Historical social dynamic simulation system and method oriented to timing constraints

Abstract

The application discloses a kind of historical social dynamic simulation system and method for time constraint, comprising: five-dimensional time sequence knowledge graph G5 module, personality modeling module is calculated, counterfactual event schema engine, multi-agent consistency checking module, multi-engine integration and double mode optimization module, the present application constructs five-dimensional time sequence knowledge graph as objective time sequence constraint field, in combination with the potential model based on neural cognitive inspiration, realize the high-fidelity simulation of individual behavior and group evolution in historical society;Introduce counterfactual event schema template library, its parameter space is defined by historical statistical distribution, ensure that the deduction conforms to the natural law of social evolution;And design PCL consistency checking protocol, through four-dimensional measurable index guarantee the logical self-consistency of multi-agent interaction;The present application solves the problems of time sequence drift, personality distortion and deduction non-physicality in the prior art, is suitable for digital humanities research and civilization evolution modeling, with clear technical essence and verifiable effect.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and social computing, and in particular to a historical social dynamic simulation system and method oriented towards time-series constraints. Background Technology

[0002] With the convergence of digital humanities and artificial intelligence technologies, digital research in the field of historical literature has made some progress, but it has not yet solved the core coordination problem of "temporal constraints-personality fidelity-counterfactual deduction," and cannot meet the rigor and high fidelity requirements of academic-level dynamic simulation: Firstly, based on agent-based modeling... While discrete event simulation systems based on ABM (Automatic Modeling Model) can characterize individual heterogeneity, they generally suffer from a fatal flaw: a lack of temporal anchoring. Agent behavior is not subject to the mandatory constraints of the historical event temporal topology. Due to the absence of a temporal constraint field and causal correlation verification mechanism that conforms to historical logic, evolutionary paths deviate significantly from real historical trajectories, making it impossible to reconstruct the life trajectories of literati with strong temporal dependencies, such as "examination → official promotion / demotion → evolution of creative style." Secondly, generative social simulations based on Large Language Models (LLM), while possessing strong semantic expressive capabilities, are prone to "time illusion" and "personality drift" due to the lack of a causal chain closed-loop verification mechanism and historical context embedding constraints. The former manifests as agents generating content that violates temporal logic, while the latter manifests as a severe misalignment between the poet's words and actions and their historical persona, failing to meet the core requirements of academic-level deduction for temporal accuracy and personality consistency. Thirdly, domestic "Chronological Maps of Tang and Song Literature" can only achieve geographical location of poets' activity trajectories, failing to establish the connection between personality traits and behavioral decisions, and thus unable to present the inherent logic of "events triggering emotions, and emotions driving creation." The "Chinese Classics Database" only supports document retrieval and collation comparison, lacking any personality simulation or interactive functions. The 3D modeling of cultural heritage projects such as the Sichuan Chen Zi'ang Hometown Digital Museum and the Du Fu Thatched Cottage Online Exhibition Hall can only statically recreate the appearance of scenes, lacking both the dynamic personality of the poets and emotional feedback; their so-called "interaction" is merely a rudimentary function of clicking to view images and text. Fourthly, international digital humanities projects such as the "Perseus Digital Library" and "Digital Classicist" focus on Western classical literature, but their constructed knowledge graphs cannot adapt to the social logic and creative patterns of Tang and Song dynasty literati. Japan's "Digital Tale of Genji" can only achieve geospatial analysis, completely neglecting the simulation of the linkage between personality, emotion, and behavior, and thus has no substantial reference value for the study of Tang and Song literature. While the Generative Agents project, open-sourced by Stanford Lab in 2023, has achieved autonomous interaction of intelligent agents, it is only suitable for modern virtual social interactions, game NPCs, and other scenarios without historical constraints. It does not take into account the historical context of the Tang and Song dynasties, such as the official system, social etiquette, and language style. Directly applying it will result in serious personality drift and time misalignment. Although knowledge graphs, ECMF frameworks, and PCL technology have been initially applied in text analysis and personality simulation, they have not been customized to the cultural characteristics of Tang and Song poets, such as their literary integrity, career ideals, and social rules, and therefore cannot support high-fidelity simulation of core personality dimensions.While Pearl's causal hierarchy and counterfactual thinking provide a theoretical basis for hypothetical historical research, current technology has not transformed them into usable tools for Tang and Song literature studies. There are neither standardized event schema templates nor dynamic deduction engines. They can only make fragmented speculations about single events and cannot form a logically consistent chain of events. They are completely unable to support in-depth hypothetical studies such as "the comprehensive impact of the failure of the An Lushan Rebellion on Du Fu's life trajectory and the evolution of his poetic style".

[0003] More importantly, although the aforementioned technologies have made breakthroughs in their respective fields, none of them explicitly model the natural evolutionary laws of the following social systems. This results in simulation results remaining at a "superficially reasonable" level, failing to align with real logic in a historically verifiable dimension. First, individual emotional responses follow a humanistic cognitive dynamic pathway of "stimulus → cognitive assessment → emotional potential accumulation → behavioral threshold triggering." This pathway corresponds to the core poetry generation logic of Tang and Song literati, such as "event reflection → emotional sedimentation → creative expression." Existing technologies have not constructed a quantitative model of this pathway, making it impossible to accurately reproduce the internal mechanism of emotion-driven creation. This leads to a severe disconnect between the generated behavior and historical personality, completely failing to support dynamic simulations of scenarios such as "Chen Zi'ang writing 'Ascending Youzhou Terrace' after being rebuked for offering advice." Second, the dissemination of group ideas follows the power-law decay and local consensus-building laws in information dissemination. This law dominates the process of thought diffusion and school formation in the social networks of Tang and Song literati. Existing technologies... First, without embedding it into a simulation framework, it is difficult to replicate the natural evolution and stylistic resonance of literary schools. Second, the impact of historical events has path dependence and critical point mutation characteristics. Small initial disturbances may trigger macroscopic state transitions, similar to thermodynamic phase transitions. The evolution of history will show significant path dependence, and mutations will occur when it reaches a critical point. Accidental events will affect the occurrence and change of inevitable events. However, a single technical path leads existing research into an "either / or" dilemma: focusing on historical facts will lose interactive flexibility, while pursuing interaction will sacrifice historical accuracy. It is impossible to take into account the core requirements of historical rigor, emotional authenticity, and interactive flexibility. It is impossible to promote the digitalization of Tang and Song literature from static presentation to dynamic interaction. The subversive impact of key events such as the An Lushan Rebellion and the Reform Movement on the fate of literati and the direction of literature is a typical manifestation of this characteristic. Existing technologies have not explicitly modeled the laws of such mutations, and cannot achieve accurate prediction of historical evolution.

[0004] Therefore, there is an urgent need for a simulation framework that deeply integrates objective temporal constraints, biological cognitive models, and statistical prior laws, so that the inference results not only "look reasonable" but also conform to the natural laws of social evolution at the level of physical verifiability. This invention proposes a systematic solution to address this technological gap. Summary of the Invention

[0005] To address the problems existing in the prior art, the purpose of this invention is to provide a historical social dynamic simulation system and method oriented towards time constraints. This invention solves problems such as time drift, personality distortion, and non-physical inference in the prior art, and is applicable to digital humanities research and civilization evolution modeling. It has clear technical substance and verifiable effects.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a historical social dynamic simulation system oriented towards time-series constraints, comprising:

[0007] The five-dimensional temporal knowledge graph G5 module is used to construct a temporal event network that satisfies the acyclic directed graph (DAG) topology based on multi-source historical data, and to use this as an objective temporal constraint field for agent behavior.

