Dynamic cognition-driven multi-subject energy differentiated transaction method and dynamic cognition-driven multi-subject energy differentiated transaction system

Abstract

The invention discloses a multi-subject energy differentiation transaction method and system driven by dynamic cognition, and relates to the field of energy internet transaction.The method comprises the steps that personalized modeling of market subject cognitive competence is achieved based on multivariate data, initial division of cognitive levels is achieved based on a clustering algorithm, and cognitive level initial labels are obtained; on the basis of a dynamic cognitive level updating mechanism of market performance and behavior evolution characteristics, obtaining a cognitive level updated tag; and determining a target transaction mode corresponding to the tag after the cognitive hierarchy is updated for the tag after the cognitive hierarchy is updated based on a differentiated transaction strategy and a cooperation mechanism of the cognitive hierarchy. In conclusion, fine identification and dynamic adaptation of the cognitive heterogeneity subject are realized, the market fairness, the operation efficiency and the system robustness are effectively improved, and key support is provided for constructing intelligent, differentiated and collaborative energy internet transaction ecology.

Description

Technical Field

[0001] This specification relates to the field of energy internet transactions, and more specifically, to a dynamic cognition-driven multi-agent energy differentiation transaction method and system. Background Art

[0002] In the existing multi-agent trading system of the Energy Internet, mainstream optimization methods are generally based on the rational assumption of homogeneous market participants. This assumption assumes that all participants (including power generation companies, electricity sales companies, aggregators, and users) possess the same information acquisition capabilities, processing depth, and decision-making complexity. Based on this assumption, the system typically adopts two approaches to scheduling optimization: a centralized scheduling model, where a central agency gathers global information and makes decisions; and a distributed strategy-solving model based on classical game theory models (such as Nash equilibrium), which simulates the optimal behavior of market participants under the premise of complete rationality.

[0003] However, in the real world, significant differences exist among various market participants in terms of cognitive abilities, technological capabilities, resource endowments, and data literacy. While some studies have attempted to describe these differences by introducing static cognitive stratification models (such as the Poisson distribution or fixed-level classification), these models are inherently static and lack the ability to dynamically update, severely limiting their adaptability and accuracy. This is particularly true in dynamic scenarios such as the dramatic fluctuations in renewable energy, dramatic changes in electricity prices, or temporary policy adjustments. Static models struggle to reflect real-time trends in participants' cognitive abilities, thus impacting the effectiveness of trading strategies.

[0004] In addition, existing solutions generally lack a refined differentiated strategy system: on the one hand, they fail to fully unleash the autonomous decision-making capabilities of high-cognitive subjects and fail to motivate them to deeply participate in market and resource optimization; on the other hand, they set excessively high operational thresholds for low-cognitive subjects, requiring them to execute complex instructions, resulting in their lack of willingness to participate and frequent operational errors, which seriously restricts the effective integration and utilization of distributed energy resources.

[0005] In this context, there is an urgent need for a systematic solution that integrates state cognitive modeling with differentiated collaboration mechanisms to achieve accurate characterization, flexible scheduling and collaborative optimization of multi-cognitive level entities, thereby promoting the evolution of the energy Internet market towards a fairer, more efficient and robust direction. Summary of the Invention

[0006] The Summary of the Invention introduces a series of simplified concepts that will be further described in the Detailed Description of the Invention. The Summary of the Invention is not intended to limit the key features and essential features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0007] In a first aspect, the present invention proposes a dynamic cognitive-driven multi-agent energy differentiation trading method, the method comprising: Based on multivariate data, personalized modeling of the cognitive capabilities of market entities is achieved, and the initial division of cognitive levels is achieved based on clustering algorithms to obtain preliminary labels of cognitive levels; The above preliminary cognitive level labels are updated based on the dynamic cognitive level update mechanism of market performance and behavioral evolution characteristics to obtain updated cognitive level labels; The target transaction mode corresponding to the updated labels at the above cognitive levels is determined based on the differentiated transaction strategy and collaboration mechanism at the cognitive levels.

[0008] In a feasible implementation, the personalized modeling of the cognitive capabilities of market entities based on multivariate data and the initial division of cognitive levels based on a clustering algorithm to obtain preliminary labels of cognitive levels include: Collect structured and unstructured data related to the cognitive capabilities of market entities through pre-set data interfaces and web crawler technology; Use a large language model to intelligently analyze and extract features from the above unstructured data to obtain quantitative feature scores; Conduct behavioral pattern mining and quantitative modeling on the aforementioned structured data, and use Fourier transform methods to conduct commentary analysis on the time series of transaction behaviors to obtain structured behavioral scores; Perform feature fusion operation and pairing on the above-mentioned feature quantitative scores and the above-mentioned structured behavioral scores to obtain the comprehensive cognitive ability score of each subject; An unsupervised algorithm is used to analyze the comprehensive cognitive ability score of each of the above-mentioned entities, and the market entities are automatically divided into several cognitive levels according to the similarity of the scores to obtain preliminary labels of the above-mentioned cognitive levels.

[0009] In a feasible implementation, the dynamic cognitive level updating mechanism based on the market performance and behavior evolution characteristics of the preliminary cognitive level labels to obtain updated cognitive level labels includes: Establish a continuous monitoring mechanism for the market environment and subject behavior to determine whether to initiate a dynamic reassessment of the target subject's multi-dimensional cognitive ability based on the preliminary labels of the above cognitive levels; Once the launch is confirmed, obtain multi-dimensional assessment results; Based on the above multi-dimensional evaluation results, the built-in cognitive level adjustment rules are called to preliminarily update the cognitive level label of the subject to obtain the preliminary updated cognitive level label; A robust control mechanism for cognitive level updating is introduced into the preliminary updated labels of the cognitive level to obtain the updated labels of the cognitive level.

[0010] In a feasible implementation, the above-mentioned continuous monitoring mechanism for market environment and subject behavior includes a market-level monitoring module and a subject behavior-level monitoring module. The market-level monitoring module is used to collect and analyze key market indicators in real time, including power market price fluctuations, power load changes, renewable energy output, grid congestion status, and changes in policies and regulations, to identify market disruption events and generate environmental risk tags. The above-mentioned subject behavior layer monitoring module is used to record the trading behavior trajectory of each market subject, including the execution of trading instructions, strategy adjustment rhythm, response time delay and profit deviation, and extract subject behavior feature labels based on the rule engine or learning model.

[0011] In a feasible embodiment, whether to initiate the multi-dimensional dynamic reassessment of the target subject's cognitive ability is determined based on any one or more triggering mechanisms selected from the group consisting of an event triggering mechanism, a periodic triggering mechanism, and an abnormal performance triggering mechanism; The above multi-dimensional evaluation results include response effectiveness evaluation results, decision-making accuracy evaluation results, learning and adaptability evaluation results, and behavioral stability evaluation results.

[0012] In a feasible implementation, the aforementioned built-in cognitive level adjustment rules include upgrade rules, downgrade rules, and maintenance rules; The robust control mechanisms for updating the above-mentioned cognitive levels include a time window smoothing mechanism and a historical behavior pattern reference mechanism.

