Smart park digital twin management platform and virtual-real interaction method

Abstract

The invention discloses a smart park digital twinborn management platform and a virtual-real interaction method, and belongs to the technical field of park management. Structural states, equipment behaviors, environment disturbance and user activity data of entity objects are hierarchically collected according to frequency, and a plurality of digital twinborn models are synchronously constructed based on a scene labeling mechanism; the platform can generate a control strategy and mark a prediction channel according to the model evolution trajectory, then acquire entity response and perform graph structure comparison with a prediction path, identify response deviation and generate a deviation graph; the platform further constructs an error learning weight according to the deviation map, drives a twin model to perform progressive optimization adjustment of confidence parameters, and can trigger structure-level network evolution to realize dynamic enhancement of a model structure and prediction capability; compared with the prior art, the invention provides a smart park digital twinning implementation scheme supporting virtual and real closed loops, self-learning evolution and structure reconstruction.

Description

Technical Field

[0001] The present invention relates to the field of park management technology, and in particular to a smart park digital twin management platform and a virtual-reality interaction method. Background Art

[0002] With the rapid development of new smart cities and industrial parks, the volume of data on equipment operating status, personnel flow, energy consumption, and security management within these parks is growing exponentially. Unified management, accurate prediction, and efficient scheduling have become key challenges in the current digital upgrade of these parks. Digital twin technology, a key means of mapping physical and virtual spaces, has seen initial application in industrial manufacturing, building operations and maintenance, and traffic scheduling. However, most existing park-level digital twin systems remain at the "static display" or "information integration" level, lacking in-depth support for real-time multi-source data fusion modeling, bidirectional virtual-reality linkage, and dynamic system response capabilities. Furthermore, parks feature diverse building types and complex operating scenarios, involving multiple subsystems such as electromechanical equipment management, energy consumption scheduling, security warnings, and environmental monitoring. Existing systems typically utilize single models or static rule bases for management, making it difficult to achieve accurate modeling and dynamic control for specific scenarios. Without dynamic feedback and model update mechanisms, the decision-making capabilities and response accuracy of digital twin platforms fall short of meeting actual management needs, hindering the realization of true "virtual-reality interaction" and an "intelligent closed loop." Summary of the Invention

[0003] The purpose of the present invention is to provide a smart park digital twin management platform and a virtual-reality interaction method to address the shortcomings of the background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solutions: a smart park digital twin management platform, comprising: A modeling module that constructs multiple self-organizing sub-digital twin models, which are dynamically associated with each other through a logical topology diagram to form a system-level digital twin network; A three-dimensional evolution engine is used to construct a temporal behavior map of physical objects based on historical data, introduce environmental disturbance factors for multi-field coupling prediction, and output multi-path trend trajectories for strategy evaluation; A strategy generation module, which combines the prediction trajectory of the evolution engine with the local anomaly criteria of each sub-model to automatically identify potential weaknesses and generate a hierarchical strategy set for space-time partitioning; The execution feedback module is used to send the policy set to the physical system, collect the feedback response after execution, compare and verify it with the digital twin evolution path in real time, and automatically update the model confidence and child twin weight parameters; The scenario trigger module is used to simulate non-steady-state operating scenarios and perform parallel disturbance simulation on multiple sub-models in the system twin network to generate a global risk map in advance; The control module is used to perform structural adjustments to the twin network structure based on execution feedback and simulation results.

[0005] Preferably, the modeling module specifically includes: according to the physical boundaries, data mobility and control dependencies of the functional blocks in the park, a structurally variable self-organizing network algorithm is used to perform unsupervised clustering of the perception data, and construct multiple sub-digital twin models with local autonomous update mechanisms; through time series correlation analysis and functional dependency mapping, a multi-dimensional logical relationship map between each sub-model is constructed, and a dynamic multi-scale graph convolution mechanism is used to form a system-level digital twin topology network with cross-domain connectivity.

[0006] Preferably, the three-dimensional evolution engine includes: Based on the long-term operation logs and state transition records of physical objects in the park, a state-event dual-channel graph modeling method is used to construct a temporal behavior graph of the objects. The graph simultaneously records the spatial migration path and the causal links of key behaviors. A set of environmental disturbance factors, including weather fluctuations, load mutations, and event cascades, is introduced. The disturbance amplitude is mapped through a Beta distribution model and integrated into the transition probability matrix of the behavior map. A finite scenario iterative simulation mechanism is adopted to simulate multiple potential evolution paths under disturbance conditions and generate multiple prediction trajectories, which cover spatial behavior deviation, abnormal state evolution and resource dissipation patterns.

[0007] Preferably, the strategy generation module includes: Combining the multi-path prediction trajectory output by the 3D evolution engine, a space-time clustering method is used to locate abnormal clusters in the trajectory and extract abnormal sub-models; For abnormal sub-models, we call its built-in local criterion library and use the structure-function-behavior ternary model deconstruction method to analyze its vulnerability causes and extract the dominant factors and influencing paths that cause the abnormality. Divide the entire park into multi-layered strategic response blocks based on spatial location, response time, and control level. Formulate strategic combinations according to priority principles to form a hierarchical strategy set. Using the selection mechanism based on scenario pressure thresholds, the strategy combination with the least resource impact and the best intervention efficiency is screened out and output to the platform scheduling layer as the final execution.