[0008] The computational personality modeling module uses the DeepPersona algorithm to encode the target person's life data, text output, and era context into a personality potential vector. Through a closed-loop logic of external event stimulation, cognitive assessment, emotional potential accumulation, and behavioral threshold triggering, the emotional accumulation process is modeled as an observable quantitative potential.

[0009] The counterfactual event graph engine's parameter space for its counterfactual event graph templates is determined by the statistical distribution of historical events. The generated virtual timeline and the five-dimensional temporal knowledge graph maintain structural isomorphism on the basic event subgraph.

[0010] The multi-agent consistency verification module integrates personality perception contrastive learning (PCL) technology for real-time deviation detection and reconciliation.

[0011] The multi-engine integration and dual-mode optimization module integrates a generative intelligent agent framework, defines the historical mode as a knowledge graph-driven deterministic subsystem, and defines the counterfactual mode as an event graph perturbation subsystem. It achieves mode linkage through an entropy conservation switching protocol and supports real-time verification and counterfactual backtracking.

[0012] As a further improvement of the present invention, the five-dimensional temporal knowledge graph G5 module includes an entity set E and a relation set R, wherein:

[0013] The entity set E is: {target person set E1, geographic node set E2, timestamp set E3, event type set E4, work set E5};

[0014] The relation set R is: any triple (e, loc, t) ∈ R, where e represents an event, loc represents a location / location identifier, and t represents a time / time stamp. It must satisfy t ∈ E3 and e ∈ E1, and the five-dimensional time-series knowledge graph as a whole is an acyclic directed graph DAG to ensure that the time sequence of historical events is irreversible.

[0015] As a further improvement of the present invention, in the computational personality modeling module, the emotion accumulation process is modeled as an observable quantitative potential energy as follows:

[0016] ,in, Let be the event stimulus intensity function. For the rate of emotion decay, For cognitive weights, t− Let t be the time when the current event occurs and the time when the kth event occurs. The time interval, This is the emotional index decay term; the formula reflects the natural law of exponential decay of emotional memory.

[0017] As a further improvement of the present invention, the quantized potential energy breaks through the dynamic threshold. Time-triggered behavior actively outputs, among which The baseline threshold, The context sensitivity coefficient. The information entropy of the current event context reflects the modulating effect of social uncertainty on the threshold for triggering behavior.

[0018] As a further improvement of the present invention, the counterfactual event graph engine includes three types of elements in its counterfactual event graph template: type, core parameters, and chain rules, specifically:

[0019] Type∈{political turmoil, natural disaster, cultural movement};

[0020] The core parameter CoreParams follows a statistical distribution fitted by historical event samples;

[0021] The ChainRule defines a perturbation function of the probability of an event on subsequent nodes, and the perturbation function satisfies the constraint of non-increasing information entropy: ,in, This represents the upper limit threshold for noise in the social system. Let be the random variable representing the state of the social system at simulation time t. Let be the random variable representing the state of the social system at the (t+1)th simulation time. For the complete historical event sequence up to time t, This is the complete historical event sequence up to time t−1.

[0022] As a further improvement of the present invention, the multi-agent consistency verification module performs real-time deviation detection and reconciliation using four-dimensional indicators: personality vector cosine similarity, contextual semantic matching degree, HMM state transition probability, and language WMD distance. Specifically, if the personality matching degree is less than a first preset threshold, the context switching module is triggered to automatically fill in the differences in historical context. If the personality matching degree is less than a second preset threshold, the interaction is terminated and a message is displayed indicating a logical conflict in the current combination, preventing the generation of effective dialogue, thereby ensuring personality consistency and logical self-consistency during the interaction process.

[0023] As a further improvement of the present invention, the verification rules of the multi-agent consistency verification module are as follows:

[0024] Identity localization unit: Calculates the cosine similarity between the current agent's personality vector P and the embedding center of the historical corpus. If the cosine similarity is less than a preset threshold, personality correction is triggered.

[0025] Scene state unit: Match the neighborhood semantic vector of the current event node t in the five-dimensional temporal knowledge graph G5 module. If the cosine similarity is less than the preset threshold, it is considered a scene misalignment.

[0026] Short-term target unit: The transition probability of the behavior sequence is calculated based on the Hidden Markov Model. If the transition probability is less than a preset threshold, the target is determined to be drifting.

[0027] Expression style unit: Calculate the WMD distance between the generated text and the text style D_style. If the WMD distance is greater than the preset threshold, style transfer correction is initiated.

[0028] As a further improvement of the present invention, when both the identity positioning unit and the expression style unit fail to be verified, the personality conflict reconciliation model is activated, and its adaptation weight is calculated according to the following formula: The difference between the personality vectors of the two agents. , Let i and j be the personality feature vectors of the i-th and j-th agents. This is the cognitive stiffness coefficient.

[0029] As a further improvement of the present invention, the system runs on a distributed computing platform and supports dynamic switching between historical mode and counterfactual mode in a standard test environment with 8 cores, 16GB RAM, and 10 concurrent users, with a switching latency of ≤1 second.

[0030] This invention also provides a time-constrained historical social dynamics simulation method, implemented using the time-constrained historical social dynamics simulation system described above. The method includes the following steps:

[0031] S1. Construct the five-dimensional temporal knowledge graph G5 module;

[0032] S2. Perform computational personality modeling for each target individual and output the personality potential vector. ;

[0033] S3. Initialize the agent state and load the five-dimensional temporal knowledge graph G5 module as a temporal constraint field;

[0034] S4. Proceed with the simulation according to the preset time step, first obtaining the current event node in each step. Calculate the emotional potential energy of each agent. ;like ≥ If this occurs, behavior generation is triggered; finally, PCL four-dimensional verification is performed to correct any abnormal behavior.

[0035] S5. Output the simulation result sequence, including the behavioral trajectories of each agent, event response records, and group evolution indicators.

[0036] The core of this invention lies in:

[0037] (1) A dual constraint mechanism of five-dimensional temporal knowledge graph + event potential field is proposed to ensure that the behavior of the agent both conforms to the temporal order of objective historical events and retains local autonomous response capabilities. Based on multi-source historical materials, a five-dimensional temporal knowledge graph that satisfies the DAG topology is constructed, and the five types of entities and their relationships, namely poets, places, times, events, and works, are transformed into an objective temporal constraint field to ensure that the behavior of the agent strictly conforms to the temporal order and causal logic of historical events. At the same time, the dynamic constraint of the event potential field is superimposed, which retains the agent's autonomous cognition and emotional response capabilities in local scenarios under the premise of locking the macro-historical trajectory, and achieves a balance between rigid historical constraints and local flexible interactions;

[0038] (2) Construct an event-emotion-behavior potential energy model inspired by neurocognition. Using the DeepPersona algorithm, a generative engine, the target person's biographical data, text output, and historical context are encoded into a personality potential energy vector. Through a closed-loop logic of "external event stimulation → cognitive evaluation → emotional potential energy accumulation → behavioral threshold triggering," the emotional accumulation process is modeled as an observable quantifiable potential energy (the mathematical expression is...). ,in For cognitive weight, (This refers to the rate of emotional decay parameter), when it exceeds a dynamic threshold. The mechanism of triggering behavior to actively output is consistent with the dopamine reward prediction error theory, accurately restoring the internal logic of Tang and Song literati's "event reflection → mood settling → creative expression";