[0013] In a feasible implementation, when the above cognitive level is updated and the label is high cognitive subject, the system performs the following operations to construct its target transaction model: Opening the advanced strategy interface module to the aforementioned high-cognition entities, where the advanced strategy interface module provides multi-dimensional API services including real-time electricity price forecasting, regional and time-period supply and demand modeling, counterparty behavior simulation, and risk hedging mechanisms; Calling the intelligent prediction and decision engine based on the large language model to conduct joint modeling and analysis of structured behavioral data and unstructured text data, performing time series modeling operations of sliding window block and multi-head attention mechanism, and generating the first target trading pattern and corresponding strategy return forecast of the above-mentioned high-cognition subject; If it is detected that the transaction behavior triggers the risk warning conditions set by the system, the soft constraint intervention mechanism will be activated to dynamically adjust the above-mentioned first target transaction mode; If the warning conditions are not triggered, the above-mentioned first target trading mode will be maintained and executed autonomously.

[0014] In a feasible implementation, when the above cognitive level is updated and the label is a medium cognitive subject, the system performs the following operations to construct its target transaction model: Connecting the aforementioned intermediate cognitive subject to an intelligent trading assistance system and generating trading recommendations based on the intermediate cognitive subject's cognitive level labels, real-time market data, and historical trading behavior, wherein the aforementioned trading recommendations include a recommended trading time window, price range, trading volume, and a market situation briefing; Calling the embedded simplified large language model trading prediction framework, performing block statistics and sequence normalization on the behavior sequence and context environment, and generating executable short-term trading operation plans; A second target transaction model is generated based on the transaction suggestion and the transaction operation plan.

[0015] In a feasible implementation, when the above cognitive level is updated and the label is low-cognitive subject, the system performs the following operations to construct its target transaction model: Guide these low-aware entities into standardized transaction paths, supporting them in choosing preset fixed-price packages or system-recommended long-term power purchase and sales agreements; For entities with basic setting capabilities, an automatic agent trading module is provided to allow them to set basic parameters, where the above basic parameters include the maximum power purchase cost and the expected transaction power; The micro-energy resources of the above-mentioned low-cognition entities are connected to the resource aggregation platform, where the above-mentioned resource aggregation platform is responsible for internal resource optimization, external centralized bidding, transaction scheduling and profit distribution.

[0016] In a second aspect, the present invention proposes a dynamic cognitive-driven multi-agent energy differentiation trading system, comprising: The first acquisition unit is used to implement personalized modeling of the cognitive capabilities of market entities based on multivariate data, and to implement initial division of cognitive levels based on clustering algorithms to obtain preliminary labels of cognitive levels; The second acquisition unit is used to update the above-mentioned preliminary cognitive level labels based on the dynamic cognitive level update mechanism of market performance and behavior evolution characteristics to obtain updated cognitive level labels; The determination unit is configured to determine a target transaction mode corresponding to the label after the above-mentioned cognitive level update based on the differentiated transaction strategy and cooperation mechanism of the cognitive level.

[0017] In summary, the dynamic, cognitive-driven, multi-agent differentiated energy trading method proposed in this paper constructs a personalized model of market agent cognitive capabilities by integrating structured behavioral data with unstructured text data. This overcomes the inability of traditional models to reflect agent differences and significantly improves the accuracy and discernibility of cognitive ability assessment. A two-tiered monitoring mechanism, combining market performance with agent behavioral evolutionary characteristics, dynamically adjusts cognitive level labels, effectively reflecting changes in agent capabilities as the environment and experience evolve, enhancing the system's adaptability and real-time performance. Differentiated strategy execution mechanisms are matched to different cognitive levels. High-cognitive agents can flexibly construct complex strategies, medium-cognitive agents can leverage intelligent decision-making assistance, and low-cognitive agents can achieve low-threshold access through fixed packages and proxy mechanisms, ensuring fair participation and efficient trading for all types of agents. Through cognitive stratification and mechanism differentiation, the risk of low-cognitive agents disrupting system stability is reduced, while the strategic contributions of high-cognitive agents are enhanced, achieving a balance between trading freedom and system coordination. Through virtual power plants and resource aggregation platforms, the distributed energy resources of medium- and low-cognitive agents are integrated, improving resource integration efficiency and trading scale, and promoting the deep participation of distributed energy in the market. In summary, the present invention achieves refined identification and dynamic adaptation of cognitive heterogeneous entities, effectively improves market fairness, operational efficiency and system robustness, and provides key support for building an intelligent, differentiated and collaborative energy Internet trading ecosystem.

[0018] The dynamic cognitive-driven multi-agent energy differentiation trading method proposed in the present invention, and other advantages, objectives and features of the present invention will be reflected in part through the following description, and in part will be understood by technical personnel in this field through research and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present description. The same reference symbols are used throughout the drawings to represent the same components. In the drawings: Figure 1 A schematic diagram of the process of a dynamic cognition-driven multi-agent energy differentiation trading method provided by an embodiment of the present invention; Figure 2 A schematic diagram of the process of initial division of cognitive levels of market entities provided by an embodiment of the present invention; Figure 3 A schematic diagram of a process for a dynamic cognitive level update mechanism provided by an embodiment of the present invention; Figure 4 A schematic diagram of a differentiated transaction strategy and collaboration mechanism based on cognitive levels provided by an embodiment of the present invention; Figure 5 A schematic diagram of a trading strategy generation and control system based on a large model provided by an embodiment of the present invention; Figure 6 A schematic diagram of the structure of a dynamic cognition-driven multi-agent energy differentiation trading system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments.

[0021] See also Figure 1 , which is a schematic diagram of the process of a dynamic cognitive-driven multi-agent energy differentiation trading method provided by an embodiment of the present invention, which may specifically include: S110. Implement personalized modeling of market entities’ cognitive capabilities based on multivariate data, and implement initial division of cognitive levels based on clustering algorithms to obtain preliminary cognitive level labels; S120, applying the preliminary cognitive level labels to a dynamic cognitive level update mechanism based on market performance and behavioral evolution characteristics to obtain updated cognitive level labels; S130: Determine a target transaction model corresponding to the updated label at the cognitive level based on the differentiated transaction strategy and collaboration mechanism at the cognitive level.

[0022] Exemplarily, step S110 is performed, which involves personalized modeling of market participants' cognitive capabilities based on multivariate data and initial categorization of cognitive hierarchies using a clustering algorithm. This process utilizes a pre-set data acquisition interface and web crawler technology to acquire information related to market participants' cognitive capabilities from multiple data sources. Data types include structured data (such as historical trading frequency, transaction accuracy, and trading strategy complexity) and unstructured data (such as market analysis reports, operational logs, and policy responses published by companies). For unstructured data, the system utilizes pre-trained or fine-tuned large language models (such as DeepSeek or ChatGPT) to perform semantic understanding and feature extraction, converting subjective expressions into quantitative feature indicators. For structured data, the system employs behavioral analysis and Fourier transform methods to model trading behavior sequences, extracting their response patterns and rhythmic characteristics in the frequency domain, and calculating their behavioral stability, proactiveness, and strategic adaptability. Finally, the system normalizes and integrates all features to form a comprehensive cognitive ability score. Using unsupervised clustering algorithms (such as K-Means or DBSCAN), the system automatically categorizes participants into several cognitive hierarchies, forming preliminary cognitive labels.

