transmitted through a preset data interface. Feedback is sent to the external semantic generation module to trigger the external semantic generation module in the next simulation step. The semantic intervention generation process.

8. The sustainable consumption behavior simulation system of claim 1, wherein, The semantic intervention information is converted into a standardized numerical signal that can be parsed by the internal behavior evolution module through a preset mapping function, specifically including an inward-outward mapping step and an outward-inward mapping step: In the said inside-out mapping step, receiving the current simulation step At the end, the current state vector characterizing the evolution state of the population of consumer agents fed back by the internal behavior evolution module ; based on the current state vector , through a template-based prompt word construction function , at least one class of external large language model agents is generated with semantic prompt words with targeted intervention targets ; the process is formally represented as: ; In the outward-to-inward mapping step, the semantic prompt words are... The input is fed into the corresponding large language model agent, triggering it to generate structured semantic response information. The semantic response information It includes at least one quantifiable semantic impact parameter; by a pre-set influence mapping matrix mapping the semantic response information at least one semantic influence parameter to an expected amount of influence on the consumer agent state vector : ; based on the expected influence quantity and the current state vector , the target state vector that the system is expected to achieve at the next simulation step is calculated, which is formally expressed as: ; the target state vector is the output normalized numerical signal.

9. The sustainable consumption behavior simulation system of claim 8, wherein, The influence mapping matrix The semantic influence parameter of different agents is linearly combined according to a preset weight to calculate a direct influence amount of different dimensions of the state vector, and an expression is as follows: ， wherein, represents the response value of the brand agent, represents the response value of the rule agent, represents the response value of the media agent; , , , are pre-calibrated influence coefficients.

10. The sustainable consumption behavior simulation system of claim 1, wherein, Each consumer agent includes two types of attribute information that are independent and related: one is static attribute information, which is used to represent the inherent social identity and underlying characteristics of consumers, remains stable throughout the simulation period, and does not change dynamically with the interaction process, including conformity, trust, and willingness to accept sustainable consumption; the other is dynamic attribute information, which is used to provide cognitive interaction and target evolution variables for consumption decision-making, including sustainable consumption cognition value, sustainable consumption behavior tendency value.

Images