Intelligent park whole life cycle management system and method based on digital twinning and internet of things

Abstract

The application discloses a full life cycle management system and method for a smart park based on digital twinning and the Internet of Things, relates to the technical field of smart park management, and comprises a sensing edge module, a data management module, an intelligent analysis module, a life cycle module and a twinning modeling module.The application supports multi-protocol access and edge computing capability, greatly improves data transmission efficiency and stability, the intelligent analysis module outputs accurate analysis results by constructing a multi-class feature matrix and a deep multi-task joint modeling mechanism, and provides data support and model guidance for park dynamic management and intelligent decision-making, the life cycle module fuses Kepler optimization algorithm, multi-agent reinforcement learning and simulated annealing algorithm, establishes a collaborative optimization mechanism, realizes the combination of global search and local fine-tuning of resource scheduling, and effectively optimizes energy consumption, response time, space utilization and safety risks, and the twinning modeling module constructs a three-dimensional model of the park, improves system interactivity and operability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart park management, and particularly relates to a smart park full life cycle management system and method based on digital twinning and Internet of Things. BACKGROUND

[0002] With the development of smart cities and intelligent parks, park management systems are gradually evolving towards digitization, networking and intelligentization. In order to achieve fine management and optimal allocation of resources in the park, more and more systems begin to introduce Internet of Things (IoT) technology and building information model (BIM) and other means to realize comprehensive sensing and centralized management of equipment, personnel, environment and other information.

[0003] However, the existing technology still has several technical bottlenecks in the full life cycle management of the park. First, a large number of heterogeneous devices deployed inside the park lack unified standards in the process of data collection, transmission and access, resulting in complex data structure, non-uniform format and low processing efficiency. Second, edge computing nodes cannot effectively guarantee data integrity in unstable network scenarios, and the upload mechanism usually lacks intelligent scheduling capability, causing waste of bandwidth resources. Third, the current analysis system is mostly based on a single model, which is difficult to cope with the joint modeling needs of multi-dimensional and multi-modal data involved in the operation of the park, and the analysis results are difficult to directly translate into executable management instructions, lacking practicality and response efficiency. In addition, the traditional resource configuration optimization mechanism fails to fully consider the operation characteristics of the park in different life cycle stages, lacks dynamic adaptation capability, and is difficult to achieve optimal scheduling and multi-objective balance. SUMMARY

[0004] In view of the problems of sensing island, data heterogeneity, insufficient analysis, rigid scheduling and virtual-real disconnection in traditional park management, the present application proposes a smart park full life cycle management system and method based on digital twinning and Internet of Things, which can realize data collection standardization, edge intelligent preprocessing, fusion modeling analysis, dynamic resource scheduling optimization and virtual-real linkage simulation response, so as to comprehensively improve the intelligent level and operation efficiency of park management.

[0005] The present application realizes the above-mentioned purpose through the following technical solutions:

[0006] The smart park full life cycle management system based on digital twinning and Internet of Things comprises:

[0007] A perception edge module is used to collect the running state data, environmental parameter data and personnel activity data of sensors and intelligent devices deployed in the park in real time, and to preprocess all the collected data on the edge computing node;

[0008] A data governance module is configured to structure and standardize the data preprocessed by the edge computing node, perform metadata labeling and data blood relationship identification, and form a unified format data set that can be traced;

[0009] An intelligent analysis module is configured to model and analyze the unified format data set, output analysis results, and transmit the analysis results to a life cycle module and a twin modeling module;

[0010] The life cycle module is configured to combine park life cycle stage data and subsystem real-time running state data, comprehensively evaluate the analysis results, adopt a collaborative optimization mechanism of fusion Kepler optimization algorithm, multi-agent reinforcement learning algorithm and simulated annealing algorithm, execute resource scheduling strategy and multi-objective optimization strategy, and output stage management instructions and resource configuration data.

[0011] The twin modeling module is configured to construct a three-dimensional digital model of the park based on building information modeling, geographic information system data and generative artificial intelligence technology, receive data collected by the perception edge module, analysis results of the intelligent analysis module, and stage management instructions and resource configuration data output by the life cycle module, and realize virtual-real synchronous updating, visualization of analysis results and simulation response of stage management instructions based on the three-dimensional data model.

[0012] As a preferred scheme of the present application, the perception edge module comprises:

[0013] A multi-protocol access unit is configured to be compatible with RS485, MODBUS, ZigBee, LoRa and NB-IoT communication protocols, and realize data access of various sensors and intelligent devices deployed in the park;

[0014] An edge computing node is configured to perform preprocessing operations on the collected raw data, including data format conversion, denoising and outlier filtering, compression and packaging, and time synchronization;

[0015] A cache management unit is configured to locally store the processed data when the network is abnormal, and perform time series labeling;

[0016] A network interruption and data transmission unit is configured to perform data loss judgment and retransmission according to the data timestamp and check code after the network is restored.

[0017] As a preferred scheme of the present application, the perception edge module further comprises:

[0018] An edge intelligent engine is embedded with a lightweight AI model, and is configured to perform preliminary abnormality detection and event judgment on the collected raw data;

[0019] A data privacy processing unit is configured to locally process sensitive features in images and location information by using regional blurring and label desensitization algorithms.

[0020] An edge uploading scheduling unit is configured to sort data according to a calculation result of an adaptive uploading scheduling function, and sequentially upload high-priority data to a data governance module and a twin modeling module.

[0021] The adaptive uploading scheduling function is as follows:

[0022]

[0023] wherein, U i is an uploading priority score of the ith data packet; P i is a data packet importance score; Q i is a current bandwidth availability factor; R i is a predicted importance index; λ is a timeliness decay coefficient, T i is a time difference between data packet generation and a current time; E i is a unit data energy consumption; D i is a data uploading delay; α is an energy consumption penalty weight, β is a delay penalty weight, γ and δ are congestion-aware adjustment factors, and C is a network congestion coefficient.

[0024] As a preferred scheme of the present application, the data governance module comprises:

[0025] A type identification unit is configured to perform format determination and source classification on data streams uploaded by the perception edge module, and extract device identifiers, timestamps and location information of the data.

[0026] A field standardization unit is configured to perform field unification and value range standardization processing on different source data according to a preset field mapping rule and unit conversion logic, and the field mapping rule is dynamically updated by a meta-learning mechanism to adapt to different device types and data structures, form a field template library and support automatic evolution.