[0008] Preferably, the execution feedback module includes: Mapping the hierarchical policy set into execution instructions matching each control terminal of the physical system through a protocol gateway; The response behavior data of physical objects is collected in real time through a multi-channel sensor cluster, and the data is synchronized to the twin platform based on the strategic time window to form a control sample; A graph matching mechanism is used to compare the node-edge structure of the collected response and the predicted evolution path to identify the behavior deviation points and response lag sections; According to the deviation magnitude and strategy response error, the confidence weight of the corresponding sub-twin model is dynamically adjusted, and the fuzzy weight incremental update mechanism is used to improve the model evolution accuracy.

[0009] Preferably, the control module includes: Based on the structural differences between execution feedback data and twin evolution trajectories, adaptive similarity matrix analysis is used to identify connection nodes and response paths in the topology that exhibit high deviation rates; For the identified weak connections, a heterogeneous functional synergy calculation method is used to construct the functional attribute vectors between the sub-twin models. Combined with the historical synergy frequency, the functional similarity index of the potential alternative edges is calculated. The candidate set of structural reconstruction is screened out according to the synergy ranking results, and the edge replacement, merging or adding scheme is determined based on the path redundancy and response reliability requirements.

[0010] The present invention also provides a virtual-reality interaction method based on the smart park digital twin management platform, comprising: S100 collects multi-source heterogeneous data of various physical objects in the park, including structural status, equipment behavior, environmental disturbances, and user behavior information. It is divided into high-frequency, medium-frequency, and low-frequency data layers based on perception accuracy and real-time performance, and then integrated and transmitted to the digital twin platform. S200, based on the collected data, a multi-scenario driving mechanism is used to synchronously build digital twin models of multiple entities; S300: Generate a control strategy based on the predicted behavior trajectory output by each digital model, and synchronously send the strategy and strategy identification code to the physical system through the park control system, while recording the predicted response trajectory as a control channel; S400: Collect entity response data, compare it with the model prediction response at the node path level, identify behavior deviations and response abnormalities, and form a deviation map; S500, based on the deviation map, establishes error learning weights according to the dimensions of region, equipment and time period, and adopts a confidence weight-driven progressive adjustment mechanism for the model parameters to achieve dynamic correction and continuous optimization of the twin model.

[0011] Preferably, S101, according to the entity object type, concurrently perceive the buildings, equipment, environment and human behavior in the park; S102. Assign high-frequency, medium-frequency, or low-frequency labels to different collected data based on their perception rate, event suddenness, and information value density, for use in determining fusion priority in subsequent processing; S103, normalizing the format and deconstructing the meaning of various types of data through an adaptive semantic mapping template to generate data units in a unified format; S104. Use spatiotemporal labels and semantic mapping structures to perform asynchronous and synchronous fusion processing on data from different frequency layers to generate a structured fusion data set and transmit it to the digital twin platform.

[0012] Preferably, the S500 includes: S501. Based on the comparison results between the entity response and the predicted trajectory, the deviation information is segmented by spatial region, device type, and time period to form a multi-dimensional deviation sub-map; S502: Establish an error weight model for each deviation sub-graph, and generate a dynamic learning weight matrix based on the deviation frequency, response lag amplitude, and impact range; S503. Mapping the learning weights to the parameter set of the corresponding twin model, and re-estimating the credibility of the key behavior functions and transition probabilities in the model through the confidence scoring mechanism; S504: Based on the weight change trend, the model parameters are gradually adjusted using an incremental correction strategy.

[0013] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. At the system level, this invention proposes a smart park digital twin management platform and virtual-reality interaction method capable of structural evolution, autonomous learning, and closed-loop virtual-reality linkage. Through core technical steps such as multi-frequency heterogeneous perception, scenario-driven modeling, strategy prediction and execution feedback comparison, deviation map generation, confidence weight learning, and progressive correction, the platform can achieve dynamic adaptive optimization of digital twin models during continuous operation, breaking through the technical bottlenecks of existing twin systems such as static modeling, response rigidity, and model aging, significantly improving the generalization capability and timeliness of digital twin systems.

[0014] 2. At the application level, this invention supports refined modeling and flexible control of complex elements within a park, including buildings, equipment, environment, and human behavior. It is widely applicable to various management scenarios, including energy optimization, security assurance, intelligent operation and maintenance, and anomaly warning. By introducing multi-model collaboration, policy identification comparison, response path comparison, and structural network evolution mechanisms, it achieves a shift from passive monitoring to active prediction and policy self-evolution, providing a highly intelligent, sustainable, and stable digital governance technology framework for new smart parks. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0016] Figure 1 Flow chart of the method of the present invention.