[0023] In step S120, the system dynamically updates the preliminary cognitive hierarchy labels based on market performance and behavioral evolution. This process incorporates a continuous dual-layer monitoring mechanism for both the market environment and agent behavior, analyzing macroeconomic variables such as electricity price fluctuations, renewable energy output changes, grid congestion status, and policy adjustments in real time. It also dynamically records indicators such as transaction response time, strategy adjustment frequency, and return deviation. If a particular agent's behavior within a certain time window is detected to be inconsistent with the current cognitive hierarchy (e.g., a significant increase in strategy maturity), the system initiates a dynamic cognitive reassessment mechanism. This mechanism feeds recent behavioral data into the updated model and assesses multiple dimensions, including response effectiveness, strategy accuracy, and learning ability. If the score deviates significantly from the original hierarchy, the system adjusts the agent's cognitive hierarchy according to pre-set rules (e.g., criteria for level promotion or demotion) to generate a new, updated cognitive hierarchy label. To avoid frequent fluctuations, the system also incorporates time smoothing and historical reference mechanisms to ensure the robustness of the update results.

[0024] In step S130, the system matches differentiated trading strategies and collaboration mechanisms based on the updated labels at the cognitive level to determine the target trading model. For high-cognition entities, the system provides advanced APIs and autonomous strategy-setting capabilities, offering support tools such as real-time electricity price forecasts, supply and demand modeling, counterparty behavior simulation, and risk hedging models. It also deploys a prediction engine based on a large language model to output personalized trading windows and operational recommendations, enabling them to freely submit advanced orders such as limit orders and arbitrage orders. For medium-cognition entities, the system deploys an intelligent trading assistance system that automatically recommends trading intervals, price ranges, and response strategies based on historical behavior and current market conditions. It also connects medium-cognition entities to the virtual power plant platform, enabling VPP-based aggregate scheduling and centralized bidding. For low-cognition entities, the system simplifies the trading path, offering only fixed packages or automated proxy trading channels. It also connects their micro-resources to a unified resource aggregation platform, enabling them to automatically participate in the market, achieving intelligent matching throughout the entire process, from cognitive perception to strategy adaptation. Ultimately, the system executes the target trading model through a unified scheduling platform, ensuring both strategic flexibility and system synergy.

[0025] In summary, this embodiment proposes a dynamic, cognitive-driven, multi-agent energy differentiated trading method. This method overcomes the limitations of the existing art of homogeneous modeling of market entities and static cognitive assumptions, enabling refined modeling and dynamic adaptation of cognitively heterogeneous entities in complex market environments, resulting in significant benefits. First, the present invention constructs a personalized model characterizing the cognitive capabilities of market entities through the joint modeling of structured behavioral data and unstructured text data. The system utilizes a large language model to perform semantic understanding and feature extraction on text data (such as market analysis reports and operation logs). It also combines Fourier modeling and behavioral analysis of structured indicators such as transaction frequency and strategy complexity to form a comprehensive set of cognitive capability indicators. After standardization, the system performs preliminary labeling of cognitive hierarchies using unsupervised clustering algorithms (such as K-Means), significantly improving the accuracy and discrimination of cognitive capability assessments. Second, to overcome the limited adaptability of static modeling in dynamic markets, the system introduces a dynamic update mechanism for cognitive hierarchies based on market performance and behavioral evolution. This mechanism monitors electricity price fluctuations, renewable energy output, grid congestion, and policy changes in real time at the market level. At the behavioral level, it continuously records indicators such as transaction response time, strategy adjustment frequency, and return deviation, forming a two-tiered monitoring system. When a significant deviation in a participant's cognitive behavior is identified, the system initiates a cognitive reassessment process, reassessing the participant based on multiple metrics such as response efficiency, strategy accuracy, learning ability, and behavioral stability. This process then integrates historical performance and a time window mechanism to smooth decisions and dynamically adjust the participant's cognitive hierarchy label, ensuring that cognitive hierarchy remains in sync with actual capabilities. Furthermore, based on the updated cognitive hierarchy labels, the system assigns differentiated trading models and collaboration mechanisms to participants with different capabilities. For participants with higher cognitive capabilities, the system provides an open advanced strategy interface, empowering them to independently construct complex trading strategies. It supports access to real-time price forecasting models, supply and demand modeling tools, counterparty behavior simulators, and risk hedging modules. A trading prediction engine driven by a large language model provides personalized strategy recommendations, meeting their demands for refined operations and optimized returns. For medium-level cognitive entities, the system deploys an intelligent trading assistance system that generates actionable trading recommendations based on behavioral trajectories and market conditions. It also provides intraday operation plans through a simplified trading prediction module. At the same time, its resources are incorporated into the virtual power plant platform, where they are centrally optimized and bid for by the central dispatching system, improving their participation efficiency and reducing trading risks. For low-level cognitive entities, the system provides fixed packages and automatic agent trading channels, allowing them to set basic trading parameters. The system then automatically executes operations on their behalf, minimizing their cognitive burden and participation threshold. At the same time, their micro-energy resources are connected to the resource aggregation platform, which centrally dispatches and distributes revenue, ensuring the effective utilization of their resources and their market accessibility.Overall, this invention achieves key breakthroughs in cognitive modeling accuracy, dynamic cognitive level updating, personalized trading strategy adaptation, and overall system collaborative optimization. It not only improves market fairness and efficiency, and enhances the system's stability and robustness to complex perturbations, but also significantly enhances the integration and utilization efficiency of distributed energy resources. This provides a solid technical foundation and practical path for building a next-generation energy internet collaborative trading system with cognitive intelligence.

[0026] In one possible implementation, Figure 2 As shown, Figure 2 A schematic diagram of the process of initial classification of cognitive levels of market entities provided by an embodiment of the present invention. Step S110 implements personalized modeling of market entity cognitive capabilities based on multivariate data and implements initial classification of cognitive levels based on a clustering algorithm to obtain preliminary cognitive level labels, including: S1101. Collect structured and unstructured data related to the cognitive capabilities of market entities through pre-set data interfaces and web crawler technology; S1102. Using a large language model to perform intelligent parsing and feature extraction on the unstructured data to obtain a quantitative feature score; S1103. Conduct behavioral pattern mining and quantitative modeling on the structured data, and use Fourier transform method to conduct commentary analysis on the time series of transaction behaviors to obtain structured behavioral scores; S1104. Perform a feature fusion operation on the aforementioned feature quantitative scores and the aforementioned structured behavior scores to obtain a comprehensive cognitive ability score for each subject; S1105. Use an unsupervised algorithm to analyze the comprehensive cognitive ability score of each of the above entities, and automatically divide the market entities into several cognitive levels based on the similarity of the scores to obtain preliminary labels for the above cognitive levels.