[0027] The meta-learning mechanism comprises: extracting structural features and context information, training a field standardization meta-model, performing standard field prediction matching, updating model parameters according to feedback results and synchronously updating the field template library.

[0028] A data fusion unit is configured to perform format conversion on semi-structured or multi-format perception data, and perform fusion processing on multiple data sources by a time window-based alignment algorithm, and output structured data in a unified format.

[0029] As a preferred scheme of the present application, the data governance module further comprises:

[0030] A metadata labeling unit is configured to label structured data to generate metadata labels including data source device identification, collection timestamp, spatial position information, data upload node identification, source credibility level label, edge processing identification, data integrity score, data continuity score, and processing version number, and form a manageable data asset catalog;

[0031] A data quality evaluation unit is configured to score data quality of each data record based on multi-dimensional indexes including data integrity, logical consistency, value range fluctuation rationality, and time continuity.

[0032] A data bloodline tracking unit is configured to record processing path of data from collection, preprocessing, standardization, fusion to output, and calculate data credibility score according to the formula:

[0033] T s =f1(C p )·ω1+f2(V r )·ω2+f3(S t )·ω3+f4(L d )·ω4+f5(I t )·ω5;

[0034] In the formula, T s is data credibility score; C p is data governance chain integrity index; V r is redundancy consistency verification score; S t is time synchronization measure; L d is processing chain depth; I t is information traceability index; f1-f5 are functions for respectively performing nonlinear conversion on score factors; and ω1-ω5 are score weights dynamically generated by a machine learning model.

[0035] An output interface unit is configured to respectively provide structured data, metadata labels, data quality score, and data credibility score to an intelligent analysis module and a life cycle module.

[0036] As a preferred scheme of the application, the intelligent analysis module comprises:

[0037] A feature extraction unit is configured to extract features from the data set in the unified format according to an analysis target, and construct four types of feature matrices including device operation feature matrix, energy consumption feature matrix, personnel behavior feature matrix, and environmental factor feature matrix.

[0038] A feature encoding unit is configured to perform modal embedded representation learning on the four types of feature matrices, convert heterogeneous features into high-dimensional representation tensors in a unified vector space through a modal mapping function, and unify the time dimension using a sliding time window and a time alignment algorithm.

[0039] A multi-task joint modeling unit includes a shared encoder and multiple task decoders, supports joint training and a multi-task dynamic weight regulation mechanism, and automatically adjusts the optimization weights of each task decoder according to the task loss; the shared encoder adopts a deep residual attention mechanism to extract general semantic features from the multi-modal fusion tensor; and the task decoders are independently trained for different modeling tasks and have task-specific attention modules.

[0040] An analysis result integration unit is configured to uniformly format the output results of the task decoders into an analysis result set, which includes device operation state prediction, energy consumption trend indicators, safety event labels, and resource configuration parameters, for calling by a life cycle module and a twin modeling module.

[0041] As a preferred scheme of the present application, the task decoder specifically includes:

[0042] A device state prediction modeler is configured to receive a device operation feature matrix, extract device topology structure features using a graph neural network, jointly model with a time series Transformer, realize state trend prediction through a sliding time window, and output device operation state prediction and fault risk score.

[0043] An energy trend analysis modeler is configured to receive an energy consumption feature matrix, extract main periodic components using a periodic pattern detection mechanism combining a time series Transformer and a fast Fourier transform, and output energy use prediction values and abnormal energy consumption warnings.

[0044] A safety behavior recognition modeler is configured to receive a personnel behavior feature matrix and an environmental factor feature matrix, adopt a space-time three-dimensional Transformer structure and embed a multi-channel cross-modal attention mechanism, perform deep semantic alignment on image frames and trajectory vectors, and output abnormal behavior types and risk level heat maps.

[0045] A resource allocation parameter generation modeler is configured to fuse the four types of feature matrices and the output results of the device state prediction modeler, the energy trend analysis modeler, and the safety behavior recognition modeler, and jointly output resource configuration parameters including spatial utilization rate, energy consumption scheduling parameters, and service load priority based on multi-objective regression and resource scheduling optimization.

[0046] As a preferred scheme of the present application, the life cycle module adopts a cooperative optimization mechanism of fusing Kepler optimization algorithm, multi-agent reinforcement learning algorithm and simulated annealing algorithm, executes resource scheduling strategy and multi-objective optimization strategy, and the method comprises the following steps:

[0047] Based on the park life cycle stage data, a life cycle code is generated as a scheduling guide signal in the optimization process, and a multi-objective weight parameter matched with the current stage is calculated to construct a multi-objective optimization function F(x) adapted to the life cycle target;

[0048] A Kepler optimization algorithm is used to perform global orbital search on the resource configuration space to generate a plurality of initial candidate solutions of resource allocation;

[0049] The initial candidate solutions are subjected to fitness evaluation, a multi-objective optimization function is calculated by using a lightweight historical scoring model, low fitness solutions are eliminated, and high-quality initial solutions are retained to form an intermediate candidate solution population;

[0050] The intermediate candidate solution population is input into a multi-agent reinforcement learning system, each subsystem corresponds to an agent, and the life cycle code is embedded into the policy network of each agent as an auxiliary state vector; each agent performs action selection based on the subsystem state vector, updates the policy to optimize four types of indexes in the multi-objective optimization function, including energy consumption fluctuation rate, task response time, spatial resource congestion degree and risk level, and outputs a set of reinforcement candidate solutions optimized by the policy;

[0051] Sparse representation is introduced to the current optimal solution in the reinforcement candidate solution set, Gaussian perturbation is performed based on the sparse solution vector, and the Metropolis criterion is used to determine whether the new solution replaces the original solution, and a local optimal solution processed by the simulated annealing algorithm is output;

[0052] The local optimal solution is calculated based on the multi-objective optimization function to output the optimal resource configuration solution in the current optimization period, and corresponding stage management instructions and resource configuration parameters are generated;

[0053] As a preferred scheme of the present application, the life cycle module further comprises:

[0054] An adaptive optimization controller is used to dynamically adjust the calling order and frequency of the Kepler optimization algorithm, the multi-agent reinforcement learning algorithm and the simulated annealing algorithm based on the life cycle code and the subsystem state vector, so as to adapt to the resource configuration optimization target of the park in different life cycle stages;

[0055] The expression of the multi-objective optimization function F(x) is:

[0056] F(x) = λ1·log(1 + E(x)) + λ2·D(x) + λ3·T(x) + λ4·exp(R(x)) + μ·E(x)·T(x);

[0057] In the formula, x is a resource configuration vector; λ1, λ2, λ3, and λ4 are multi-objective weight parameters; E(x) is an energy fluctuation rate; D(x) is a spatial resource congestion degree; T(x) is a task response time; R(x) is a risk level; and μ is a coupling penalty coefficient.