[0017] Figure 2 This is a flow chart of the management platform system module of the present invention. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] Example 1, please refer to Figure 1 As shown, the smart park digital twin management platform described in this embodiment includes: A modeling module that constructs multiple self-organizing sub-digital twin models, which are dynamically associated with each other through a logical topology diagram to form a system-level digital twin network. A three-dimensional evolution engine is used to construct a temporal behavior map of physical objects based on historical data, introduce environmental disturbance factors for multi-field coupling prediction, and output multi-path trend trajectories for strategy evaluation; A strategy generation module, which combines the prediction trajectory of the evolution engine with the local anomaly criteria of each sub-model to automatically identify potential weaknesses and generate a hierarchical strategy set for space-time partitioning; The execution feedback module is used to send the policy set to the physical system, collect the feedback response after execution, compare and verify it with the digital twin evolution path in real time, and automatically update the model confidence and child twin weight parameters; The scenario trigger module is used to simulate non-steady-state operating scenarios and perform parallel disturbance simulation on multiple sub-models in the system twin network to generate a global risk map in advance; The control module is used to perform structural adjustments to the twin network structure based on execution feedback and simulation results.

[0020] In this embodiment, the smart campus is divided into several functional areas, such as office areas, energy management areas, security monitoring areas, and smart parking areas. The platform first extracts multidimensional sensory data from each area based on sensor layout, equipment operation and maintenance relevance, and control logic flow, including ambient temperature, power load, security event frequency, and personnel distribution.

[0021] The system then introduced a structurally variable self-organizing network (SV-SOM) algorithm. Unlike traditional K-Means or hierarchical clustering methods, this algorithm automatically determines the number and boundaries of sub-model divisions based on the topological stability and structural coupling strength of the data. During training, the model node structure is scalable and integrated with the actual spatial layout of the campus, enabling the automatic generation of multiple autonomous sub-digital twin models.

[0022] To ensure each sub-model possesses a degree of autonomous behavior, the platform incorporates local control units based on finite game strategies. For example, a local game controller in the energy sub-model can formulate load balancing strategies based on the energy consumption game of various air conditioning devices; in the security sub-model, multi-camera resource preemption enables self-negotiation.

[0023] Each sub-model optimizes its local objectives and considers the degree of coupling with neighboring sub-models (e.g., equipment sharing, power connectivity, etc.) to form a weighted decision-making game graph. This mechanism enables multiple sub-models to self-optimize and coordinate control through local judgment without intervention from the global control system.

[0024] After each sub-model is constructed, the platform uses temporal correlation analysis to combine the temporal synchronization, event propagation paths, and functional dependencies of each model to generate an initial multi-dimensional logical topology. To enable intelligent cross-domain connectivity and knowledge transfer, this topology is further embedded in a multi-scale graph convolutional neural network to analyze the influence propagation relationships across sub-models.

[0025] During platform operation, this logical topology is not fixed. The system periodically evaluates indicators such as communication frequency, response latency, and joint feedback accuracy between sub-models. It also applies a fuzzy boundary fusion algorithm to adaptively expand or contract fuzzy boundary areas.

[0026] For example, when the smart parking system frequently coordinates with the security system (such as nighttime intrusion detection and gate closure), the platform can dynamically increase the edge weights between the two or even merge sub-models to form a more stable coupled unit. Conversely, when certain models have extremely low data correlation for a long time, they can be disconnected or scheduled independently.

[0027] In this embodiment, in order to achieve high-precision prediction of the operating trends of key physical objects in the park (such as cold and heat source systems, security gateways, main power hubs, etc.), the platform introduced a three-dimensional evolution engine, which is mainly composed of "behavior trajectory modeling sub-unit", "disturbance implantation module", "multi-field coupling simulation engine" and "path evaluation module".

[0028] The platform extracts state data (such as running / standby / abnormal), event data (such as switch operation, alarm records) and location migration data from the object's operation log. Through the state-event dual-channel graph modeling method, a multi-dimensional graph structure is constructed: Graph nodes represent object states or positions; Graph edges represent event-driven state transitions; Each edge contains a probability parameter and an event trigger identifier.

[0029] This map can clearly show the object's "time-behavior-space" change path.

[0030] To achieve the ability to predict responses to complex real-world disturbances, the platform builds an environmental disturbance factor library, including: meteorological disturbances (sharp temperature changes, sharp fluctuations in humidity); Load disturbances (holiday peaks, power surges); System-level cascading events (e.g., air conditioning system failure leading to increased energy consumption).

[0031] Each perturbation factor uses a Beta distribution model to represent the uncertainty of its perturbation magnitude, and the perturbation impact matrix is ​​constructed by combining the transition edges in the behavior graph. Before prediction, the system integrates the perturbation model with the graph to form a multi-scenario perturbation transition probability map.

[0032] The platform launches a limited scenario iterative simulation mechanism, runs multiple rounds of evolutionary simulations with different perturbation combinations as initial conditions, and outputs multiple potential trend trajectories. Each trajectory includes: Spatial path deviation trends (e.g., movement trajectory variations); State evolution path (e.g., from normal to potential failure); Resource consumption change trends (such as the rate of increase in power consumption).

[0033] Finally, the platform sends the trajectory to the path assessment module, scores and ranks it according to dimensions such as impact level, system risk exposure and resource utilization efficiency, and provides it to the platform scheduling module in the form of a graphical report.