[0027] For example, steps S1101 to S1105 are used to complete the personalized modeling of market entities' cognitive capabilities and the initial division of cognitive levels, breaking the limitations of traditional static cognitive assumption models (such as fixed Poisson distribution modeling). The system introduces multi-source heterogeneous data fusion and intelligent analysis mechanisms to achieve in-depth identification of cognitive heterogeneity, specifically including the following continuous processing flow: First, in step S1101, the system uses pre-set data interfaces and web crawler technology to comprehensively collect structured and unstructured data related to the cognitive capabilities of market entities. Unstructured data primarily comes from official corporate websites, annual reports, industry research reports, authoritative media, and social media platforms, reflecting the entity's soft power, such as management capabilities, information technology development level, and strategic vision. Structured data, primarily from energy trading platforms, includes historical transaction behavior data and market environment variables such as quotation frequency, transaction records, response delays, grid load, price fluctuations, and renewable energy output. Through this multi-dimensional data collection mechanism, the system is able to comprehensively acquire cognitive-related information from both the "explicit behavior" and "implicit capabilities" dimensions.

[0028] Next, in step S1102, the system introduces a large language model (such as DeepSeek-R1) to perform semantic understanding and structural analysis on the collected unstructured text data. The system uses natural language processing (NLP) technology to extract key information, such as the educational background, strategic role, and cross-industry experience of corporate executives. The extracted information is then quantitatively scored based on built-in evaluation rules. This scoring system encompasses multiple dimensions, including strategic decision-making ability, risk mitigation, information technology and AI application capabilities, and market influence. These metrics are then generated to provide soft information support for cognitive modeling.

[0029] Subsequently, in step S1103, the system performs behavioral pattern mining and frequency-domain modeling on the structured historical trading behavior data. By constructing time series data containing variables such as price fluctuation responses, trading volume changes, and strategy adjustments, the system uses a Fourier transform to map this data from the time domain to the frequency domain, revealing the periodicity and reaction speed of the behavior. The extracted frequency-domain features, such as dominant frequency, energy distribution, and high-frequency components, can be used to quantify key capability indicators such as "high-frequency response agility," "behavioral stability," and "ability to adapt to market disturbances." Based on this, a structured behavioral score is generated for each entity.

[0030] In step S1104, the system performs a feature fusion of the cognitive ability scores generated by unstructured data analysis and the behavioral scores generated by structured data. This fusion process integrates the various scoring dimensions using a weighted average or multi-layer weighting function. The fusion weights can be set based on the completeness of the feature data, model confidence, or domain expert experience. The fusion result is the "comprehensive cognitive ability score" for each market entity, which is used to uniformly measure its overall performance in terms of strategic capabilities, behavioral execution, and market adaptability.

[0031] Finally, in step S1105, the system performs unsupervised clustering based on the comprehensive cognitive ability scores. The system analyzes the subject set using K-means, hierarchical clustering, or other similarity-driven clustering algorithms, dividing the subjects into several cognitive levels based on the similarity of their cognitive scores. Each cluster represents a group of subjects with similar cognitive characteristics. The resulting classification serves as a preliminary label for the cognitive level, providing a basis for subsequent cognitive dynamics updates and differentiated trading strategy configuration.

[0032] Through this continuous process, the system achieves dynamic modeling and hierarchical identification of cognitive capabilities, building an integrated cognitive analysis framework from data acquisition, feature extraction, frequency domain analysis, to intelligent clustering. The large language model improves the efficiency of understanding unstructured data, the Fourier transform introduces a time-sensitive characterization of market behavior, and feature fusion and cluster modeling ensure the objectivity and operability of evaluation results, forming the core technical foundation for the initial delineation of cognitive hierarchies.

[0033] In one possible implementation, Figure 3 As shown, Figure 3 This is a flow chart of a dynamic cognitive hierarchy update mechanism provided by an embodiment of the present invention. Step S120 above uses the dynamic cognitive hierarchy update mechanism based on market performance and behavioral evolution characteristics to obtain updated cognitive hierarchy labels, including: S1201. Establish a continuous monitoring mechanism for the market environment and subject behavior to determine whether to initiate a dynamic reassessment of the target subject's multi-dimensional cognitive ability based on the aforementioned preliminary cognitive level labels; S1202. When the startup is confirmed, obtain multi-dimensional evaluation results; S1203: Based on the above multi-dimensional evaluation results, call the built-in cognitive level adjustment rules to preliminarily update the cognitive level label of the subject to obtain a preliminary updated cognitive level label; A robust control mechanism for cognitive level updating is introduced into the preliminary updated labels of the cognitive level to obtain the updated labels of the cognitive level.

[0034] In a feasible implementation, the above-mentioned continuous monitoring mechanism for market environment and subject behavior includes a market-level monitoring module and a subject behavior-level monitoring module. The market-level monitoring module is used to collect and analyze key market indicators in real time, including power market price fluctuations, power load changes, renewable energy output, grid congestion status, and changes in policies and regulations, to identify market disruption events and generate environmental risk tags. The above-mentioned subject behavior layer monitoring module is used to record the trading behavior trajectory of each market subject, including the execution of trading instructions, strategy adjustment rhythm, response time delay and profit deviation, and extract subject behavior feature labels based on the rule engine or learning model.

[0035] In a feasible embodiment, whether to initiate the multi-dimensional dynamic reassessment of the target subject's cognitive ability is determined based on any one or more triggering mechanisms selected from the group consisting of an event triggering mechanism, a periodic triggering mechanism, and an abnormal performance triggering mechanism; The above multi-dimensional evaluation results include response effectiveness evaluation results, decision-making accuracy evaluation results, learning and adaptability evaluation results, and behavioral stability evaluation results.

[0036] In a feasible implementation, the aforementioned built-in cognitive level adjustment rules include upgrade rules, downgrade rules, and maintenance rules; The robust control mechanisms for updating the above-mentioned cognitive levels include a time window smoothing mechanism and a historical behavior pattern reference mechanism.

[0037] Exemplarily, step S120 is used to dynamically update the cognitive hierarchy of market entities, breaking through the limitations of traditional cognitive models, which only assess entity capabilities during the initialization phase and then maintain static configuration over the long term. This mechanism ensures that cognitive labels always match the entity's true capabilities through continuous monitoring of the market environment and entity behavior, multi-dimensional capability assessment, rule-driven hierarchy adjustment, and robust control measures.

[0038] Specifically, step S120 includes the following processing flow: In step S1201, the system first establishes a continuous monitoring mechanism for the market environment and subject behavior, serving as the foundation for dynamic updates at the cognitive level. The market-level monitoring module continuously collects key indicators, including electricity market price fluctuations, power load changes, renewable energy output such as wind and solar power, grid congestion, and changes in policies and regulations. This is used to identify potential market disruptions and generate corresponding environmental risk tags. The subject-level monitoring module focuses on transaction details, including the execution status of trade orders (e.g., whether or not they were executed), the pace of quotation strategy adjustments, response latency, and revenue fluctuations. The system analyzes this data using a rules engine or learning model, extracting behavioral feature tags to support subsequent capability assessment and adjustments.