[0058] The method comprises the following steps:

[0059] Real-time collection of running state data, environmental parameter data, and personnel activity data of sensors and intelligent devices in the park, and execution of preprocessing operations such as denoising, compression, caching, uploading, and offline uploading on edge computing nodes;

[0060] Format unification, field standardization, metadata annotation, and data credibility score calculation are performed on the preprocessed data to form a traceable data set in a unified format;

[0061] The data set in the unified format is modeled and analyzed to construct a device running feature matrix, an energy consumption feature matrix, a personnel behavior feature matrix, and an environmental factor feature matrix, perform multi-modal feature encoding and multi-task joint modeling, and output analysis results including device running state prediction, energy consumption trend indicators, safety event labels, and resource configuration parameters;

[0062] Based on the park life cycle stage data and the real-time running state data of the subsystems, the analysis results are comprehensively evaluated, a collaborative optimization mechanism of the fusion Kepler optimization algorithm, the multi-agent reinforcement learning algorithm, and the simulated annealing algorithm is adopted, resource scheduling strategies and multi-objective optimization strategies are executed, and stage management instructions and resource configuration data are output;

[0063] A three-dimensional digital model of the park is constructed based on the building information model, geographic information system data, and generative artificial intelligence technology, virtual-real synchronous updating, visualization of analysis results, and simulation response of stage management instructions are realized based on the three-dimensional data model.

[0064] The application has the beneficial effects that: the multi-protocol access and edge computing capabilities are supported, various types of heterogeneous devices in the park can be efficiently accessed, intelligent preprocessing such as format conversion, denoising filtering, compression and packaging, network interruption and continuous transmission is completed at the edge node, the data transmission efficiency and stability are greatly improved, the central server pressure is reduced, the data integrity and real-time performance are guaranteed, the multi-source data is standardized managed and quality controlled through field standardization, metadata labeling, data quality evaluation and bloodline tracking, a unified and reliable data set is constructed, and the data utilization efficiency and analysis accuracy are significantly improved, the comprehensive analysis and prediction of device state, energy consumption trend, personnel behavior and environmental factors are realized through the construction of multi-type feature matrix and deep multi-task joint modeling mechanism, data support and model guidance are provided for park dynamic management and intelligent decision-making, the global search and local fine-tuning of resource scheduling are combined through the establishment of a collaborative optimization mechanism by combining the Kepler optimization algorithm, multi-agent reinforcement learning and simulated annealing algorithm, the energy consumption, response time, space utilization and safety risk are effectively optimized, and the park operation efficiency and management level are significantly improved, the real-time synchronous update of perception data and analysis results is supported by constructing a three-dimensional model of the park, the dynamic reflection of the virtual space on the actual park operation state and the visualization simulation of management instructions are realized, and the system interactivity and operability are improved. BRIEF DESCRIPTION OF DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0066] Among them:

[0067] Figure 1 The system structure diagram of the present application is shown in the figure.

[0068] Figure 2 The method flowchart in the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme of the embodiments of the present application will be described clearly and completely in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0070] As Figure 1As shown, an embodiment of the present application provides a digital twin and Internet of Things-based intelligent park full life cycle management system, which comprises:

[0071] A perception edge module is configured to collect running state data, environmental parameter data and personnel activity data of sensors and intelligent devices deployed in the park in real time, and pre-process all the collected data at an edge computing node;

[0072] A data governance module is configured to structure and standardize the data pre-processed at the edge computing node, and perform metadata labeling and data blood relationship identification to form a data set in a unified format that can be traced;

[0073] An intelligent analysis module is configured to model and analyze the data set in the unified format, output analysis results and transmit them to a life cycle module and a twin modeling module;

[0074] The life cycle module is configured to comprehensively evaluate the analysis results in combination with park life cycle stage data and real-time running state data of subsystems, execute resource scheduling strategies and multi-objective optimization strategies by using a collaborative optimization mechanism of a Kepler optimization algorithm, a multi-agent reinforcement learning algorithm and a simulated annealing algorithm, and output stage management instructions and resource configuration data;

[0075] The twin modeling module is configured to construct a three-dimensional digital model of the park based on a building information model, geographic information system data and generative artificial intelligence technology, receive data collected by the perception edge module, analysis results of the intelligent analysis module, and stage management instructions and resource configuration data output by the life cycle module, and realize virtual-real synchronous updating, visualization of analysis results and simulation response of stage management instructions based on the three-dimensional data model.

[0076] (1) Perception edge module

[0077] The perception edge module is located at the front end of the data collection and primary processing of the park intelligent management system, and plays a key role in inputting data flow from the physical world to the digital twin model. Through the cooperation of the Internet of Things perception layer and the edge computing node, the module realizes real-time capture and computing offload processing of the park running state, and is the technical basis for realizing stable uploading of high-frequency data and efficient analysis of pre-processing.

[0078] The perception edge module comprises:

[0079] A multi-protocol sensor network (RS485, MODBUS, ZigBee, NB-IoT, etc.);

[0080] An edge computing node device (with local processing, caching and networking upload capabilities);

[0081] Data interface with platform core system (MQTT, HTTPS, Kafka, etc. streaming protocol);

[0082] Anomaly detection and event-driven mechanism component.

[0083] The complete data collection and edge processing process is as follows:

[0084] 1. Data collection layer

[0085] Various IoT sensing terminals are deployed within the park to collect the following information in real time:

[0086] Running state data: such as device start-stop state, current and voltage value, device health code (referring to hot water pump, air conditioner, power distribution cabinet, etc.);

[0087] Environmental parameter data: such as temperature and humidity, illumination, PM2.5, CO2 concentration, etc.

[0088] Personnel activity data: such as access control records, camera recognition tracks, Bluetooth / wireless tag location information.

[0089] Devices are connected to edge computing nodes through edge access protocols.

[0090] 2. Edge computing node processing flow

[0091] The edge computing node performs local preprocessing on the collected data, and the process is as follows:

[0092] Data pre-filtering and format conversion: basic legality verification of original data; unified data format, cleaning redundant fields; preliminary shunt processing according to device type / source.