[0034] For example, if a trajectory faces an increased risk of cooling plant overload due to the "high temperature + sudden increase in power load" disturbance, the system recommends adjusting the cooling strategy and prioritizing switching to backup equipment, while also reducing the air conditioning power in non-core areas to prevent global failures.

[0035] The three-dimensional evolution engine in this embodiment not only supports complex predictions under multiple disturbance conditions, but also has the ability to explain behavioral trajectories and generate closed-loop strategies. Compared with traditional linear prediction models, it significantly improves the foresight and adaptability of park-level intelligent decision-making.

[0036] In this embodiment, in order to achieve adaptive control of the operating status of complex campuses, the platform deploys a multi-dimensional strategy generation mechanism based on predicted trajectories and local anomaly identification. It relies on the state feedback of the three-dimensional evolution engine and the sub-digital twin model to complete a set of "identification-analysis-construction-optimization" strategy generation processes.

[0037] The system first receives multiple trend trajectories output by the evolution engine, including trends in rising temperatures, sudden power surges, and functional chain disruptions. Based on these trends, the platform applies spatial-temporal clustering methods (e.g., based on ST-DBSCAN) to cluster regions with high-frequency deviations within these trajectories, rapidly identifying multiple high-density areas of abnormal load, such as the simultaneous occurrence of abnormal loads between the air conditioning system and the power distribution nodes. The identified results are then localized to specific sub-digital twin model units, such as the "central cooling plant model" and the "office area load model," which serve as targets for strategic intervention.

[0038] The platform then accesses the pre-set local anomaly criteria library in each target sub-model (for example, temperature fluctuation > ±3°C is considered thermal stability out of control; load exceeding 10% is the critical value). Using the structure-function-behavior ternary deconstruction method, the model is divided into: The structural layer (e.g., chiller plant topology); Functional layer (cooling efficiency, valve status); Behavioral layer (control response delay and intervention effect deviation).

[0039] Based on this, the platform determines the main cause of the abnormality. For example, "valve response delay + slow fan operation" may be a combined cause of the sudden drop in air conditioning efficiency.

[0040] The system builds multi-layered strategic response blocks based on the physical location of the anomaly model, control response delay, and system level of the affected objects, such as: Level 1 strategic area (core system, requiring immediate response); Secondary strategy area (secondary system, requiring coordinated adjustments); Buffer zone (small impact, just observe).

[0041] Within each block, the system applies the "urgent response + wide diffusion" priority model to generate multiple types of strategies, including: Control strategies (e.g., reducing chiller frequency); Load limiting strategies (such as shutting down equipment in non-core areas); Avoidance strategies (such as redistributing load flow).

[0042] The system applies a scenario pressure threshold screening mechanism to all candidate strategy combinations: based on the time required for strategy intervention, the amount of resources affected, and the degree of local conflict, it calculates the "comprehensive cost-response index" of each strategy combination, and finally selects the best one as the execution plan.

[0043] For example, in the scenario of high temperature + load fluctuation, the system selects the "prioritized unloading of non-office area loads + delayed opening of some air-conditioning circuits" solution to replace the original "load limit in the entire area" high-impact strategy to achieve low-disturbance, high-efficiency intervention.

[0044] In this embodiment, in order to achieve closed-loop evaluation of strategy execution effects and dynamic correction of digital twin model accuracy, the smart park platform deploys an execution feedback response tracking and model confidence correction mechanism, enabling the digital model to have perception-comparison-correction capabilities during operation.

[0045] When the hierarchical policy set generated by the platform is confirmed as an executable solution by the scheduling layer, the system first formats the policy commands into a standardized instruction set that adapts to the control interfaces of different physical devices. For example: The temperature control strategy is mapped to the Modbus command of the air conditioning controller; The energy diversion strategy is converted into PLC logic control rules.

[0046] Each set of execution instructions is assigned a unique execution identification code (SID) to associate subsequent feedback data with this strategy operation.

[0047] The platform's multi-channel sensor cluster (including power sensors, temperature sensors, and control feedback modules) synchronously uploads real-time monitoring data to the digital twin platform. The system sets a policy response time window (e.g., 5 to 30 minutes after execution) and extracts feedback data corresponding to the SID only within this time period, creating a feedback data sample set.

[0048] For example, for the air conditioner's "reduced load operation" strategy, the system collects indicators such as the air conditioner current change, room temperature change curve, and air valve response time before and after execution, and performs data cleaning and time alignment.

[0049] The system calls the original predicted trajectory in the three-dimensional evolution engine and structures it as a node-edge behavior evolution graph. At the same time, it converts the feedback sample into the actual execution path graph. Through the graph matching and comparison mechanism, the two graphs are compared to identify the following information: Behavior deviation node (such as the expected state does not appear); Response delay path segment (e.g., air valve response lag of 5 minutes); Abnormal transfer paths (e.g., temperature increases when it should be cooling).

[0050] The system quantifies the deviation information into a response error matrix as the basis for subsequent model correction.

[0051] For sub-digital twin models with errors, the platform applies a fuzzy weight incremental update mechanism. Instead of directly replacing modeling parameters, it fine-tunes the following indicators by "enhancing confidence for low errors and reducing weights for high errors": State transition probability; The model outputs a confidence score; Behavioral pattern confidence factor.