[0039] Next, in step S1202, the system determines whether to perform a cognitive re-evaluation operation on the target entity based on three types of trigger mechanisms: first, the event trigger mechanism, when the market fluctuates violently (such as prices breaking through upper and lower limits, abnormal load growth) or the entity participates in major trading behavior, the evaluation is immediately initiated; second, the cycle trigger mechanism, the system regularly evaluates the cognitive status of each entity according to the preset evaluation cycle (such as daily or monthly); third, the performance abnormality trigger mechanism, when the entity continuously exhibits abnormal behaviors such as unfulfilled transactions, strategy errors or returns lower than expected, reaching the set threshold, the system actively initiates the re-evaluation process.

[0040] In step S1203, once the system determines to initiate the re-evaluation process, it evaluates the subject's cognitive ability based on four core dimensions: (1) Response Effectiveness: measuring the speed and effectiveness of the subject's decision-making response to market disturbances; (2) Decision Accuracy: analyzing the degree of deviation between its trading results and the market's optimal solution; (3) Learning and Adaptability: identifying whether the subject has optimized its strategy through experience, such as the degree of improvement in its behavior in similar situations in the past; and (4) Behavioral Stability: observing the consistency of its performance in multi-stage market situations to prevent occasional anomalies from affecting the evaluation results. The system comprehensively calculates these evaluation indicators to form a multi-dimensional evaluation result.

[0041] Based on the above evaluation results, the system calls the built-in cognitive level adjustment rules to perform a preliminary label update. This rule set includes: (1) Upgrade rules: When the subject continues to perform well, responds quickly, and has an adaptability trend, its cognitive level is adjusted upward; (2) Downgrade rules: If the subject responds slowly, deviates significantly, and has no optimization trend for multiple consecutive cycles, the cognitive level is adjusted downward; (3) Maintain rules: When there is no significant change or slight fluctuation, the current cognitive label remains unchanged. This type of rule-based decision logic can improve the system's ability to distinguish between different types of subjects.

[0042] To enhance the assessment's resistance to volatility, the system introduces a robust control mechanism for cognitive level updates after the initial label update to reconfirm the label update results. This mechanism includes two core designs: (1) Time window smoothing mechanism: Before actually executing the label update, the system sets a time lookback window (such as the last month or several complete trading cycles). Only when the performance within this window deviates significantly and persistently from the current level will the label change be finally confirmed, avoiding frequent adjustments caused by short-term incidental factors; (2) Historical behavior pattern reference mechanism: While analyzing the current behavior, the system also looks back at the subject's past long-term trading behavior to ensure the continuity, stability, and rationality of the cognitive level adjustment decision, thereby reducing the misjudgment rate and overfitting risk.

[0043] Finally, after confirming the updated cognitive label, the system pushes the label to the differentiated trading strategy module and resource collaboration module, providing a weight basis for the subsequent opening of personalized trading interfaces, auxiliary decision-making tool configuration, resource aggregation and incentive allocation strategies, thereby realizing real-time cognitive hierarchical management and dynamic capability shaping of market entities.

[0044] In summary, this embodiment, through the coordinated operation of multiple submodules in step S120, implements a dynamic cognitive tag update mechanism that responds to market and behavioral changes. Its technical advantage lies in its real-time perception, flexible stratification, and stable assessment of subject cognitive capabilities, providing a sustainable and intelligent decision-making support foundation for differentiated multi-subject participation in the energy trading market.

[0045] After completing the identification and dynamic update of the cognitive level, the system implements differentiated transaction participation methods and resource scheduling strategies based on the cognitive ability labels of each market entity. This stage aims to stimulate the potential of high-cognitive entities, reduce the complexity of medium-cognitive entities, and protect the basic rights and interests of low-cognitive entities through matching strategy design, while at the same time using collaborative mechanisms (such as virtual power plants and resource aggregation platforms) to achieve overall market efficiency improvements. Figure 4 As shown, Figure 4 This is a schematic diagram of a differentiated transaction strategy and collaboration mechanism based on cognitive levels provided by an embodiment of the present invention. Specific strategies are divided into the following three categories: In a feasible implementation, when the above cognitive level is updated and the label is high cognitive subject, the system performs the following operations to construct its target transaction model: S210. Opening an advanced strategy interface module to the aforementioned high-cognition subject, wherein the advanced strategy interface module provides multi-dimensional API services including real-time electricity price forecasting, regional and time-period supply and demand modeling, counterparty behavior simulation, and risk hedging mechanisms; S220: Invoke an intelligent prediction and decision engine based on a large language model to conduct joint modeling and analysis of structured behavioral data and unstructured text data, perform time series modeling operations using sliding window segmentation and a multi-head attention mechanism, and generate a first target trading pattern and corresponding strategy return forecast for the aforementioned high-cognition subject. S230: If it is detected that the transaction behavior triggers the risk warning condition set by the system, the soft constraint intervention mechanism is activated to dynamically adjust the first target transaction mode; S240: If the warning condition is not triggered, the first target transaction mode is maintained for autonomous execution.

[0046] For example, when the cognitive level is updated and the label is identified as a "high cognitive subject", the system identifies the subject as a core participant with high strategic planning capabilities, information processing capabilities and market behavior understanding capabilities. Therefore, the system builds a highly autonomous and intelligent target trading model for it to give full play to its cognitive advantages and promote market efficiency.

[0047] Specifically, the system performs the following operations: In step S210, the system first opens the advanced strategy interface module to the high-cognition subject. This module integrates multiple service APIs for building advanced trading strategies, including but not limited to: (1) a real-time electricity price forecast module, which provides refined, regional, and time-based electricity price forecasts based on historical price trends and current market conditions; (2) a supply and demand modeling engine, which analyzes energy supply and demand relationships in different geographical regions and time scales, and assists in strategy formulation and spatiotemporal arbitrage judgment; (3) a competitor behavior simulator, which models the historical trading behavior of other market players and predicts potential trading reactions and quotation behaviors; (4) a risk assessment and hedging module, which quantifies the risks of price fluctuations, capacity bottlenecks, or policy changes that may be faced during the execution of the subject's strategy and provides corresponding hedging solutions. Through this module, the high-cognition subject can freely call the above interfaces and independently construct complex trading combinations and profit maximization paths.

[0048] In step S220, to further enhance the decision-making intelligence of the highly cognitive agent, the system invokes an intelligent predictive decision engine built on a large language model (such as DeepSeek). This engine supports joint modeling of structured behavioral data (such as electricity price series, load time series, and energy storage status) and unstructured data (such as policy announcements, grid notices, and news and public opinion). The model uses a sliding window partitioning mechanism to perform local, fine-grained slicing of time series data and incorporates a multi-head attention mechanism to perform deep interactive modeling of internal and external features of the sequence, effectively extracting key strategic signals across multiple spatiotemporal dimensions. The output includes the highly cognitive agent's optimal trading operation sequence, recommended execution window, and strategy return forecast for the current and future trading periods, forming its primary target trading model.