[0093] De-noising and outlier elimination: based on sliding window and median filtering algorithm to remove abnormal points;

[0094] Data compression and structured packaging: use lightweight protocols (such as LZ4 compression, binary JSON packaging) to compress the size of data packets; classify and label, and attach timestamp, device ID, and measurement point location information.

[0095] Local cache and offline transmission mechanism: establish a ring buffer to support offline transmission; when the network is interrupted, cache data temporarily locally; when the network is restored, automatically determine the synchronization state and resend the unuploaded data packets to ensure data continuity and integrity.

[0096] In one embodiment, the sensing edge module includes:

[0097] A multi-protocol access unit is configured to be compatible with RS485, MODBUS, ZigBee, LoRa, NB-IoT and other communication protocols, and to realize data access to various sensors and intelligent devices deployed in the park;

[0098] An edge computing node is configured to perform preprocessing operations on the collected raw data, including data format conversion, denoising and outlier filtering, compression and packaging, and time synchronization.

[0099] A cache management unit is configured to locally store processed data when the network is abnormal, and to mark the time sequence.

[0100] A network outage resuming unit is configured to judge and retransmit data missing according to data timestamps and check codes after network recovery.

[0101] An edge intelligent engine is embedded with a lightweight AI model, and is configured to perform preliminary anomaly detection and event judgment on the collected raw data.

[0102] A data privacy processing unit is configured to use regional blurring and annotation desensitization algorithms to locally process sensitive features in images and location information.

[0103] An edge upload scheduling unit is configured to sort according to the calculation results of an adaptive upload scheduling function, and sequentially upload high-priority data to the data governance module and the twin modeling module. The adaptive upload scheduling function is:

[0104]

[0105] In the formula, U i is the upload priority score of the ith data packet; P i is the data packet importance score, i.e., the static or dynamic priority level set according to the business scenario (e.g., security alarm > energy monitoring > ordinary environmental parameters); Q i is the current bandwidth availability factor, i.e., the proportion of available bandwidth in the current channel, reflecting the upload capacity; R i is the predicted importance index, which is the score of the potential value of the data to the edge AI, such as the influence on anomaly detection and energy consumption prediction; λ is the timeliness decay coefficient, T i is the time difference between the data packet generation and the current time, is the timeliness decay factor, i.e., the data value decaying with the delay of data packet generation, reflecting "the newer the better"; E i is the unit data energy consumption, i.e., the energy required to upload the data, which depends on the communication method and data size; D iData upload delay, the current estimated data packet transmission delay, in seconds, reflecting the current network delay level; a is the energy consumption penalty weight, controlling the influence of energy consumption in priority calculation (the larger the more not inclined to high energy consumption upload) ; β is the delay penalty weight, controlling the penalty intensity of delay on upload priority; γ, δ are both congestion awareness adjustment factors, γ is used to adjust E i The growth coefficient of weight under congestion condition (reflecting the sensitivity to energy saving), used to adjust D i The growth coefficient of weight under congestion condition (reflecting the demand for rapid release of network pressure) ; C is the network congestion coefficient, that is, the current overall network pressure index of the system, which can be evaluated by the average delay, congestion rate and buffer occupancy rate of the network.

[0106] For example: in a smart park, the edge device needs to judge the upload priority of data packet A (emergency equipment exception) and data packet B (daily temperature and humidity), and the specific parameters are shown in Table 1:

[0107] Table 1 Parameter values of data packet A and data packet B

[0108] Parameter P i ]]> Q i ]] [R i ]] [TECHNICAL FIELD] i ]] E i ]]> D i ]]> C A 0.95 (very high) 0.8 0.9 (high predicted risk of failure) 3 seconds 1.2J 0.6 seconds 0.7 (network congestion is relatively heavy) B 0.4 (average) 0.8 0.2 (predicted insensitivity) 10 seconds 0.5J 0.4 seconds 0.7

[0109] Parameter setting suggestion: a = 1.0, β = 1.5, λ = 0.2, γ = 1.0, δ = 1.0, put into the adaptive upload scheduling function, and U A ≈0.1051, U B ≈0.00463;

[0110] According to the calculation result, the upload priority of data packet A is much higher than that of data packet B, and the scheduling strategy is reasonable and effective.

[0111] (2) Data governance module

[0112] The data governance module plays a "data hub" role between the perception layer and the intelligent analysis layer in the smart park system, and its task is to process a large number of heterogeneous and original Internet of Things data into structured and trusted analysis input data sets. Its core goal is to improve the quality, security and fusibility of data, so as to guarantee the correctness and traceability of the basic data of AI analysis and decision.

[0113] As one of the embodiments of the present application, the data governance module comprises:

[0114] The type identification unit is used for format determination and source classification of the data stream uploaded by the perception edge module, and extracts the device identifier, timestamp and location information of the data;

[0115] A field standardization unit is configured to perform field unification and value range standardization processing on different source data according to preset field mapping rules and unit conversion logic, and the field mapping rules are dynamically updated through a meta-learning mechanism to adapt to different device types and data structures, form a field template library and support automatic evolution.

[0116] The meta-learning mechanism includes extracting structural features and context information, training a field standardization meta-model, performing standard field prediction matching, updating model parameters according to feedback results and synchronously updating the field template library.

[0117] A data fusion unit is configured to perform format conversion on semi-structured or multi-format perception data, and perform fusion processing on multiple data sources through a time window-based alignment algorithm to output structured data in a unified format.

[0118] A metadata labeling unit is configured to label structured data to generate metadata labels including a data source device identifier (device unique identifier such as device ID or MAC address), a collection timestamp (standard UTC time at the time of original data collection, accurate to milliseconds), spatial location information (physical deployment location coordinates of a sensor or device), a data upload node identifier (indicating which edge computing node uploaded the data), a source trust level label (based on device authentication history, vendor reputation, and stability score to form a graded label), an edge processing identifier (indicating whether the data has undergone preliminary reasoning or filtering by an edge-side AI model), a data integrity score (based on field missing rate and format correctness score), a data continuity score (based on historical upload interval and time continuity score based on packet loss), and a processing version number (indicating the data governance rule version ID used in the data processing process) to form a manageable data asset directory. The above metadata labels are common items in the embodiments of the present application, and other necessary fields can be extended according to specific data types in actual application.

[0119] A data quality evaluation unit is configured to score each data record based on multi-dimensional indicators including data integrity, logical consistency, value range fluctuation rationality, and time continuity.