[0052] For example, if the same strategy predicts a deviation greater than a threshold of 5% in the model three times in a row, the evolution output of the sub-model will be marked as "moderately credible" and the system will deploy other related models as auxiliary prediction references.

[0053] This embodiment simulates sudden non-steady-state operation scenarios, conducts parallel disturbance tests on each sub-digital twin model in the platform, and outputs a global risk mapping map to achieve early prediction, high-sensitivity area identification and linked risk monitoring.

[0054] The platform has a historical event library and external interfaces, continuously collecting data from security systems, operation and maintenance systems, and third-party platforms, such as extreme temperature records, equipment cascading failure cases, and records of people gathering at large-scale events.

[0055] The system encodes events into sets of unsteady factor vectors based on time, space, and coupling strength. Each factor vector contains the trigger type (e.g., high temperature, power outage), scope (a set of model numbers), and disturbance level (mild / moderate / severe), which serve as simulation startup conditions.

[0056] During the simulation initialization phase, the platform uses an asynchronous perturbation injection mechanism to concurrently apply perturbation signals to multiple selected sub-twin models with set timing delays and intensity changes. For example: Simulate "condenser efficiency drop" in the air conditioning system submodel; Simulate "load shock + lighting anomaly" in the lighting system sub-model; Apply “density mutation” perturbation in the people flow sub-model.

[0057] This mechanism allows the simulation to not follow linear synchronization, but instead present a multi-path evolution trend of "event cascade + asymmetric conduction", which is close to real complex scenarios.

[0058] During the simulation process, the platform records the state change sequence of each sub-model (such as a sudden increase in energy consumption curve, equipment operating frequency deviation, response delay fluctuation, etc.). Then, the fuzzy collaborative aggregation method is used to uniformly classify the responses of multiple models: Clustering out common abnormal behavior patterns; Extract suspicious coupling paths (such as abnormal conduction from air conditioning to lighting systems); Locate potential abnormal transmission chains.

[0059] This process generates a set of “anomaly response maps” for risk hotspot identification.

[0060] The platform overlays the above map information onto the park's spatial GIS structure to generate a real-time updated global risk map. The map is annotated according to the following elements: Highly sensitive areas: areas that respond strongly to disturbances and frequently experience abnormalities; Response hysteresis area: risk transmission hysteresis and monitoring blind spot area; Linkage hotspot: multi-model coupling high-frequency channel node.

[0061] This map serves as the early warning input for the subsequent strategy generation module, providing basic support for preventive maintenance, emergency strategy deployment and resource scheduling.

[0062] In order to improve the stability and adaptability of the digital twin platform in long-term operation in a complex campus environment, this embodiment provides a control module implementation solution to realize the platform's structural-level self-evolution and reorganization capabilities.

[0063] During the daily operation of the platform, the system continuously compares the simulated prediction trajectory with the actual execution feedback data to build a difference map between the two. The control module uses an adaptive similarity matrix analysis method to identify structural weaknesses from the following two aspects: The response coordination between twin nodes continues to decline; There is a high incidence of errors or delayed propagation in the transmission paths across sub-models.

[0064] For example, during five consecutive strategy executions, the data transmission stability between the cold station model and the load control model dropped to 60%, triggering the reconstruction condition.

[0065] The platform calculates the heterogeneous functional synergy (HFC) of the identified unstable connection regions based on the functional attributes of each sub-model and historical collaborative behavior data. The HFC scoring system is used to screen a set of candidate structural links that can be replaced or supplemented.

[0066] The method for obtaining heterogeneous functional synergy is to extract the functional attribute characteristics of each sub-twin model Mi, including but not limited to: Type of managed object (e.g. energy, security, environment); Control response mode (such as real-time control, timed scheduling, event triggering); Behavioral output type (e.g., Boolean event, continuous variable, probabilistic prediction); Historical linkage frequency (number of times it responds together with other models).

[0067] Forming the feature vector for each model ,in is the encoding value of the kth functional feature.

[0068] Select any two models , calculate their functional similarity. Use weighted cosine similarity algorithm: ;in, is the weight coefficient of the kth functional feature, which is preset according to its importance in the collaborative task (for example, the weight of "control response method" is greater than that of "object type"); n is the total number of items.

[0069] In order to reflect the influence of historical linkage experience, the system introduces a behavior weighting coefficient , which is based on the frequency or coverage of the two models participating in the linkage in the historical strategy: ;in For the model The number of coordinated responses in the past T strategy cycles. Finally, the degree of heterogeneous functional coordination is defined as: ; The higher the HFC value, the more it indicates that although the two models are structurally heterogeneous, they have potential synergy in functional performance and are suitable for being included in the candidate connection set for twin network structure reconstruction.

[0070] For example, if the lighting system and security system are found to have high-frequency resonance behavior during a nighttime emergency, the platform will automatically add them to the candidate connection path to build a redundant response chain.

[0071] The system statistics of each sub-model: Execution response timeliness; Evolution prediction accuracy; Frequency of control strategy adoption.