[0039] (3) Design a counterfactual timeline template library. The template types cover three core scenarios: political turmoil, natural disasters, and cultural movements. The core parameter space is determined by the statistical distribution fitted to historical event samples (e.g., the success rate of rebellion follows a Beta distribution, and the disaster-affected area follows a Gamma distribution). The event perturbation function defined by the chain rule satisfies the non-increasing information entropy constraint H( | )≤H( | )+ϵ (ϵ is the upper limit threshold of noise in the social system), ensuring that all counterfactual inferences are within the boundaries of the natural laws of social evolution, and that the generated virtual timeline and the five-dimensional temporal knowledge graph maintain structural isomorphism on the basic event subgraph;

[0040] (4) The PCL-Plus consistency verification protocol is introduced, integrating a four-layer verification mechanism of identity positioning, context adaptation, style transfer, and conflict reconciliation. Real-time deviation detection is performed using four-dimensional indicators: cosine similarity of personality vectors (threshold ≥ 0.85), contextual semantic matching degree (threshold ≥ 0.75), HMM state transition probability (threshold ≥ 0.7), and language WMD distance (threshold ≤ 0.35). When both identity positioning and expression style unit verification fail, the personality conflict reconciliation model (adaptation weight) is activated. , where ΔP is the difference in personality vectors between the two agents), dynamically corrects the bias in the semantic-personality joint embedding space trained based on historical corpora, solves the cognitive discrepancy problem in multi-agent interaction, and ensures the consistency of personality and logical self-consistency in the interaction of poets across time and space and across different schools of thought.

[0041] The beneficial effects of this invention are:

[0042] 1. This invention aims to solve the technical problem of how to construct an AI-driven social simulation framework with high causal fidelity, logical self-consistency, and cross-modal consistency under the constraints of natural laws such as the inherent irreversibility of temporal sequence in historical social systems, the closed nature of causal chains of individual behavior, and the convergence of group evolution paths. Its core lies in utilizing the objective feedback mechanism, information entropy constraints, and neurocognitive dynamics in the evolution of social systems to achieve calculable reconstruction and counterfactual deduction of complex historical situations. It is applicable to scenarios with strict scientific empirical requirements, such as digital humanities research, impact assessment of major historical events, and modeling of civilization evolution paths.

[0043] 2. This invention is verified by quantifiable indicators: Under standard testing environment (Intel Xeon 8 cores / 16GB RAM / 10 concurrent connections), the historical mode restoration F1-score is ≥91.3%, and the counterfactual chain logic self-consistency rate is ≥86.7% (manual blind review + automatic conflict detection dual verification), which is significantly better than the existing ABM and LLM solutions. Attached Figure Description

[0044] Figure 1 This is a flowchart of the system processing in an embodiment of the present invention;

[0045] Figure 2 This is a closed-loop logic diagram of "external event - emotional potential energy accumulation unit - threshold triggering behavior output" in an embodiment of the present invention;

[0046] Figure 3 This is a flowchart of the self-questioning standardization test in an embodiment of the present invention;

[0047] Figure 4 This is a flowchart illustrating the multi-engine integration and system optimization process in an embodiment of the present invention.

[0048] Figure 5 This is a flowchart illustrating the construction of the poet's personality model in an embodiment of the present invention;

[0049] Figure 6 This is a diagram of the five-dimensional collaborative architecture of the system in this embodiment of the invention;

[0050] Figure 7 This is a schematic diagram of the spatial distribution of event graph parameters of the three modules in an embodiment of the present invention;

[0051] Figure 8 This is a schematic diagram of the rebellion scenario parameters in an embodiment of the present invention, taking the An Lushan Rebellion as an example;

[0052] Figure 9 This is a schematic diagram of the historical mode temporal constraint field in an embodiment of the present invention;

[0053] Figure 10 This is a logic diagram for generating a counterfactual timeline in an embodiment of the present invention;

[0054] Figure 11 This is a schematic diagram of the event-potential-behavior triggering dynamics model in an embodiment of the present invention. Detailed Implementation

[0055] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0056] Example 1

[0057] like Figure 1 As shown, a historical social dynamics simulation system oriented towards time-series constraints is described. The system's process execution steps are as follows:

[0058] S1. Construct a five-dimensional temporal knowledge graph and historical simulation mechanism: Based on the chronological map data of Tang and Song literature, integrate historical documents, poet biographies and works records to construct a five-dimensional temporal knowledge graph covering poets, places, times, events and works, and define core relationships such as "born at", "traveled to", and "inspired creation"; design an intelligent agent navigation mechanism under the constraints of the knowledge graph, disable the macro long-term planning module of the intelligent agent, guide macro actions with historical nodes as mandatory instructions, and retain autonomous cognitive functions such as reflection and emotional perception.

[0059] S2, a computational personality model for poets based on DeepPersona: Through data mining, personality modeling, and consistency verification, it integrates the poet's life trajectory, key events, poetic style, and historical context to extract core personality dimensions and establish a model like... Figure 2 The external event-internal emotion-behavioral decision-making closed-loop linkage logic is shown; the constructed personality model is tested for consistency with historical records, poetry works, and social relationships, and if it fails, the data weights are adjusted or supplementary data is added.

[0060] S3, Counterfactual Event Schema Definition and Dynamic Timeline Generation: Define standardized templates for Event Schemas for core scenarios such as rebellion, reform, and natural disasters, and clarify key controllable variables (such as the success rate of rebellion and the scope of disaster); develop a schema-guided event simulation engine that supports custom assumptions such as modifying historical event parameters and placing poets across generations, and dynamically generate logically consistent event chains and timelines.

[0061] S4, Construction of a Multi-Agent Interaction Consensus Mechanism: Applying PCL technology to establish a mechanism such as... Figure 3 The self-questioning verification process shown before interaction covers four dimensions: core identity positioning, scenario status, short-term goals, and expression style. For cross-temporal and spatial interaction scenarios, a personality conflict reconciliation model is constructed to quantify trait differences and design appropriate communication strategies.

[0062] S5, such as Figure 4 As shown, multi-engine integration and dual-mode optimization are implemented: an integrated generative agent framework is used to construct a closed loop of "memory storage - deep reflection - behavior planning"; the ECMF framework is integrated to simulate the dynamic driving of external events on the poet's emotions; and the optimized mode is a switchable world model to achieve seamless switching between historical mode (knowledge graph guidance) and counterfactual mode (event schema guidance).

[0063] In S1, the construction of the temporal knowledge graph requires data cleaning using Python pandas and numpy libraries, including missing value imputation, duplicate data removal, and outlier filtering to ensure data structuring; the intelligent agent navigation mechanism locks the physical journey by forcibly writing historical node memories, while allowing autonomous generation of inner monologues and emotional feedback.

[0064] In S2, the core personality dimensions are extracted, including fine-grained features such as political inclination, emotional patterns, and creative style. Emotion quantification is performed by calibrating the intensity of emotions through the ECMF framework, and behavioral decisions need to be adapted to the contextual constraints of officialdom rules and social etiquette in the Tang and Song dynasties.