[0120] A data bloodline tracking unit is configured to establish processing path records from data collection, preprocessing, standardization, fusion to output the whole process, and calculate a data trust score.

[0121] An output interface unit is configured to provide structured data, metadata labels, data quality scores, and data trust scores to an intelligent analysis module and a life cycle module, respectively.

[0122] To realize the quantifiable credibility evaluation of the park perception data after structured processing, the application comprehensively considers the performance of the data in the transmission path, processing consistency, time characteristics and traceability, and calculates the overall credibility score of each data record through a nonlinear function conversion and a machine learning weight adjustment mechanism, and the formula is:

[0123] T s =f1(C p )·ω1+f2(V r )·ω2+f3(S t )·ω3+f4(L d )·ω4+f5(I t )·ω5;

[0124] In the formula, T s is the data credibility score; C p is the data governance chain integrity index, reflecting whether the data has undergone a complete perception to upload process, and its value is dynamically generated according to whether the processing path is closed loop and whether there is a broken chain record; V r is the redundancy consistency verification score, evaluating the consistency degree of the same data item in multiple collection or processing paths, and is determined based on the consistency comparison or hash verification result of the redundant sampling result; S t is the time synchronization measure, used to evaluate the delay difference between data collection time and upload time, platform receiving time, and the smaller the error, the higher the synchronization; L d is the processing chain depth, which is the depth and complexity of the data in the edge node, cache unit, data governance module and other processing nodes, and is obtained after combining the node weight; I t is the information traceability index, quantifying whether the data is accompanied by complete processing logs, version identification and source device records, and the record with high traceability has a higher score; f1-f5 are functions for nonlinear conversion of the score factors respectively, used to improve the sensitivity and response range of the model in the abnormal boundary value section, avoid score misjudgment caused by linear weighting, and the nonlinear function can include logarithmic function (such as log(x+1)), exponential decay function (such as e -x ) or S-shaped function (such as sigmoid), and the specific selection is set according to the data distribution characteristics; ω1-ω5 are score weights, which are not statically set, but are dynamically optimized based on historical score results and score accuracy feedback using a machine learning model, supporting gradual optimization of the weight contribution of each factor through incremental training, so as to improve the discrimination accuracy of the model for data credibility.

[0125] The final data credibility score T s can be used as a reference for the system in executing data screening, life cycle decision reasoning and visual display sorting, and also supports traceability analysis and repair suggestion generation for low credibility data.

[0126] (3) Intelligent analysis module

[0127] The intelligent analysis module serves as the intelligent decision-making center of the system, and faces the high-quality data set after governance. It conducts comprehensive analysis through multiple types of AI modeling methods, and outputs structured results according to the task categories. While ensuring the accuracy of the model, a credibility scoring mechanism is introduced to realize the explainability and evaluability of the analysis results. Through the model in the center-edge heterogeneous deployment, a hierarchical response mechanism is formed to improve the decision-making efficiency, reliability and resource adaptability of the system.

[0128] In one embodiment, the intelligent analysis module includes:

[0129] A feature extraction unit is configured to extract features from the unified format data set according to the analysis target, and construct four types of feature matrices, including a device operation feature matrix, an energy consumption feature matrix, a personnel behavior feature matrix, and an environmental factor feature matrix.

[0130] The device operation log matrix records time series data such as device state, start-stop frequency, voltage and current, etc.

[0131] The energy consumption feature matrix records real-time energy consumption data, peak-valley load and historical period statistics of each subsystem in the park, such as daily / monthly energy consumption of the subsystem, electric meter data, peak-valley load, etc.

[0132] The personnel behavior feature matrix includes personnel trajectory point set, residence time, behavior state label, moving speed, activity hot zone distribution, etc.

[0133] The environmental factor feature matrix includes temperature and humidity, air quality, noise, water, electricity and gas indicators, etc.

[0134] A feature encoding unit is configured to perform modal embedded representation learning on the four types of feature matrices, convert heterogeneous features into high-dimensional representation tensors in a unified vector space through a modal mapping function, and unify the time sequence dimension using a sliding time window and a time alignment algorithm.

[0135] A multi-task joint modeling unit includes a shared encoder and multiple task decoders, supports joint training and multi-task dynamic weight regulation mechanism, and automatically adjusts the optimization weight of each task decoder according to the task loss. The shared encoder adopts a deep residual attention mechanism to extract general semantic features from the multi-modal fusion tensor. The task decoder is independently trained for different modeling tasks and has a task-specific attention module.

[0136] An analysis result integration unit is configured to format the output results of each task decoder into an analysis result set, including:

[0137] Device operation state prediction: operation trend, health degree, remaining life estimation, etc.

[0138] Energy consumption trend indicators: multi-period energy consumption change curve, short-term energy efficiency prediction, etc.

[0139] Security event labels: intrusion detection, cluster identification, wandering behavior marking, etc.

[0140] Resource configuration parameters: priority, scheduling time window, coverage range suggestion, etc.

[0141] The life cycle module and the twin modeling module are called.

[0142] The task decoder specifically includes:

[0143] The device state prediction modeler is configured to receive a device operation feature matrix, extract device topology structure features using a graph neural network, jointly model with a time series Transformer, realize state trend prediction through a sliding time window, and output device operation state prediction and fault risk score.

[0144] The energy trend analysis modeler is configured to receive an energy consumption feature matrix, extract main periodic components using a time series Transformer combined with a fast Fourier transform periodic pattern detection mechanism, and output energy use prediction values and abnormal energy consumption warnings.

[0145] The security behavior recognition modeler is configured to receive a personnel behavior feature matrix and an environmental factor feature matrix, use a space-time three-dimensional Transformer structure, and embed a multi-channel cross-modal attention mechanism to perform deep semantic alignment on image frames and trajectory vectors, and output abnormal behavior types and risk level heat maps.

[0146] The resource allocation parameter generation modeler is configured to fuse four types of feature matrices, as well as the output results of the device state prediction modeler, the energy trend analysis modeler, and the security behavior recognition modeler, and based on multi-objective regression and resource scheduling optimization, output resource configuration parameters including spatial utilization rate, energy consumption scheduling parameters, and service load priority.

[0147] (4) Life cycle module

[0148] Park life cycle stage data refers to information reflecting the current operation stage of the park, including but not limited to:

[0149] Planning stage (not yet constructed, management needs mainly simulation and prediction);

[0150] Construction stage (device installation and commissioning stage);

[0151] Operation stage (normal operation stage, focusing on energy efficiency and fault warning);

[0152] Maintenance period (aging and failure repair stage);

[0153] Retirement period (stage of demolition, relocation or functional transformation).