[0072] Combining these metrics, each model in the twin network is assigned a dynamic confidence weight. Models with higher weights are given greater control rights or participate in determining global evolution trends. For example, in the summer, the weight of the energy consumption control model increases, while in scenarios with a sudden increase in occupancy, the weight of the access management model increases.

[0073] Based on the above assessment results, the platform performs structural level operations, including: Delete the connecting edges that show instability; Add the newly generated high HFC edge; Merge highly correlated models into composite twins (e.g., air conditioning + ventilation into an HVAC composite model); All changes are pre-verified through simulation testing to ensure that the overall stability of the system is not weakened.

[0074] Ultimately, the platform automatically completes the update of the twin network structure, realizing the system evolution closed loop of "structure reconstruction-path rearrangement-functional rebalancing".

[0075] Example 2, please refer to Figure 2 As shown, the virtual-reality interaction method based on the smart park digital twin management platform described in this embodiment includes: S100 collects multi-source heterogeneous data of various physical objects in the park, including structural status, equipment behavior, environmental disturbances, and user behavior information. It is divided into high-frequency, medium-frequency, and low-frequency data layers based on perception accuracy and real-time performance, and then integrated and transmitted to the digital twin platform. S200, based on the collected data, a multi-scenario driving mechanism is used to synchronously build digital twin models of multiple entities; S300: Generate a control strategy based on the predicted behavior trajectory output by each digital model, and synchronously send the strategy and strategy identification code to the physical system through the park control system, while recording the predicted response trajectory as a control channel; S400: Collect entity response data, compare it with the model prediction response at the node path level, identify behavior deviations and response abnormalities, and form a deviation map; S500, based on the deviation map, establishes error learning weights according to the dimensions of region, equipment and time period, and adopts a confidence weight-driven progressive adjustment mechanism for the model parameters to achieve dynamic correction and continuous optimization of the twin model.

[0076] In this embodiment, to meet the digital twin platform's multi-dimensional, differentiated, and timely requirements for park entity data, the system implements a data acquisition mechanism with perceptual precision stratification and semantic fusion capabilities. This mechanism divides raw perceptual information into four data sources: structure, behavior, environment, and human factors, and achieves structured fusion through multiple steps.

[0077] The campus platform deploys dedicated sensor devices on different objects, such as: Install displacement and stress sensors on building structures; Install switch status detectors and operating frequency collectors on the equipment; Deploy temperature, humidity, air pressure and noise detection nodes at the environmental level; Deploy behavior recognition cameras and passage counting modules along the pedestrian passage paths.

[0078] These sensing terminals are connected to the edge computing gateway and output to the data bus using a unified interface protocol to ensure the integrity of the data source and real-time access.

[0079] The platform assigns frequency labels to each type of data based on three indicators (sampling frequency, event suddenness, and data value density): High-frequency data: such as equipment operating status (updated per second) and indoor environment fluctuations; Medium-frequency data: such as personnel entry and exit records, power load changes (minute level); Low-frequency data: such as building structure deformation monitoring and historical energy consumption data (daily / weekly); Various types of data are labeled “HF”, “MF”, and “LF” to provide a basis for subsequent data fusion priorities.

[0080] Because devices from different manufacturers have different output formats, the platform uses an adaptive semantic mapping template mechanism to convert raw data into a unified triple structure: Data unit = ⟨object ID, attribute label, value / unit>; This normalization mechanism is compatible with text, numerical, image, or Boolean data input and forms a logically consistent data structure.

[0081] On the data center side, the platform fuses data from different sources based on data timestamps, frequency tags, and spatial locations. The fusion logic is as follows: High-frequency data is used first to drive real-time updates of the twin state; Medium and low frequency data serve as trend correction and background modeling support; The data under the same object are packaged into fused data blocks according to the timeline for subsequent model module calls.

[0082] The fused data structure not only retains the source attributes, but also carries contextual spatiotemporal information, which facilitates model training and dynamic simulation.

[0083] In this example, the digital twin platform first extracts scene attributes from different types of data in the park. Taking office buildings, power distribution rooms, and air conditioning systems as examples, the system extracts the following typical scene attribute factors: Type of space area (enclosed / semi-open / outdoor); Object category (structural class / operating equipment / user behavior); Control logic complexity (single point / linkage / automatic feedback); The platform uses a rule engine to combine and classify the above factors to form "scenario labels", such as: "office building-HVAC-cyclical load type".

[0084] Based on the label, the platform automatically matches the corresponding template from the twin model library, including the model's: Parameter structure (temperature, current, frequency of use); Behavioral functions (dynamic changes in heat load); Environmental coupling (linkage with external meteorological systems); Subsequently, the platform adopted a nested modeling mechanism to simultaneously construct the main model and its corresponding sub-model in the same scenario, such as: "central air-conditioning main model + regional terminal sub-model", and established a linkage interface between the two through data mapping relationships, ultimately forming a set of digital twin models with scenario adaptability.

[0085] During each strategy generation cycle, the platform processes the predicted behavior trajectories evolved by each sub-model, including state transition sequences, triggering condition points, and expected response trends. For example: The cooling water pump is predicted to experience a power increase within 30 minutes; The terminal air supply is expected to switch from high speed to medium speed within 10 minutes.