[0049] In step S230, the system deploys a trading behavior monitoring and risk control module to track and analyze the target trading patterns executed by high-cognition entities in real time. If the system detects that their trading behavior triggers risk warning conditions, such as frequent submission of extreme bids, strategies that disrupt market equilibrium, or the emergence of systemic risk trends associated with high concentration, the system automatically activates a soft constraint intervention mechanism. This mechanism does not directly veto the entity's actions, but rather guides the entity's strategy back to rationality and maintains overall market coordination by limiting trading frequency, temporarily increasing bidding transparency, and providing notifications about the entity's strategy risk level.

[0050] In step S240, if the system detects no early warning risk conditions, it assumes that the target trading model is within a controllable risk range and allows it to execute autonomously throughout the entire process. This empowerment mechanism ensures that highly cognitive entities can fully utilize their cognitive and trading capabilities without violating system boundary conditions, achieving synergy between strategic innovation and market efficiency improvement.

[0051] In summary, this embodiment, through steps S210-S240, constructs a comprehensive trading mechanism for highly cognitive market participants that is intelligent, scalable, autonomous, and risk-controlled. This mechanism not only empowers highly cognitive participants with ample operational space and modeling capabilities, but also maintains system security through decision-making assistance using a large language model and soft constraint mechanisms. This provides effective institutional and technical support for a differentiated and diversified energy trading market.

[0052] In a feasible implementation, when the above cognitive level is updated and the label is a medium cognitive subject, the system performs the following operations to construct its target transaction model: S310: Connecting the intermediate cognitive subject to the intelligent trading assistance system and generating trading suggestions based on the cognitive level labels, real-time market data, and historical trading behavior of the intermediate cognitive subject. The trading suggestions include a recommended trading time window, price range, trading volume, and a market situation briefing. S320: Invoke the embedded simplified large language model transaction prediction framework to perform block statistics and sequence normalization on the behavior sequence and context environment to generate an executable short-term transaction operation plan; S330: Generate a second target transaction model based on the transaction suggestion and the transaction operation plan.

[0053] For example, when the system identifies a market entity with an updated cognitive level and labels it as a "medium cognitive entity," it considers it to possess certain data understanding and strategic response capabilities, but still has limitations in high-dimensional modeling and complex game strategy construction. Therefore, the system provides intelligent auxiliary tools and platform collaboration mechanisms to build a highly adaptable and operational target trading model to improve market participation efficiency and trading decision-making quality.

[0054] Specifically, the system performs the following steps: In step S310, the system first connects the cognitive subject to the intelligent trading assistance system. Based on the cognitive level label of the subject, the assistance system combines its real-time market environment data (such as the current regional electricity price trend, demand load situation, bidding heat) and historical behavior trajectory (such as past transaction price preferences, transaction rate, response time), and generates structured trading suggestions through an embedded rule engine, lightweight machine learning model or game simplification deduction module. The trading suggestion includes the following core elements: (1) recommended trading time window, which is used to guide the subject to trade in the most cost-effective period; (2) recommended price range, which provides a reasonable quotation range under the current market fluctuations; (3) recommended trading volume, which combines the dispatchable resources and market capacity to recommend an appropriate order volume; (4) market situation briefing, which uses natural language to generate a summary of the key points of the current market situation to lower the information processing threshold of such subjects.

[0055] In step S320, to further enhance the operational feasibility and strategic responsiveness of the medium-cognition subject, the system calls an embedded simplified version of the large language model trading prediction framework. Unlike the complete strategy modeling system used by high-cognition subjects, this module is designed specifically for medium-capability subjects, emphasizing practicality and lightweight deployment. The system extracts trading behavior events from the subject's recent behavior sequence and environmental context, extracts the characteristics of trading activity cycles and operation density through block statistics, and uses sequence normalization to eliminate the impact of time span differences. Subsequently, the system generates directly executable short-term trading operation plans, such as intraday declaration strategies, short-term arbitrage paths, load adjustment plans, etc., to provide subjects with specific and feasible operational guidance and reduce their autonomous modeling costs.

[0056] In step S330, the system integrates the recommendations generated by the trading assistance system with the operational plans output by the predictive model to generate a second target trading model. This model preserves the autonomous decision-making space of the intermediate cognitive agent while enhancing the rationality of its strategy and the timeliness of its response through intelligent assistance. The intermediate cognitive agent can confirm, fine-tune, or directly execute the recommendations in this model, thus achieving high-quality market participation with a low barrier to entry.

[0057] In summary, this implementation helps mid-level cognitive entities bridge the gap between insufficient modeling capabilities and market complexity by providing them with a transaction support system and a streamlined intelligent forecasting framework. This mechanism not only improves their transaction efficiency and profitability, but also enhances the system's intensive management capabilities for mid-level market participants through a platform-based and structured approach, laying a solid foundation for the stable operation of a multi-level cognitive market structure.

[0058] In a feasible implementation, when the above cognitive level is updated and the label is low-cognitive subject, the system performs the following operations to construct its target transaction model: S410: Guide the aforementioned low-aware entities to enter a standardized transaction path, and support them in selecting a preset fixed-price package or a system-recommended long-term power purchase and sales agreement; S420: Provide an automatic proxy trading module to entities with basic setting capabilities, allowing them to set basic parameters, wherein the basic parameters include the maximum electricity purchase cost and the expected transaction volume; S430. Connect the micro-energy resources of the above-mentioned low-cognition entities to the resource aggregation platform, wherein the above-mentioned resource aggregation platform is responsible for internal resource optimization, external centralized bidding, transaction scheduling and profit distribution.

[0059] For example, when the system identifies a market entity with an updated cognitive level and labels it as a "low-cognitive entity," it will be included in the simplified path management model, aiming to protect its market access rights and basic trading capabilities while reducing its participation threshold and operational complexity. The process for constructing this target trading model is as follows: In step S410, the system first guides the low-cognition subject into a standardized transaction path and provides it with a set of preset, simple transaction options to meet its basic needs for electricity trading. This path mainly includes the following two typical solutions: (1) Fixed price package, the subject can directly choose the low-complexity quotation combination provided by the system, such as the basic solution of "fixed electricity purchase price + off-peak period discount", without participating in real-time bidding; (2) Long-term power purchase and sales agreement (PPA), the system automatically matches and recommends available standardized long-term contracts based on the subject's energy demand, electricity consumption cycle and historical preferences. The subject only needs to confirm to sign and execute. Through the above standardized design, the system effectively avoids high cognitive burdens such as complex quotations and timing optimization, ensuring that low-cognition subjects can participate in transactions within a controllable risk range.