[0154] Source: project construction schedule, facility commissioning records, asset depreciation information, life cycle identification tags, etc.

[0155] Real-time subsystem operating state data refers to the current operating condition parameters of each subsystem (such as energy management, security system, HVAC, office system, etc.);

[0156] Typical data includes: device switch status, power consumption, alarm information, load rate, online rate, fault log, etc.

[0157] Source: real-time upload of perception edge module and processing results of data governance module.

[0158] In one embodiment, a collaborative optimization mechanism is adopted, which integrates Kepler optimization algorithm, multi-agent reinforcement learning algorithm and simulated annealing algorithm, to execute resource scheduling strategy and multi-objective optimization strategy. The method includes:

[0159] Based on the park life cycle stage data, life cycle coding is generated as a scheduling guide signal in the optimization process, and the multi-objective weight parameters matched with the current stage are calculated, which are used to construct the multi-objective optimization function F(x) that adapts to the life cycle target, the expression is:

[0160] F(x) = λ1·log(1+E(x))+λ2·D(x)+λ3·T(x)+λ4·exp(R(x))+μ·E(x)·T(x);

[0161] In the formula, x is the resource configuration vector, representing the current candidate solution, including resource allocation decisions of all subsystems, such as energy consumption scheduling parameters, service priority, space usage ratio, etc. λ1, λ2, λ3, λ4 are multi-objective weight parameters, used to express the importance of each sub-target in the current life cycle stage. They can be set by rules or adjusted adaptively by reinforcement learning; μ is the coupling penalty coefficient, used to control the weight influence of the target coordination term (such as the product of energy consumption and response time) in the total optimization function;

[0162] E(x) is the energy consumption fluctuation rate, which measures the energy consumption balance and fluctuation degree of each subsystem under configuration x, which can be defined as:

[0163]

[0164] In the formula, e i (x) is the energy consumption of the i th subsystem; E is the average energy consumption of the whole system; n is the number of subsystems.

[0165] D(x) is the spatial resource congestion degree, reflecting the balance degree of resource distribution or local congestion phenomenon in the park, which can be defined as:

[0166] D(x) = max z∈Z ρ z (x) - min z∈Z ρ z (x) ;

[0167] wherein, ρ z (x) is the resource utilization rate of the region z; Z is a set of spatial regions;

[0168] T(x) is the task response time, which can be defined as:

[0169] wherein, RT j (x) is the response time of the jth task under the configuration x; m is the total number of tasks;

[0170] R(x) is the risk level, which can be defined as:

[0171] wherein, Risk k (x) is the predicted occurrence probability of the kth risk event; ω k is the corresponding risk weight; p is the number of risk types;

[0172] The Kepler optimization algorithm is used to perform global orbital search on the resource configuration space to generate multiple initial candidate solutions of resource allocation;

[0173] The initial candidate solutions are evaluated, a lightweight historical scoring model is used to calculate the multi-objective optimization function, low fitness solutions are removed, and high-quality initial solutions are retained to form an intermediate candidate solution population;

[0174] The intermediate candidate solution population is input into a multi-agent reinforcement learning system, each subsystem corresponds to an agent, and the life cycle code is embedded into the policy network of each agent as an auxiliary state vector; each agent performs action selection based on the subsystem state vector, updates the policy to optimize the four indicators of energy fluctuation rate, task response time, spatial resource congestion degree and risk level in the multi-objective optimization function, and outputs a set of reinforced candidate solutions optimized by the policy;

[0175] The current optimal solution in the reinforced candidate solution set is introduced into sparse representation, Gaussian perturbation is performed based on the sparse solution vector, and the Metropolis criterion is used to determine whether the new solution replaces the original solution, and a locally optimal solution processed by the simulated annealing algorithm is output;

[0176] Based on the multi-objective optimization function, the final score of the local optimal solution is calculated, and the optimal resource allocation solution in the current optimization period is output, and the corresponding phased management instruction and resource allocation parameter are generated.

[0177] The life cycle module further comprises an adaptive optimization controller for dynamically adjusting the calling order and frequency of the Kepler optimization algorithm, the multi-agent reinforcement learning algorithm and the simulated annealing algorithm based on the life cycle code and the state vector of each subsystem to adapt to the resource allocation optimization target of the park in different life cycle stages.

[0178] (5) Twin modeling module

[0179] The twin modeling module can adaptively adjust the modeling granularity and simulation strategy according to the life cycle stage of the park, for example, focusing on structure layout simulation and capacity prediction in the planning stage, focusing on equipment scheduling visualization and load analysis in the operation stage, and supporting fault evolution simulation and resource redundancy configuration evaluation in the maintenance stage, further realizing the closed-loop control capability of "modeling-analysis-prediction-optimization-response".

[0180] The system can also include:

[0181] 1) Energy management and energy efficiency evaluation module

[0182] Function: Execute energy optimization algorithm to realize dynamic optimization control of park energy management, calculate AIoT system life cycle energy consumption indicators, provide energy efficiency evaluation report and deployment optimization suggestion.

[0183] Input: Device running state, energy consumption data.

[0184] Output: Energy efficiency evaluation report and optimization control instruction.

[0185] 2) Comprehensive security and public safety module

[0186] Function: Optimize security monitoring layout based on AI and simulation, realize intelligent identification and real-time monitoring of security scenes such as intelligent video analysis, perimeter alarm, and smart fire fighting.

[0187] Input: Real-time monitoring video data of the park, security device state data.

[0188] Output: Real-time security alarm and linkage control instruction.

[0189] 3) Intelligent property and equipment maintenance module

[0190] Function: Perform life cycle management and equipment maintenance strategy optimization of assets in the park, automatically generate maintenance priority, and recommend maintenance strategy.

[0191] Input: Real-time running data of equipment, asset management information.

[0192] Output: Maintenance plan, asset management report.

[0193] 4) Enterprise and industry chain service module

[0194] Function: Support dynamic maintenance of enterprise archives, enterprise growth analysis, construction of industry chain correlation graph, assist in investment decision-making and optimization of enterprise services.

[0195] Input: Enterprise operation data, industry chain related data.

[0196] Output: Enterprise growth analysis report, industry chain service recommendation.

[0197] 5) User interaction and intelligent service portal

[0198] Function: Provide a unified digital service portal and mobile application for park users, such as online repair, visitor reservation, equipment reservation, information publishing, etc.