[0086] The platform structures these trajectories into a trajectory graph and generates strategies (such as reducing pump speed or adjusting air conditioning volume). Each strategy is assigned a unique identification code (such as SID-230615-01) for feedback execution.

[0087] The control strategy and strategy identification code are sent to the corresponding device controller through the communication interface; at the same time, the platform retains the predicted trajectory as a "channel control sample" in the strategy cache module for subsequent response comparison.

[0088] Within the set time window after the policy is executed (e.g., 10 to 30 minutes), the platform continuously collects entity response data, including whether the control instructions are executed and whether the status changes after execution are in line with expectations.

[0089] The collected response sequence is structured into a "response trajectory graph", and the platform calls the "path-level structure comparison algorithm" to perform node-edge comparison on the graph with the previously saved predicted trajectory graph. For example: Predicted path: A → B → C → D (normal load reduction); Actual response: A → B → X (abnormal pause); The system identifies problems such as "missing point C", "path deviation", and "response time delayed by 5 minutes" and marks them in the "deviation map".

[0090] The deviation map will be used to: Model confidence adjustment; Control strategy correction optimization; Suggested input for restructuring of sub-models.

[0091] In this embodiment, the Smart Park digital twin platform systematically learns execution deviations to build a confidence-driven, dimensionally decomposed, and dynamically evolving model-adaptive correction mechanism. Based on the deviation map generated by comparison, this mechanism applies weighted corrections based on different dimensions, enabling continuous and precise optimization of the digital twin model.

[0092] After obtaining the deviation graphs after executing multiple strategies, the platform first divides the graphs according to the following three dimensions: Spatial dimension: for example, "Office Area East Wing" vs. "Equipment Room Area B"; Equipment dimensions: such as chillers, fan coil units, electricity sub-meters, etc.; Time dimension: such as weekday peak (9-11 am) and nighttime trough; Through segmentation, each sub-graph reflects the deviation behavior characteristics under specific conditions, such as "the chiller in area B has a response lag when operating at night."

[0093] The system counts the frequency and magnitude of deviations (e.g., temperature error greater than 3°C) and the scope of influence (single device / multi-device / cross-model) in each sub-graph, and calculates the error learning weight W based on this, satisfying: Where: F is the deviation frequency; D is the response delay amplitude; S is the number of affected system nodes; and λi is an empirical weighting parameter. This weight is used to guide the priority of model parameter correction.

[0094] The platform maps the learning weights to relevant parameters in the twin model (such as behavior transition probability and state response function coefficients), and re-estimates the confidence scores of these parameters based on historical execution accuracy to form a "high confidence parameter set" and a "low confidence parameter set".

[0095] For example, if the function of "air conditioning air supply speed → temperature drop" deviates frequently during peak hours, the confidence of the function is reduced and it is prioritized in the correction queue.

[0096] For low-confidence parameters, the platform adopts an incremental adjustment strategy, that is, it does not directly replace the model structure or all parameters, but instead: Inject corrections into the simulation cycle in batches; Each adjustment range is controlled within the dynamic tolerance range; Real-time monitoring of adjustment effects and system stability; In this way, the system achieves "light disturbance, high convergence" modeling accuracy evolution, avoiding model oscillation or control logic instability.

[0097] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. The smart park digital twin management platform features: include: A modeling module that constructs multiple self-organizing sub-digital twin models, which are dynamically associated with each other through a logical topology diagram to form a system-level digital twin network; A three-dimensional evolution engine is used to construct a temporal behavior map of physical objects based on historical data, introduce environmental disturbance factors for multi-field coupling prediction, and output multi-path trend trajectories for strategy evaluation; A strategy generation module, which combines the prediction trajectory of the evolution engine with the local anomaly criteria of each sub-model to automatically identify potential weaknesses and generate a hierarchical strategy set for space-time partitioning; The execution feedback module is used to send the policy set to the physical system, collect the feedback response after execution, compare and verify it with the digital twin evolution path in real time, and automatically update the model confidence and child twin weight parameters; The scenario trigger module is used to simulate non-steady-state operating scenarios and perform parallel disturbance simulation on multiple sub-models in the system twin network to generate a global risk map in advance; The control module is used to perform structural adjustments to the twin network structure based on execution feedback and simulation results.

2. The smart park digital twin management platform according to claim 1, characterized in that: The modeling module specifically includes: based on the physical boundaries, data mobility and control dependencies of the functional blocks in the park, a structurally variable self-organizing network algorithm is used to perform unsupervised clustering of the perception data, and to construct multiple sub-digital twin models with local autonomous update mechanisms; through time series correlation analysis and functional dependency mapping, a multi-dimensional logical relationship map between each sub-model is constructed, and a dynamic multi-scale graph convolution mechanism is used to form a system-level digital twin topology network with cross-domain connectivity.