[0065] In S3, the rebellion scenario template includes parameters such as initiator, outbreak area, and success rate, while the natural disaster template covers elements such as occurrence time, affected area, and duration. The event simulation engine uses a chain reaction algorithm to ensure that the counterfactual timeline is logically consistent with the historical context and character traits.

[0066] In S4, PCL technology ensures that interactive behavior matches historical personality through cosine similarity matching of pre-trained personality vectors; the personality conflict reconciliation model designs differentiated transformation strategies to address the differences in expression among poets of different schools and generations, maintaining the self-consistency of dialogue logic.

[0067] In S5, multi-engine integration achieves data adaptation by reconstructing the underlying logic. The mode switching trigger mechanism is based on user operation commands to ensure stable system operation during the switching process. The historical accuracy is ≥90%, and the self-consistency rate of the counterfactual event chain logic is ≥85%.

[0068] This embodiment achieves high-precision replication of the life trajectories and creative backgrounds of Tang and Song Dynasty poets through a five-dimensional temporal knowledge graph and historical constraint mechanism. The historical accuracy of typical poets is ≥90%, overcoming the limitations of traditional static digital presentation and providing a precise historical tool for academic research. Based on the DeepPersona personality modeling method and consistency verification mechanism, a computational personality model library (containing ≥25 core poets) with deep binding of "life-emotion-behavior" is constructed, effectively avoiding personality drift and improving the authenticity and dynamic responsiveness of the poets' images. Standardized counterfactual event schemas and dynamic inference engines provide adjustable events. The variable library and logically consistent hypothetical scenario generation tool transform abstract counterfactual speculations into visual results, expanding the hypothetical exploration dimension of literary studies. PCL technology and personality conflict reconciliation models ensure the consistency of multi-agent interactions, with an accuracy rate of ≥85% in confirming that the interactions and behaviors of poets across time and genres conform to historical personalities, restoring the intellectual collisions and temporal tensions within the literary field. The multi-engine integration and dual-mode switching design balance historical rigor with interactive flexibility, and can be widely applied to scenarios such as academic assistance, immersive education, and cultural dissemination, promoting the digitalization of Tang and Song literature towards dynamism and interactivity, and contributing to the living inheritance of traditional culture.

[0069] Example 2

[0070] A time-constrained historical social dynamics simulation method is proposed, which verifies the social dynamics simulation of historical period P. Historical period P (600-900 AD) is selected as the algorithm verification benchmark. This period is characterized by high-density documentary records, clear event sequences, significant cultural discontinuities, and a representative group of figures, making it suitable as a technical verification scenario. The method includes the following steps:

[0071] S1: Constructing a five-dimensional temporal knowledge graph Based on authoritative chronicles, extract entity sets:

[0072] (1) Integrate Tang and Song literature chronological map data, historical documents, poet biographies and works records to construct a five-dimensional temporal knowledge graph. At the same time, define standardized templates for event diagrams of core scenarios such as rebellion, reform, and natural disasters, and clarify key controllable variables.

[0073] (2) Based on the DeepPersona methodology, such as Figure 5 As shown, a computational personality model of a poet is constructed by integrating the poet's life trajectory, key events, poetic style and historical background, and constructing a closed-loop linkage of "external events - internal emotions - behavioral decisions". Consistency verification is completed through historical records, poetry works and social relationships.

[0074] (3) Construct a dual-mode world model. In the historical mode, knowledge graphs constrain the macroscopic action trajectory of the intelligent agent and retain the autonomous cognitive function. In the counterfactual mode, logically consistent event chains and timelines are dynamically generated based on event graph templates and user-defined parameters.

[0075] (4) Based on PCL technology, construct a standardized verification process for self-questioning before multi-agent interaction and a personality conflict reconciliation model to ensure personality consistency and logical self-consistency during the interaction process;

[0076] (5) Integrate the generative intelligent agent framework and the ECMF framework, optimize the mode switching mechanism, achieve seamless connection between historical mode and counterfactual mode, and complete the system integration and optimization of multi-engine collaboration; the system's five-dimensional collaborative architecture diagram is shown below. Figure 6 As shown.

[0077] ={Target Person 1, Target Person 2, ..., Target Person N} (N≥25 including core poets such as Li Bai, Du Fu, and Chen Zi'ang);

[0078] ={Geographic Node 1, Geographic Node 2, ...} (including core locations such as Chang'an, Luoyang, and Shu).

[0079] ={timestamp , ..., (According to the Gregorian calendar, accurate to the year / month);

[0080] ={Event Type A, B, C, ...} (including imperial examinations, demotion and exile, war and migration, etc.);

[0081] ={works , ...} (including core poems such as "Song of Climbing Youzhou Terrace" and "Spring View");

[0082] A relation set is defined as:

[0083] R = {born in ( , ), travel to ( , , [ [, t_q]), inspiring creation ( , W_m, , t_r)}

[0084] All triples have been cross-verified with historical data to ensure that the temporal topology strictly satisfies the DAG (Directed Acyclic Graph) structure, reflecting the natural law of the irreversibility of history.

[0085] S2: Target Person's Calculated Personality Modeling:

[0086] For target person i (taking Chen Ziang as an example), the complete implementation process is as follows:

[0087] Step S2-1: Multi-source data fusion:

[0088] Movement trajectory D_traj = {location sequence (Shu region → Chang'an → Juyan Lake → Youzhou), timestamp sequence (leaving Shu in 679 → joining the army in 686 → ascending Youzhou Terrace in 696)};

[0089] Key events D_evt = {participation in political events, changes in social networks, major turning points} (failed the imperial examination in 679, recommended for submitting a book in 681, rebuked for offering advice in 694, resigned from office and went into seclusion in 698);

[0090] Text style D_style = {Imagery spectrum vector (high-frequency images such as "sword", "heaven and earth", "merit" account for ≥35%), emotional polarity curve (sorrowful feeling accounts for 68%, ambition accounts for 57%), grammatical complexity (short sentences account for 72%, which conforms to the expression characteristics of generosity and sorrow)};

[0091] The historical context _ctx = {Institutional constraint matrix (the early Tang Dynasty's imperial examination pass rate was ≤5%, and border policies allowed scholars to join the army), language norms vector (rhyme rules of seven-character ancient poems in the early Tang Dynasty), and principal components of ideological trends (the trend of establishing merit and achieving success has a weight of 0.7, and the trend of Buddhist and Taoist reclusion has a weight of 0.3)};

[0092] Step S2-2: Data weight adjustment and missing value imputation:

[0093] Weighting: Key events recorded in official history (such as being rebuked for offering advice) are weighted at 0.8; the results of the analysis of the poetry text (such as the spectrum of imagery) are weighted at 0.6; and later commentaries (such as the evaluation of Chen Zi'ang in "Chronicles of Tang Poetry") are weighted at 0.3.

[0094] Missing value imputation: Social preference data was supplemented by Lu Zangyong’s “Biography of Chen Zi’ang”, a contemporary poet, and the era influence factor of his creative style was supplemented by a study of the frontier poetry style of the early Tang Dynasty.

[0095] Step S2-3: Generation of Personality Potential Vector:

[0096] The fused data is input into the DeepPersona architecture, which outputs a personality potential vector. ∈R d The core dimensions and their corresponding quantified values ​​are as follows:

[0097] d1: Tendency to obey authority (reflecting the strength of constraints within the official system): 0.3 (low obedience, consistent with his historical reputation for upright remonstrance).