[0060] In step S420, for low-cognition entities with certain basic setting capabilities, the system further provides an automatic agent trading module. This module allows the entity to set a small number of core trading parameters through a simple interface, including but not limited to: (1) the maximum acceptable power purchase cost, which is used to set the price upper limit protection; (2) the expected monthly trading electricity volume, which is used to plan a reasonable supply and demand matching target. The system automatically performs transaction matching and declaration operations based on the set parameters, current market price trends and electricity consumption trends, and the entity does not need to make trading decisions in real time. This module encapsulates the strategy logic in advance, reduces operational complexity, and ensures its basic trading efficiency and system compatibility.

[0061] In step S430, the system connects the micro-energy resources of multiple low-cognition entities to a resource aggregation platform. The platform supports centralized management of fragmented resources such as household photovoltaics, small energy storage devices, and household adjustable loads. The platform deploys a resource scheduling optimization engine and a revenue distribution module, with specific responsibilities including: (1) Internal resource optimization: dynamically coordinate the resource status of each household and reasonably configure output or load response; (2) External centralized bidding: participate in market matching with the platform as the overall trading entity to improve bargaining power; (3) Transaction scheduling and execution: generate minimalist scheduling instructions (such as "start charging" or "delay electricity use") based on market transaction results, and automatically transmit them to smart terminals for execution; (4) Revenue distribution mechanism: calculate and settle economic benefits based on unified rules or algorithm models (such as contribution degree, output accuracy) to ensure that resource owners receive reasonable returns.

[0062] Through the above implementation, the system establishes a fully managed transaction participation path for low-cognition entities, enabling them to effectively trade energy and realize resource value without having to understand the details of market mechanisms or perform complex operations. This mechanism not only improves the fairness of market participation for these entities, but also significantly enhances resource utilization efficiency. It also enhances the overall market's regulatory flexibility and transaction stability through the resource aggregation effect.

[0063] At the same time, this resource aggregation platform, together with the virtual power plant platform (VPP) supported by the medium- and low-cognitive entities mentioned above, forms a dual-pillar architecture that supports collaborative transactions between medium- and low-cognitive entities. This not only meets the diverse needs of entities with different cognitive capabilities, but also reduces the system operating load and market volatility risks through intensive and platform-based means, thereby enhancing the resilience and coordination capabilities of the entire energy market.

[0064] Through the implementation of strategies in this phase, the system's trading model design achieves a precise match between cognitive ability differences and participation complexity. This not only fully unleashes the market potential of high-cognition entities, but also effectively mitigates the risks and frustrations faced by medium- and low-cognition entities, ensuring a mutually beneficial collaboration among entities of varying capabilities within the same market system. Furthermore, the introduction of collaborative mechanisms strengthens the system's ability to integrate fragmented resources, providing a core support for promoting the deep integration of distributed energy resources and the high-proportion consumption of new energy.

[0065] like Figure 5 As shown, Figure 5 This diagram illustrates a large-model-based trading strategy generation and control system, provided by an embodiment of the present invention. This diagram illustrates the overall design of a market-based trading platform for new energy stations. By introducing intelligent analysis and prediction mechanisms driven by large language models (such as DeepSeek), this platform achieves an intelligent closed-loop process from data perception to transaction execution, addressing challenges faced by new energy stations in complex market environments, such as difficult strategy formulation, slow response times, and low transaction efficiency.

[0066] First, in terms of the transaction process support module, the platform has four major functions: "time-sharing transaction organization", "rolling matching transaction", "historical settlement review" and "expert intervention control". The time-sharing transaction organization is used to support new energy stations to participate in high-frequency electricity market quotation and matching, and meet the response needs of the spot market and intraday market. The rolling matching transaction module realizes the continuous optimization and update of transaction orders based on the real-time prediction model, improving the timeliness and competitiveness of the pending order strategy. The historical settlement review module not only provides training data for large models, but also provides a data basis for expert analysis and strategy backtesting. The expert intervention control module reserves an entry for manual intervention while the model strategy is being executed, supporting human-machine collaborative decision-making when the strategy fails or an emergency occurs.

[0067] Secondly, in its intelligent trading prediction and optimization module, the platform integrates large language models such as DeepSeek to integrate structured data (such as historical electricity prices, output curves, and load changes) with unstructured data (such as enterprise operation logs, policy announcements, and expert commentary). The system first feeds the unstructured data into a pre-trained large model for semantic parsing and keyword segmentation to extract potential trading signals. Simultaneously, the structured time series data is normalized and processed using a sliding window to capture dynamic features such as price trends and trading rhythms. Subsequently, the system achieves multimodal information fusion through sequence unification and block-wise word embedding. The model incorporates a multi-head attention mechanism and a linear prediction layer to simultaneously extract key contextual features at both the time series and semantic levels. Ultimately, it outputs the optimal trading operation prediction for a specific new energy station, including the trading time window, strategy category, and expected return path. After forward inference, the system feeds the output back to the model's backend for reinforcement learning and weight updates, continuously optimizing the quality of strategy generation.

[0068] Third, at the strategy implementation and execution level, the platform maps the model's prediction results into executable operational instructions, including limit orders, arbitrage orders, combination orders, and time-slot scheduling commands. It automatically matches the most appropriate execution mode based on the characteristics of each trading station. For trading stations with independent decision-making capabilities, the system provides a complete API call interface, supporting the integration of their customized strategy modules with external systems. For small and medium-sized trading stations that rely on the platform for proxy trading, they can directly call the prediction output results to automatically generate orders and connect them to the unified scheduling module for matching.

[0069] Finally, in the scenario adaptation and expansion module, the platform pre-configures adaptation paths for various trading models of new energy sites, including traditional wind farms, photovoltaic sites, energy storage systems, distributed energy aggregation platforms, and virtual power plants. The platform selects appropriate prediction paths and optimizers based on site type and trading model, and introduces modules such as revenue estimation, risk assessment, and production constraint judgment to verify the feasibility of strategies. Ultimately, through the integration of a "site scenario prediction model," a "strategy generation subgraph module," and a "real-time trading instruction module," the platform has implemented an intelligent decision-making system for new energy trading applicable to multiple scenarios and cognitive levels.

[0070] In summary, the platform has realized the intelligent closed loop of "data perception-cognitive modeling-transaction prediction-strategy execution-feedback optimization" for new energy stations, significantly improving the flexibility, response speed and transaction efficiency of new energy resources participating in the market, while reducing decision-making complexity and transaction risks, and providing systematic support for building a future-oriented distributed intelligent energy trading system.

[0071] Second, as Figure 6 As shown, the present invention proposes a dynamic cognitive driven multi-agent energy differentiation trading system, including: A first acquisition unit 21 is used to implement personalized modeling of the cognitive capabilities of market entities based on multivariate data, and to implement initial division of cognitive levels based on a clustering algorithm to obtain preliminary labels of cognitive levels; The second acquisition unit 22 is used to update the above-mentioned preliminary cognitive level labels based on the dynamic cognitive level update mechanism of market performance and behavior evolution characteristics to obtain updated cognitive level labels; The determining unit 23 is configured to determine a target transaction mode corresponding to the label after the cognitive level is updated based on the differentiated transaction strategy and collaboration mechanism of the cognitive level.