[0199] Input: User demand request, data query and interaction operation.

[0200] Output: Service response and user experience feedback, information publishing and notification push.

[0201] As shown in Figure 2 Another embodiment of the present application provides a full life cycle management method for a smart park based on digital twinning and the Internet of Things, based on the full life cycle management system for a smart park based on digital twinning and the Internet of Things as described above, comprising:

[0202] Real-time collection of running state data, environmental parameter data and personnel activity data of sensors and intelligent devices in the park, and execution of preprocessing operations such as denoising, compression, cache uploading and offline transmission on edge computing nodes;

[0203] Format unification, field standardization, metadata annotation and data credibility score calculation are performed on the preprocessed data to form a unified format data set that can be traced;

[0204] Modeling analysis is performed on the unified format data set to construct device running feature matrix, energy consumption feature matrix, personnel behavior feature matrix and environmental factor feature matrix, multi-modal feature encoding and multi-task joint modeling are performed, and analysis results including device running state prediction, energy consumption trend index, safety event label and resource configuration parameters are output;

[0205] In combination with the park life cycle stage data and the subsystem real-time running state data, the analysis result is comprehensively evaluated, a synergistic optimization mechanism of fusion Kepler optimization algorithm, multi-agent reinforcement learning algorithm and simulated annealing algorithm is adopted, resource scheduling strategy and multi-objective optimization strategy are executed, and stage management instruction and resource configuration data are output.

[0206] Based on the building information model, geographic information system data and generative artificial intelligence technology, a three-dimensional digital model of the park is constructed, and based on the three-dimensional data model, virtual-real synchronous updating, analysis result visualization and simulation response of stage management instruction are realized.

[0207] In summary, through multi-module collaborative design and multi-technology fusion innovation, the present application constructs a complete structure, closed-loop function and real-time response intelligent park full life cycle management system, which has significant practical value and broad popularization prospect in improving the intelligent level of park management, resource allocation efficiency and safety guarantee capability.