[0098] d2: Emotion decay rate (fitted by the variance of text sentiment fluctuation): 0.65 (moderate decay, corresponding to the accumulation of long-term feelings of being unappreciated).

[0099] d3: Cognitive resilience coefficient (calculated from cross-event behavioral consistency): 0.4 (low resilience, reflecting its stable value of "prioritizing achievement and success").

[0100] d4: Social openness (quantified based on social interaction records): 0.7 (high openness, suitable for his close interactions with Qiao Zhizhi and Lu Zangyong);

[0101] d5: Creation trigger threshold (emotional potential threshold): 0.8 (high threshold, strong stimulus is required to trigger creation, which fits the creative background of "Ascending Youzhou Terrace").

[0102] →This modeling process directly maps the natural law of “trait-situation interaction” in social cognition, achieving a deep binding of “life experience-personality traits-behavioral logic”.

[0103] Step S2-4: Triple Consistency Check

[0104] Historical record verification: The model output of "low authority compliance" is consistent with the record of "repeatedly submitting memorials to remonstrate" in the Old Book of Tang, and the verification is passed;

[0105] Poetry verification: The model's prediction of "high social openness" matches the number of poems exchanged between poets during the same period (12 existing related poems), and the verification is successful;

[0106] If any validation fails, return to Step S2-2 to readjust the data weights (e.g., supplement with data on military service experience on the frontier to optimize the rate of emotional decay).

[0107] S3: Counterfactual Event Schema-Driven Deduction:

[0108] Define event graph template Σ=<Type, CoreParams, ChainRule> :

[0109] Type ∈ {political turmoil, natural disasters, cultural movements} (using the An Lushan Rebellion as a prototype to fit the early Tang Dynasty setting, defined as "frontier rebellion"), such as Figure 7 As shown;

[0110] Core parameters CoreParams ~ Distribution (e.g., probability of rebellion success ~ Beta) =2.1, β=3.7), the parameters are obtained by statistical fitting of historical events.

[0111] ChainRule: Describes how an event changes the probability of subsequent nodes (e.g., "famine → increase in refugees → shift in poetic style towards compassion").

[0112] Specific parameters: Initiator = Governor of Yingzhou, Outbreak area = Yingzhou, Success rate = 35%, External aid speed = 45 days, Affected area = Hebei / Liaoning;

[0113] The inference engine calls Σ to generate counterfactual events first T. l and forced T l Maintain isomorphism with G5 on the basic event subgraph (such as the basic event subgraphs "Chen Zi'ang was born in Shu" and "He presented his book and was recommended in 681") to ensure that the deduction does not violate the basic conservation law of social evolution.

[0114] S4: Consistency Guarantee in Multi-Agent Interactions

[0115] Using the PCL-Plus protocol, taking Chen Zi'ang and Li Bai (a cross-generational interaction scenario) as an example, as shown in the table below:

[0116]

[0117] In S1, the process of constructing the five-dimensional temporal knowledge graph G5 and the standardized event graph template is as follows:

[0118] (1-1) Data integration and preprocessing: Taking the "Chronological Map of Tang and Song Literature" as the core data source, we integrated official history, poet biographies, and texts from "Complete Tang Poems" and "Complete Song Lyrics" to extract five-dimensional data of "poet-location-time-event-work"; we used the pandas library of Python to impute missing values, the numpy library to remove duplicate data, and filtered outliers through logical checks to form a standardized structured dataset;

[0119] (1-2) Construction of a five-dimensional temporal knowledge graph: Based on a graph database (such as Neo4j), a graph is constructed, defining five core relationships: "born", "traveled to", "participated", "inspired to create", and "resigned from office and retired", to replicate the poet's life trajectory and the internal logic of his creation;

[0120] (1-3) Standardized template definition for event schemas: Key parameters and controllable variables are defined for three core scenarios: rebellion, natural disaster, and reform. The rebellion scenario is based on the An Lushan Rebellion, such as... Figure 8 As shown, the natural disaster scenario is based on the Yellow River breach in 1048, and the reform scenario is based on Wang Anshi's reforms, clearly defining the core elements and adjustable dimensions of each scenario;

[0121] Referring to the "Poet Personality Model Construction Flowchart," in S2, the target poet's personality model construction module includes three sub-modules: data extraction, model construction, and consistency verification. The detailed process of each module is as follows:

[0122] The data extraction module process includes: mining multi-dimensional data on the poet's life trajectory, key personal events, poetic style, socio-cultural characteristics, and historical background. Among them, the extraction of poetic style combines text semantic analysis, imagery preference identification, and emotional tendency quantification. The historical background data covers core information such as the Tang Dynasty's official system, social etiquette, language habits, and ideological trends.

[0123] The model building module process includes: adjusting data weights (historical events weight 0.8, poetry text analysis weight 0.6, and later commentary weight 0.3), supplementing missing information to form a basic dataset, extracting core dimensions such as personality traits, value system, and emotional patterns based on the DeepPersona algorithm, and constructing a closed-loop linkage model of "external events - internal emotions - behavioral decisions".

[0124] The consistency verification module process includes three verifications: historical record verification, poetry work verification, and social relationship verification. If the verification fails, the process returns to the data integration stage to readjust the data weights or supplement the data to ensure that the personality model is consistent with the historical image.

[0125] In S3, such as Figure 10As shown, the construction of the dual-mode world model includes the following steps:

[0126] Step (3-1) Implementation of historical mode: Load the five-dimensional temporal knowledge graph, disable the agent's macro long-term planning module, use historical nodes as mandatory instructions to guide macro actions, retain autonomous cognitive functions such as reflection, emotional perception, and inner monologue generation, and achieve a balance between historical constraints and autonomous response;

[0127] Step (3-2) Counterfactual mode user-defined settings: Supports three types of operations - modifying parameters of key historical events (such as the success rate of the An Lushan Rebellion), placing poets across generations (such as placing Li Shangyin in the prosperous Tang Dynasty), and customizing the birth date, personality traits, and family background of characters;

[0128] Step (3-3) Event Chain Generation and Timeline Adaptation: Based on the event diagram template rules and user settings, generate a logically consistent event chain with chain reactions, dynamically adjust the poet's life trajectory, mood changes and creative theme evolution path to ensure high adaptation with the set scenario.

[0129] In S4, the construction of the multi-agent interaction consensus mechanism includes the following steps:

[0130] Step (4-1) Self-questioning standardized verification process: Before the interaction is triggered, the agent completes four-dimensional verification of core identity positioning, current scene state, short-term behavioral goals, and exclusive expression style. The cosine similarity matching of pre-trained personality vectors (threshold ≥ 0.85) ensures that the words and actions are consistent with personality traits.

[0131] Step (4-2) Construction of a personality conflict reconciliation model: Quantitatively analyze the differences in the core traits of different personalities (such as the differences in expression between Du Fu's realism and Li Shangyin's aestheticism), design differentiated communication adaptation strategies, and resolve logical conflicts while preserving the independence of personality;

[0132] Step (4-3) Interaction Consistency Verification: Through cross-temporal and spatial interaction scenario testing (such as a dialogue between Du Fu and Li Shangyin), verify that the accuracy of the intelligent agent's words and actions in conforming to its own personality traits is ≥85%, ensuring no personality drift or homogenization of expression. When the accuracy is lower than the threshold, automatically backtrack the dimensional weights of the personality model, supplement the corresponding historical data for the personality dimension with the largest deviation (such as political inclination, emotional pattern) and retrain until the target is met.