[0072] It is understandable that a dynamic cognition-driven multi-agent energy differentiation trading system can also execute the steps of any method described in the first aspect.

[0073] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A dynamic cognitive driven multi-agent energy differentiation trading method, characterized by: include: Based on multivariate data, personalized modeling of the cognitive capabilities of market entities is achieved, and the initial division of cognitive levels is achieved based on clustering algorithms to obtain preliminary labels of cognitive levels; The preliminary cognitive level labels are updated based on a dynamic cognitive level update mechanism based on market performance and behavioral evolution characteristics to obtain updated cognitive level labels; The target transaction mode corresponding to the updated label at the cognitive level is determined based on the differentiated transaction strategy and cooperation mechanism of the cognitive level.

2. The multi-agent energy differentiation trading method according to claim 1 is characterized in that: The personalized modeling of the cognitive ability of market entities based on multivariate data and the initial division of cognitive levels based on clustering algorithms to obtain preliminary labels of cognitive levels include: Collect structured and unstructured data related to the cognitive capabilities of market entities through pre-set data interfaces and web crawler technology; Using a large language model to intelligently analyze and extract features from the unstructured data to obtain a quantitative feature score; Conducting behavioral pattern mining and quantitative modeling on the structured data, and using Fourier transform methods to conduct commentary analysis on the time series of transaction behaviors to obtain structured behavioral scores; performing a feature fusion operation on the feature quantitative score and the structured behavioral score to obtain a comprehensive cognitive ability score for each subject; An unsupervised algorithm is used to analyze the comprehensive cognitive ability score of each subject, and the market subjects are automatically divided into several cognitive levels according to the similarity of the scores to obtain preliminary labels of the cognitive levels.

3. The multi-agent energy differentiation trading method according to claim 1 is characterized in that: The dynamic cognitive level updating mechanism based on market performance and behavioral evolution characteristics of the preliminary cognitive level labels to obtain updated cognitive level labels includes: Establish a continuous monitoring mechanism for the market environment and subject behavior to determine whether to initiate a dynamic reassessment of the target subject's multi-dimensional cognitive ability based on the initial labeling of the cognitive level; Once the launch is confirmed, obtain multi-dimensional assessment results; Based on the multi-dimensional evaluation results, calling the built-in cognitive level adjustment rules to preliminarily update the cognitive level label of the subject to obtain a preliminary updated cognitive level label; A robust control mechanism for cognitive level update is introduced into the initial cognitive level update label to obtain the cognitive level updated label.

4. The multi-agent energy differentiation trading method according to claim 3 is characterized in that: The continuous monitoring mechanism for market environment and subject behavior includes a market-level monitoring module and a subject behavior-level monitoring module. The market-level monitoring module is used to collect and analyze in real time a variety of key market indicators, including power market price fluctuations, power load changes, renewable energy output, grid congestion status, and policy and regulatory changes, to identify market disturbance events and generate environmental risk tags; The subject behavior layer monitoring module is used to record the trading behavior trajectory of each market subject, including the execution of trading instructions, strategy adjustment rhythm, response time delay and profit deviation, and extract subject behavior feature labels based on the rule engine or learning model.

5. The multi-agent energy differentiation trading method according to claim 3 is characterized in that: Whether to initiate a dynamic reassessment of the target subject's multi-dimensional cognitive ability is determined based on one or more triggering mechanisms selected from the group consisting of an event triggering mechanism, a periodic triggering mechanism, and an abnormal performance triggering mechanism; The multi-dimensional evaluation results include response effectiveness evaluation results, decision accuracy evaluation results, learning and adaptability evaluation results, and behavioral stability evaluation results.

6. The multi-agent energy differentiation trading method according to claim 3 is characterized in that: The built-in cognitive level adjustment rules include upgrade rules, de-escalation rules and maintenance rules; The robust control mechanism for updating the cognitive hierarchy includes a time window smoothing mechanism and a historical behavior pattern reference mechanism.

7. The multi-agent energy differentiation trading method according to claim 1 is characterized in that: When the cognitive level is updated and the label is a high cognitive subject, the system performs the following operations to construct its target transaction model: Opening an advanced strategy interface module to the high-cognition subject, wherein the advanced strategy interface module provides multi-dimensional API services including real-time electricity price forecasting, regional and time period supply and demand modeling, opponent behavior simulation, and risk hedging mechanism; Invoke an intelligent prediction and decision engine based on a large language model to conduct joint modeling and analysis of structured behavioral data and unstructured text data, perform time series modeling operations using sliding window segmentation and a multi-head attention mechanism, and generate the first target trading pattern and corresponding strategy return forecast of the high-cognition subject; If it is detected that the transaction behavior triggers the risk warning conditions set by the system, the soft constraint intervention mechanism is activated to dynamically adjust the first target transaction model; If the warning condition is not triggered, the first target transaction mode is maintained and executed autonomously.

8. The multi-agent energy differentiation trading method according to claim 1 is characterized in that: When the cognitive level is updated and the label is a medium cognitive subject, the system performs the following operations to construct its target transaction model: Connecting the intermediate cognitive subject to an intelligent trading assistance system and generating trading suggestions based on the intermediate cognitive subject's cognitive level tags, real-time market data, and historical trading behavior, wherein the trading suggestions include a recommended trading time window, price range, trading volume, and a market situation briefing; Calling the embedded simplified large language model trading prediction framework, performing block statistics and sequence normalization on the behavior sequence and context environment, and generating executable short-term trading operation plans; A second target transaction model is generated based on the transaction proposal and the transaction operation plan.

9. The multi-agent energy differentiation trading method according to claim 1 is characterized in that: When the cognitive level is updated and the label is low cognitive subject, the system performs the following operations to construct its target transaction model: Guiding low-aware entities into standardized transaction paths, supporting them in choosing preset fixed-price packages or system-recommended long-term power purchase and sales agreements; For entities with basic setting capabilities, an automatic agent trading module is provided to allow them to set basic parameters, where the basic parameters include the maximum power purchase cost and the expected transaction power; The micro-energy resources of the low-cognition subject are connected to a resource aggregation platform, wherein the resource aggregation platform is responsible for internal resource optimization, external centralized bidding, transaction scheduling and profit distribution.

10. A dynamic cognitive driven multi-agent energy differentiation trading system, characterized by: include: The first acquisition unit is used to implement personalized modeling of the cognitive capabilities of market entities based on multivariate data, and to implement initial division of cognitive levels based on clustering algorithms to obtain preliminary labels of cognitive levels; A second acquisition unit is configured to update the initial cognitive level labels based on a dynamic cognitive level update mechanism of market performance and behavior evolution characteristics to obtain updated cognitive level labels; The determining unit is configured to determine a target transaction mode corresponding to the label after the cognitive level is updated based on the differentiated transaction strategy and collaboration mechanism of the cognitive level.

Images