[0208] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of various changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A smart park full life cycle management system based on digital twin and Internet of Things, characterized in that, The system comprises: a perception edge module for collecting real-time operation state data, environmental parameter data and personnel activity data of sensors and intelligent devices deployed in the park, and pre-processing all collected data at an edge computing node; a data governance module for structuring and standardizing the data pre-processed by the edge computing node, and performing metadata labeling and data blood relationship identification to form a data set in a unified format that is traceable; an intelligent analysis module for modeling and analyzing the data set in the unified format, outputting analysis results and transmitting them to a lifecycle module and a twin modeling module; the lifecycle module for comprehensively evaluating the analysis results in combination with park lifecycle stage data and real-time operation state data of subsystems, executing resource scheduling strategies and multi-objective optimization strategies using a collaborative optimization mechanism that integrates Kepler optimization algorithm, multi-agent reinforcement learning algorithm and simulated annealing algorithm, and outputting stage management instructions and resource configuration data; the twin modeling module for constructing a three-dimensional digital model of the park based on building information modeling, geographic information system data and generative artificial intelligence technology, receiving data collected by the perception edge module, analysis results of the intelligent analysis module, and stage management instructions and resource configuration data output by the lifecycle module, and realizing virtual-real synchronous updating, visualization of analysis results and simulation response of stage management instructions based on the three-dimensional digital model; the intelligent analysis module comprises: a feature extraction unit for extracting features from the data set in the unified format according to analysis targets, and constructing four types of feature matrices, i.e., device operation feature matrix, energy consumption feature matrix, personnel behavior feature matrix and environmental factor feature matrix; a feature encoding unit for performing modal embedded representation learning on the four types of feature matrices, converting heterogeneous features into high-dimensional representation tensors in a unified vector space through a modal mapping function, and unifying the time dimension using a sliding time window and a time alignment algorithm; a multi-task joint modeling unit including a shared encoder and multiple task decoders, supporting joint training and multi-task dynamic weight regulation mechanism, and automatically adjusting the optimization weights of each task decoder according to task loss; the shared encoder adopts a deep residual attention mechanism to extract general semantic features from the high-dimensional representation tensors; the task decoders are independently trained for different modeling tasks and have task-specific attention modules; an analysis result integration unit for formatting the output results of each task decoder into an analysis result set, which includes device operation state prediction, energy consumption trend indicators, safety event labels and resource configuration parameters, for calling by the lifecycle module and the twin modeling module; the lifecycle module uses a collaborative optimization mechanism that integrates Kepler optimization algorithm, multi-agent reinforcement learning algorithm and simulated annealing algorithm to execute resource scheduling strategies and multi-objective optimization strategies, and the method comprises: Generating a lifecycle encoding based on park lifecycle stage data as a dispatch guidance signal in the optimization process and calculating multi-objective weight parameters matching the current stage for building a multi-objective optimization function adapted to the lifecycle objectives ; performing global orbital search on the resource configuration space using Kepler optimization algorithm to generate multiple initial candidate solutions of resource allocation; The initial candidate solution is subjected to fitness evaluation, a lightweight historical scoring model is used to calculate a multi-objective optimization function, low fitness solutions are eliminated, and high-quality initial solutions are retained to form an intermediate candidate solution population; The intermediate candidate solution population is input into a multi-agent reinforcement learning system, each subsystem corresponds to an agent, and the life cycle code is embedded into the agent policy network as an auxiliary state vector; each agent performs action selection based on the subsystem state vector, updates the policy to optimize four types of indicators in the multi-objective optimization function, including energy consumption fluctuation rate, task response time, spatial resource congestion degree, and risk level, and outputs a set of reinforced candidate solutions optimized by the policy; The current optimal solution in the set of reinforced candidate solutions is introduced into sparse representation, Gaussian perturbation is performed based on the sparse solution vector, and the Metropolis criterion is used to determine whether the new solution replaces the original solution, and a locally optimal solution processed by the simulated annealing algorithm is output; The locally optimal solution is subjected to final score calculation based on the multi-objective optimization function, and the optimal resource allocation solution in the current optimization period is output, and corresponding phased management instructions and resource allocation parameters are generated. 2.The digital-twin and IoT-based smart park full life cycle management system according to claim 1, characterized in that, The perception edge module comprises: A multi-protocol access unit is configured to be compatible with RS485, MODBUS, ZigBee, LoRa, and NB-IoT communication protocols, and to realize data access to various sensors and intelligent devices deployed in the park; An edge computing node is configured to perform preprocessing operations on collected raw data, including data format conversion, denoising and outlier filtering, compression and packaging, and time synchronization; A cache management unit is configured to locally store processed data when the network is abnormal, and to mark the time sequence; A network outage transmission unit is configured to judge and retransmit data missing according to the data timestamp and check code after the network is restored. 3.The digital-twin and IoT-based smart park full life cycle management system of claim 2, wherein, The perception edge module further comprises: An edge intelligent engine is embedded with a lightweight AI model and is configured to perform preliminary anomaly detection and event judgment on collected raw data; A data privacy processing unit is configured to use regional blur and label desensitization algorithms to locally process sensitive features in images and location information; An edge upload scheduling unit is configured to sort according to the calculation results of an adaptive upload scheduling function, and sequentially upload high-priority data to the data governance module and the twin modeling module; The adaptive upload scheduling function is: ； wherein, is the upload priority score of the data packet of the first class; is the data packet importance score; is the data packet importance score; is the current bandwidth availability factor; is the predicted importance index; is the time-to-live decay coefficient, is the time difference between the data packet generation and the current time; is the unit data energy consumption; is the data upload delay; is the energy consumption penalty weight, is the delay penalty weight, are the congestion-aware adjustment factors, is the network congestion coefficient. 4.The digital-twin and IoT-based smart park full life cycle management system of claim 1, wherein, The data governance module comprises: A type identification unit is configured to determine the format and source of the data stream uploaded by the perception edge module, and extract the device identifier, timestamp, and location information of the data; A field standardization unit is configured to perform field unification and value range standardization processing on data from different sources according to a preset field mapping rule and unit conversion logic, and the field mapping rule is dynamically updated through a meta-learning mechanism to adapt to different device types and data structures, forming a field template library and supporting automatic evolution; The meta-learning mechanism comprises: extracting structural features and context information, training a field standardization meta-model, performing standard field prediction matching, updating model parameters according to feedback results, and synchronously updating the field template library. A data fusion unit is configured to perform format conversion on semi-structured or multi-format perception data, and to perform fusion processing on multiple data sources through a time window-based alignment algorithm, and to output structured data in a unified format. 5.The digital-twin and IoT-based smart park full life cycle management system of claim 4, wherein, The data governance module further includes: A metadata labeling unit is configured to label structured data to generate metadata labels including data source device identification, collection timestamp, spatial location information, data upload node identification, source credibility level label, edge processing identification, data integrity score, data continuity score, and processing version number, forming a manageable data asset directory; A data quality assessment unit is configured to score the data quality of each data record based on multi-dimensional indicators including data integrity, logical consistency, value range fluctuation rationality, and time continuity; A data bloodline tracking unit is configured to record the processing path of data from collection, preprocessing, standardization, fusion to output, and to calculate a data credibility score, with the formula being: ； wherein, is a data trust score; is a data governance chain integrity indicator; is a redundancy consistency check score; is a time synchronization measure; is a processing chain depth; is an information traceability indicator; is a function that respectively performs a non-linear transformation on the score factors; is a dynamically generated score weight for a machine learning model; An output interface unit is configured to provide structured data, metadata labels, data quality scores, and data credibility scores to an intelligent analysis module and a life cycle module, respectively. 6.The digital-twin and IoT-based smart park full life cycle management system of claim 1, wherein, The task decoder specifically includes: A device state prediction modeler is configured to receive a device operation feature matrix, extract device topology structure features using a graph neural network, jointly model with a time series Transformer, achieve state trend prediction through a sliding time window, and output device operation state prediction and fault risk score; An energy trend analysis modeler is configured to receive an energy consumption feature matrix, extract main periodic components using a periodic pattern detection mechanism combining a time series Transformer and a fast Fourier transform, and output energy use prediction values and abnormal energy consumption alerts; A safety behavior recognition modeler is configured to receive personnel behavior feature matrix and environmental factor feature matrix, use a space-time three-dimensional Transformer structure, and embed a multi-channel cross-modal attention mechanism, perform deep semantic alignment on image frames and trajectory vectors, and output abnormal behavior types and risk level heat maps; A resource allocation parameter generation modeler is configured to fuse the four types of feature matrices, and the output results of the device state prediction modeler, the energy trend analysis modeler, and the safety behavior recognition modeler, and based on multi-objective regression and resource scheduling optimization, output resource allocation parameters including space utilization rate, energy consumption scheduling parameters, and service load priority. 7.The digital-twin and IoT-based smart park full life cycle management system of claim 1, wherein, The life cycle module further includes: An adaptive optimization controller is configured to dynamically adjust the calling order and frequency of Kepler optimization algorithm, multi-agent reinforcement learning algorithm, and simulated annealing algorithm based on life cycle coding and state vectors of each subsystem, to adapt to resource configuration optimization objectives in different life cycle stages of the park. The multi-objective optimization function The expression is: ； In the formula, is a resource configuration vector; is a multi-objective weight parameter; is an energy consumption fluctuation rate; is a spatial resource congestion degree; is a task response time; is a risk level; is a coupling penalty coefficient.

8. The method of claim 1-7, wherein the system is characterized in that, The method includes: Real-time collection of sensor and intelligent device operation state data, environmental parameter data, and personnel activity data in the park, and execution of preprocessing operations such as denoising, compression, caching, uploading, and network outage continuation on edge computing nodes; The preprocessed data is subjected to format unification, field standardization, metadata labeling and data credibility score calculation to form a traceable data set in a unified format; The data set in the unified format is subjected to modeling analysis to construct a device operation feature matrix, an energy consumption feature matrix, a personnel behavior feature matrix and an environmental factor feature matrix, multi-modal feature coding and multi-task joint modeling are performed, and analysis results including device operation state prediction, energy consumption trend index, safety event label and resource configuration parameter are outputted; The analysis results are comprehensively evaluated in combination with park life cycle stage data and subsystem real-time operation state data, a collaborative optimization mechanism fusing Kepler optimization algorithm, multi-agent reinforcement learning algorithm and simulated annealing algorithm is adopted, resource scheduling strategy and multi-objective optimization strategy are executed, and stage management instructions and resource configuration data are outputted; A three-dimensional digital model of the park is constructed based on building information model, geographic information system data and generative artificial intelligence technology, virtual-real synchronous updating, analysis result visualization and simulation response of stage management instructions are realized based on the three-dimensional digital model.

Images