[0133] As a preferred embodiment of the present invention, in step (5), the multi-engine integration and system optimization are supported by a five-layer technical architecture. This architecture systematically answers five ontological questions of the digital poet from bottom to top, namely, who he is, what he is thinking, where he is, what he has experienced, and how he speaks. The specific implementation includes the following steps:

[0134] Step (5-1) Core cognitive engine integration, corresponding to the cognitive layer. Based on the generative intelligent agent framework, a full-process cognitive loop of "perception-memory flow-retrieval-reflection-planning" is constructed, enabling the digital poet to evolve from short-term memory to long-term beliefs in continuous events. It uses growth-oriented logical judgment to respond to situations such as failing the imperial examinations or going to war, capturing the poet's past experiences and integrating scene information to deduce subsequent autonomous decisions. Through the above customized technical optimization, the digital poet can complete the cognitive evolution from short-term memory to long-term beliefs in a continuous event sequence. It responds to typical scenarios such as failing the imperial examinations, going to war, and being demoted with growth-oriented logical judgment adapted to the characteristics of Tang and Song literati. It accurately captures the poet's past experiences and scene information, deducing autonomous decisions that combine historical authenticity and personality consistency, forming an essential difference from the generalized cognitive logic of general generative intelligent agents, thus building a strong technological protection barrier.

[0135] Step (5-1-1) Construction of Scene Priority Perception Mechanism. A three-tiered perception priority system is established for the core life scenes of Tang and Song literati: Six categories of events strongly related to historical facts and creative works—such as taking the imperial examinations, official promotions and demotions, exile, migrations during wartime, literati social interactions, and leisurely travels—are designated as Level 1 perception priority. Key attributes such as time, location, participants, and historical context are prioritized using event feature extraction algorithms. Non-core events such as daily life and diet are designated as Level 3 perception priority, with core information extracted only through sampling at key nodes. The remaining events are designated as Level 2 priority, achieving differentiated allocation of perception resources, avoiding invalid data consuming cognitive computing power, and ensuring that the perception process accurately focuses on the core life trajectories and creative scenes of Tang and Song poets.

[0136] Step (5-1-2) Memory Storage Mechanism Adapted to Historical Fact Relationship Weights. An innovative "Historical Event Weight Enhancement Strategy" is designed, assigning a 30% higher memory storage weight than general generative agents to core historical events such as passing / failing the imperial examinations, demotion / appointment, outbreak of war, and separation from close friends; and a 20% lower storage weight to everyday trivial events, optimizing memory resource allocation efficiency. Simultaneously, a "Core Historical Fact Memory Decay Inhibition Mechanism for Literati" is constructed, reducing the natural decay rate of core historical facts by 50%. Through dynamic adjustment of the memory decay coefficient, this ensures that poets form long-term stable memories of key events that influence their life trajectories, avoiding personality drift caused by memory loss, and adapting to the cognitive orientation characteristics of Tang and Song poets based on historical experience.

[0137] Step (5-1-3) Construction of a Two-Dimensional Coupled Retrieval Model. A two-dimensional retrieval model, "Personality Traits - Historical Context," is constructed. In addition to considering the proximity and relevance of memories, the retrieval process introduces two dimensions: personality matching degree and historical context adaptability. Based on a pre-trained database of poet personality vectors, core personality traits are matched using cosine similarity calculation (e.g., Du Fu's patriotism and Li Bai's romanticism). Simultaneously, historical context constraints such as the Tang Dynasty's official system, social etiquette, and language norms are embedded to establish a historical adaptability verification mechanism for the retrieval results. For example, when retrieving memories of Du Fu's social interactions, priority is given to matching data on his interactions with contemporary literati such as Fang Guan and Li Bai. Furthermore, social rules are used to ensure that the retrieval results conform to the social paradigm of Tang Dynasty literati "exchanging poems," preventing the recall of memories out of place in the historical context.

[0138] Step (5-1-4) Customizes the reflection triggering and content generation mechanism. A "dual-condition trigger threshold" is set: the reflection process is triggered when two consecutive negative event stimuli are received (such as the combination of demotion and the death of a close friend), or when a major historical event is detected (such as the outbreak of the An Lushan Rebellion). The reflection content focuses on the core cognitive concerns of Tang and Song literati through a theme-constrained model, including the contradictions between career ideals and real-world dilemmas, the correlation between the turmoil of the era and individual fate, and the intellectual resonance and cognitive inspiration in social interactions. This generates high-level cognitive insights that align with the spirit of the literati (such as Du Fu's reflections on the people's livelihood in a chaotic world, and Chen Zi'ang's cognitive refinement of unfulfilled talent), distinguishing it from the generalized reflection output of a general intelligent agent.

[0139] Step (5-1-5) is a planning engine that integrates the rules of the era with personality constraints. The planning process embeds a knowledge base of the rules of the era, such as the official system, social etiquette, and geographical transportation of the Tang and Song dynasties. Through rule matching algorithms, it ensures that the action plan conforms to the historical context (e.g., planning the route to the post based on the distribution data of the post roads in the Tang dynasty, and designing social behaviors according to the norms of official etiquette). At the same time, a personality trait-action planning mapping model is established to transform the poet's core personality traits into planning constraints: for example, Li Bai's planning output is biased towards "wandering famous mountains and rivers, socializing with literati, and impromptu creation", while Du Fu's planning output focuses on "observing people's livelihood, recording war, and submitting remonstrances". This achieves a dual fit between the planning results and the logic of the era and the historical persona, avoiding the unconstrained planning defects of general intelligent agents.

[0140] Step (5-2) integrates the emotional cognition engine, corresponding to the emotional layer and the personality layer. The ECMF framework (emotional layer) is integrated to simulate the dynamic driving of emotions through the chain of "external event stimulation - internal cognitive assessment - emotional value accumulation - expression desire trigger". When emotions such as indignation and grief exceed the threshold, creation is triggered, so that the generated text reflects the emotional cause and effect; at the same time, it is associated with the DeepPersona technology (personality layer) in step (2) to encode the poet's life, complete works and historical evaluation into a high-dimensional vector, and to perform style transfer and value shaping on the large model to ensure that poets such as Chen Zi'ang and Li Bai show distinct differences in language style and decision-making preferences, and are highly similar to the historical prototypes;

[0141] Step (5-3) optimizes the mode switching, corresponding to the world layer and the interaction layer. The world layer provides a stage for the agent's activities. In historical mode, an insurmountable spatiotemporal fence is constructed using a knowledge graph. In counterfactual mode, a virtual event chain conforming to social rules is generated using event graphs, balancing historical rigor with the logical consistency of possible worlds. The interaction layer integrates PCL technology to monitor tone and word choice in real time during dialogues between poets and users or across generations of poets, preventing characters from breaking character and ensuring an immersive cross-temporal and spatial interactive experience. Based on this, the underlying technical logic is reconstructed, dual-mode data adaptation is strengthened, and the switching trigger mechanism is optimized. In a standard test environment with an 8-core 16G server and 10 concurrent users, the switching latency is controlled to ≤1 second, achieving seamless connection and stable operation between historical mode and counterfactual mode. The event-potential-behavior triggering dynamics model is as follows: Figure 11 As shown.