3. The smart park digital twin management platform according to claim 1, characterized in that: The three-dimensional evolution engine includes: Based on the long-term operation logs and state transition records of physical objects in the park, a state-event dual-channel graph modeling method is used to construct a temporal behavior graph of the objects. The graph simultaneously records the spatial migration path and the causal links of key behaviors. A set of environmental disturbance factors, including weather fluctuations, load mutations, and event cascades, is introduced. The disturbance amplitude is mapped through a Beta distribution model and integrated into the transition probability matrix of the behavior map. A finite scenario iterative simulation mechanism is adopted to simulate multiple potential evolution paths under disturbance conditions and generate multiple prediction trajectories, which cover spatial behavior deviation, abnormal state evolution and resource dissipation patterns.

4. The smart park digital twin management platform according to claim 1, characterized in that: The strategy generation module includes: Combining the multi-path prediction trajectory output by the 3D evolution engine, a space-time clustering method is used to locate abnormal clusters in the trajectory and extract abnormal sub-models; For abnormal sub-models, we call its built-in local criterion library and use the structure-function-behavior ternary model deconstruction method to analyze its vulnerability causes and extract the dominant factors and influencing paths that cause the abnormality. Divide the entire park into multi-layered strategic response blocks based on spatial location, response time, and control level. Formulate strategic combinations according to priority principles to form a hierarchical strategy set. Using the selection mechanism based on scenario pressure thresholds, the strategy combination with the least resource impact and the best intervention efficiency is screened out and output to the platform scheduling layer as the final execution.

5. The smart park digital twin management platform according to claim 4 is characterized by: The execution feedback module includes: Mapping the hierarchical policy set into execution instructions matching each control terminal of the physical system through a protocol gateway; The response behavior data of physical objects is collected in real time through a multi-channel sensor cluster, and the data is synchronized to the twin platform based on the strategic time window to form a control sample; A graph matching mechanism is used to compare the node-edge structure of the collected response and the predicted evolution path to identify the behavior deviation points and response lag sections; According to the deviation magnitude and strategy response error, the confidence weight of the corresponding sub-twin model is dynamically adjusted, and the fuzzy weight incremental update mechanism is used to improve the model evolution accuracy.

6. The smart park digital twin management platform according to claim 1, characterized in that: The control module includes: Based on the structural differences between execution feedback data and twin evolution trajectories, adaptive similarity matrix analysis is used to identify connection nodes and response paths in the topology that exhibit high deviation rates; For the identified weak connections, a heterogeneous functional synergy calculation method is used to construct the functional attribute vectors between the sub-twin models. Combined with the historical synergy frequency, the functional similarity index of the potential alternative edges is calculated. The candidate set of structural reconstruction is screened out according to the synergy ranking results, and the edge replacement, merging or adding scheme is determined based on the path redundancy and response reliability requirements.

7. A virtual-reality interaction method based on a smart park digital twin management platform, used to implement the smart park digital twin management platform according to any one of claims 1 to 6, characterized in that: include: S100 collects multi-source heterogeneous data of various physical objects in the park, including structural status, equipment behavior, environmental disturbances, and user behavior information. It is divided into high-frequency, medium-frequency, and low-frequency data layers based on perception accuracy and real-time performance, and then integrated and transmitted to the digital twin platform. S200, based on the collected data, a multi-scenario driving mechanism is used to synchronously build digital twin models of multiple entities; S300: Generate a control strategy based on the predicted behavior trajectory output by each digital model, and synchronously send the strategy and strategy identification code to the physical system through the park control system, while recording the predicted response trajectory as a control channel; S400: Collect entity response data, compare it with the model prediction response at the node path level, identify behavior deviations and response abnormalities, and form a deviation map; S500, based on the deviation map, establishes error learning weights according to the dimensions of region, equipment and time period, and adopts a confidence weight-driven progressive adjustment mechanism for the model parameters to achieve dynamic correction and continuous optimization of the twin model.

8. The virtual-reality interaction method based on the smart park digital twin management platform according to claim 7 is characterized by: include: S101. Based on the entity object type, concurrently perceive the buildings, equipment, environment, and human behavior in the park; S102. Assign high-frequency, medium-frequency, or low-frequency labels to different collected data based on their perception rate, event suddenness, and information value density, for use in fusion priority determination in subsequent processing; S103, normalizing the format and deconstructing the meaning of various types of data through an adaptive semantic mapping template to generate data units in a unified format; S104. Use spatiotemporal labels and semantic mapping structures to perform asynchronous and synchronous fusion processing on data from different frequency layers to generate a structured fusion data set and transmit it to the digital twin platform.

9. The virtual-reality interaction method based on the smart park digital twin management platform according to claim 7 is characterized by: include: The S500 includes: S501. Based on the comparison results between the entity response and the predicted trajectory, the deviation information is segmented by spatial region, device type, and time period to form a multi-dimensional deviation sub-map; S502: Establish an error weight model for each deviation sub-graph, and generate a dynamic learning weight matrix based on the deviation frequency, response lag amplitude, and impact range; S503. Mapping the learning weights to the parameter set of the corresponding twin model, and re-estimating the credibility of the key behavior functions and transition probabilities in the model through the confidence scoring mechanism; S504. Based on the weight change trend, the model parameters are gradually adjusted using an incremental correction strategy.

Images