[0142] Example 3

[0143] A historical social dynamic simulation system oriented to time-series constraints, cross-period generalization verification (historical period Q: 1200-1400 AD). To verify the cross-period adaptability and technical module reusability of the present invention, the system was migrated to another historical period Q, and only the knowledge graph G5 and the context parameter D_ctx were replaced, while the remaining modules were reused.

[0144] This period is characterized by moderately complete historical materials, dramatic social changes (such as the Song-Yuan transition and adjustments to the imperial examination system), and distinct literary schools (such as the continuation of the Jianghu poetry school and the bold and unrestrained ci poetry school), which can effectively test the system's adaptability to different historical contexts; specifically including:

[0145] S1: Data replacement and adaptation (only the input layer is modified, the technical logic remains unchanged):

[0146] S1-1: G5 Reconstruction of the Five-Dimensional Temporal Knowledge Graph:

[0147] (1) Entity set update:

[0148] E1 = {Lu You, Xin Qiji, Wen Tianxiang, Jiang Kui and ≥20 other core literati of the Song and Yuan dynasties}, supplemented with exclusive attributes such as official positions and resumes (e.g., Wen Tianxiang's "Right Chancellor and Privy Councilor") and anti-Yuan deeds in the context of the "Song-Yuan transition".

[0149] E2 = {Geographical nodes of the Song and Yuan dynasties such as Lin'an, Jingkou, Ganzhou, and Yashan}, which are associated with Song and Yuan administrative divisions (such as "Jiangnan West Circuit" and "Fujian Circuit") and war-related migration routes (such as "Lin'an → Yashan").

[0150] E3 = {key time stamps such as 1205 (Lu You's dismissal from office) and 1279 (Battle of Yamen)}, accurate to the year / month, highlighting the chronological nodes of the Song-Yuan transition;

[0151] E4 = {Exclusive event types such as the Song-Yuan War, the abolition of the imperial examination system, and the formation of Jianghu poetry societies}, supplemented with special events such as "Resisting the Yuan Dynasty and Serving the Emperor" and "Remnant People Living in Seclusion";

[0152] E5 = {Core Song and Yuan Dynasty works such as "Showing My Son", "Yong Yu Le: Nostalgia at Beigu Pavilion in Jingkou", and "Passing Lingdingyang"}, with added emotional tone attributes such as "Poems of the Remnant People" and "Lingering Charm of Bold and Unrestrained Lyrics".

[0153] (2) Relationship set adaptation: core relationships such as “born in”, “traveled to”, and “inspired to create” are retained, and new Song and Yuan exclusive relationships such as “participating in the resistance against the Yuan”, “living in seclusion and not serving in office” and “forming a poetry society” are added. All ternary groups have been cross-verified by the History of Song, the Collected Drafts of the Song Dynasty, the Continued Comprehensive Mirror for Aid in Government, the History of Yuan, and the Song and Yuan Academic Cases, and strictly follow the DAG topology.

[0154] S1-2: Adjustment of the era context parameter D_ctx:

[0155] (1) Institutional constraint matrix: The admission rate of the imperial examination in the Song and Yuan dynasties was ≤3% (lower than that in the Tang dynasty), the recruitment policy for anti-Yuan volunteer armies, and the abolition of the imperial examination in the Yuan dynasty (1279-1315), etc.

[0156] (2) Language norm vector: Song Ci metrical rules (such as Jiang Kui's self-composed songs), language characteristics of Song and Yuan vernacular poetry, and the expression paradigm of "sorrow and desolation" of the Ming loyalists;

[0157] (3) Main components of ideological trends: the anti-Yuan and restoration of Song thought has a weight of 0.6, the reclusive and escapist thought has a weight of 0.3, and the influence of Neo-Confucianism has a weight of 0.1 (adapted to the background of the flourishing of Neo-Confucianism in the Song and Yuan dynasties).

[0158] S2: Computational Personality Modeling (Reusing the DeepPersona Algorithm): Taking Wen Tianxiang as an example, modeling was completed based on G5 data:

[0159] Multi-source data fusion: Extract the action trajectory of "starting a rebellion to serve the emperor in 1275 (E4(Q)) → being captured in 1278 (E4) → being executed in 1283 (E4)" from G5, and text style data such as "Passing Lingdingyang" and "Ode to Righteousness" (75% of the sentiment is pathos and 82% of the sentiment is loyalty).

[0160] Personality potential vector P 文 Output: Authority obedience tendency (0.1, low obedience, consistent with the anti-Yuan stance), emotional decay rate (0.8, slow decay, corresponding to long-term adherence to loyalty and righteousness), creative trigger threshold (0.9, high threshold, strong stimulation after capture triggers "Ode to Righteousness").

[0161] Triple consistency check: Consistent with the "loyalty and unyielding" recorded in the "Biography of Wen Tianxiang" in the History of Song Dynasty; ≥93% match with the emotional tone of "Passing Lingdingyang"; and matches the social records of Lu Xiufu and Zhang Shijie. The check passed.

[0162] S3: Counterfactual Deduction (Reusing Event Schema Template Library): Selecting "Victory at the Battle of Yamen in 1279" as the counterfactual assumption, and calling the "Political Turmoil" class template:

[0163] Core parameter adjustment: The success rate of the rebellion (probability of Yuan army defeat) is set to 40% (following a Beta (α=2.3, β=3.5) distribution, based on statistical fitting of the comparison of Song and Yuan military strength).

[0164] The chain rule is generated as follows: "Song army wins the Battle of Yamen → Southern Song Dynasty continues → Wen Tianxiang is reinstated → Anti-Yuan sentiment cools down → Poetic style shifts from grief and indignation to revival". The virtual timeline maintains structural isomorphism with the basic subgraphs in G5 such as "Wen Tianxiang's life before being captured" and "Jianghu poetry society".

[0165] Logical self-consistency verification: The generated events such as "Wen Tianxiang's creation of 'Ode to the Restoration'" and "Lu You's reinstatement in his later years" do not conflict with the Song and Yuan systems or the personality of literati, with a self-consistency rate of ≥84%.

[0166] S4: PCL-Plus Consistency Verification (Reusing Four-Dimensional Metrics): Taking the "Intergenerational Interaction between Lu You (Southern Song Dynasty) and Wen Tianxiang (Song-Yuan Transition)" as the test scenario:

[0167] Identity verification: The cosine similarity between Lu You's personality vector and the embedding center of the Southern Song Dynasty poets' corpus is 0.87 (≥0.85), which passes the verification.

[0168] Scene status: The semantic matching degree between the "Jingkou Poetry Gathering" scene and the G5 neighborhood is 0.78 (≥0.75), and the verification is passed;

[0169] Short-term objective: The probability of Lu You's "exchanging experiences in resisting the Jin" and the historical action being transferred is 0.74 (≥0.7), verification passed;

[0170] Expression style: The WMD distance between the generated text and Lu You's "bold and tragic" style is 0.32 (≤0.35), and the verification passes;

[0171] Conflict reconciliation: No validation failure, directly output interactive content, words and actions are consistent with the historical personalities of the two (Lu You "an old horse in the stable" and Wen Tianxiang "loyal and passionate").

[0172] Tested in a standard testing environment (Intel Xeon 8 cores / 16GB RAM / 10 concurrent connections), the following core metrics are output:

[0173] Accuracy of personality modeling; F1-score for historical pattern reconstruction; self-consistency rate of counterfactual chain logic; latency of pattern switching; accuracy of intergenerational interaction personality consistency.

[0174] Tests show that:

[0175] The accuracy of personality modeling decreased by ≤3.2% (due to differences in the completeness of historical data);

[0176] The self-consistency rate of counterfactual reasoning still reached 84.1%.

[0177] This embodiment demonstrates that the core technical framework of this invention (five-dimensional temporal knowledge graph constraints, DeepPersona personality modeling, event graph counterfactual deduction, and PCL-Plus verification) possesses cross-period generalization capabilities. By simply replacing the historical data and contextual parameters of the input layer, it can adapt to the social dynamics simulation needs of different historical stages without exhibiting technical adaptation failures or significant performance degradation. This verifies the system's versatility and stability, demonstrating non-special case modeling.

[0178] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A historical social dynamics simulation system oriented towards time-constrained conditions, characterized in that, include: The five-dimensional temporal knowledge graph G5 module is used to construct a temporal event network that satisfies the acyclic directed graph (DAG) topology based on multi-source historical data, and to use this as an objective temporal constraint field for agent behavior. The computational personality modeling module uses the DeepPersona algorithm to encode the target person's life data, text output, and era context into a personality potential vector. Through a closed-loop logic of external event stimulation, cognitive assessment, emotional potential accumulation, and behavioral threshold triggering, the emotional accumulation process is modeled as an observable quantitative potential. The counterfactual event graph engine's parameter space for its counterfactual event graph templates is determined by the statistical distribution of historical events. The generated virtual timeline and the five-dimensional temporal knowledge graph maintain structural isomorphism on the basic event subgraph. The multi-agent consistency verification module integrates personality perception contrastive learning (PCL) technology for real-time deviation detection and reconciliation. The multi-engine integration and dual-mode optimization module integrates a generative intelligent agent framework, defines the historical mode as a knowledge graph-driven deterministic subsystem, and defines the counterfactual mode as an event graph perturbation subsystem. It achieves mode linkage through an entropy conservation switching protocol and supports real-time verification and counterfactual backtracking.

2. The historical social dynamics simulation system oriented towards time-constrained scenarios according to claim 1, characterized in that, The five-dimensional temporal knowledge graph G5 module includes an entity set E and a relation set R, wherein: The entity set E is: {target person set E1, geographic node set E2, timestamp set E3, event type set E4, work set E5}; The relation set R is: any triple (e, loc, t) ∈ R, where e represents an event, loc represents a location / location identifier, and t represents a time / time stamp. It must satisfy t ∈ E3 and e ∈ E1, and the five-dimensional time-series knowledge graph as a whole is an acyclic directed graph DAG to ensure that the time sequence of historical events is irreversible.

3. The historical social dynamics simulation system oriented towards time-constrained scenarios according to claim 1, characterized in that, In the computational personality modeling module, the process of emotion accumulation is modeled as an observable quantifiable potential energy as follows: ,in, Let be the event stimulus intensity function. For the rate of emotion decay, For cognitive weights, t− Let t be the time when the current event occurs and the time when the kth event occurs. The time interval, This is the emotional index decay term; the formula reflects the natural law of exponential decay of emotional memory.

4. The historical social dynamics simulation system oriented towards time-constrained scenarios according to claim 3, characterized in that, The quantized potential energy breaks through the dynamic threshold Time-triggered behavior actively outputs, among which The baseline threshold, The context sensitivity coefficient. The information entropy of the current event context reflects the modulating effect of social uncertainty on the threshold for triggering behavior.

5. The historical social dynamics simulation system oriented towards time-constrained scenarios according to claim 1, characterized in that, In the counterfactual event schema engine, the counterfactual event schema template includes three elements: type, core parameters, and chain rules. Type∈{political turmoil, natural disaster, cultural movement}; The core parameter CoreParams follows a statistical distribution fitted by historical event samples; The ChainRule defines a perturbation function of the probability of an event on subsequent nodes, and the perturbation function satisfies the constraint of non-increasing information entropy: ,in, This represents the upper limit threshold for noise in the social system. Let be the random variable representing the state of the social system at simulation time t. Let be the random variable representing the state of the social system at the (t+1)th simulation time. For the complete historical event sequence up to time t, This is the complete historical event sequence up to time t−1.

6. The historical social dynamics simulation system oriented towards time-series constraints according to claim 1, characterized in that, The multi-agent consistency verification module performs real-time deviation detection and reconciliation using four-dimensional indicators: personality vector cosine similarity, contextual semantic matching degree, HMM state transition probability, and language WMD distance. Specifically, if the personality matching degree is less than a first preset threshold, the context switching module is triggered to automatically fill in the differences in era background. If the personality matching degree is less than a second preset threshold, the interaction is terminated and a message is displayed indicating that there is a logical conflict in the current combination and that an effective dialogue cannot be generated, thereby ensuring personality consistency and logical self-consistency during the interaction process.

7. The historical social dynamics simulation system oriented towards time-series constraints according to claim 6, characterized in that, The specific verification rules of the multi-agent consensus verification module are as follows: Identity localization unit: Calculates the cosine similarity between the current agent's personality vector P and the embedding center of the historical corpus. If the cosine similarity is less than a preset threshold, personality correction is triggered. Scene state unit: Match the neighborhood semantic vector of the current event node t in the five-dimensional temporal knowledge graph G5 module. If the cosine similarity is less than the preset threshold, it is considered as scene misalignment. Short-term target unit: The transition probability of the behavior sequence is calculated based on the hidden Markov model. If the transition probability is less than the preset threshold, the target is determined to be drifting. Expression style unit: Calculate the WMD distance between the generated text and the text style D_style. If the WMD distance is greater than the preset threshold, style transfer correction is initiated.

8. The historical social dynamics simulation system oriented towards time-constrained scenarios according to claim 7, characterized in that, When both the identity positioning unit and the expression style unit fail to be validated, the personality conflict reconciliation model is activated, and its adaptation weight is calculated according to the following formula: The difference between the personality vectors of the two agents. , Let i and j be the personality feature vectors of the i-th and j-th agents. This is the cognitive stiffness coefficient.

9. The historical social dynamics simulation system oriented towards time-constrained scenarios according to any one of claims 1-8, characterized in that, The system runs on a distributed computing platform and supports dynamic switching between historical and counterfactual modes in a standard test environment with 8 cores, 16GB RAM, and 10 concurrent users, with a switching latency of ≤1 second.

10. A method for simulating historical social dynamics oriented towards time-series constraints, characterized in that, The method is implemented using the historical social dynamics simulation system oriented towards time constraints as described in any one of claims 1-9, and includes the following steps: S1. Construct the five-dimensional temporal knowledge graph G5 module; S2. Perform computational personality modeling for each target individual and output the personality potential vector. ; S3. Initialize the agent state and load the five-dimensional temporal knowledge graph G5 module as a temporal constraint field; S4. Proceed with the simulation according to the preset time step, first obtaining the current event node in each step. Calculate the emotional potential energy of each agent. ;like ≥ If this occurs, behavior generation is triggered; finally, PCL four-dimensional verification is performed to correct any abnormal behavior. S5. Output the simulation result sequence, including the behavioral trajectories of each agent, event response records, and group evolution indicators.

Images