Intelligent campus comprehensive service management method and system

Abstract

The application provides a kind of intelligent campus comprehensive service management method and system, it is related to digital twin technical field, it is applied to the campus physical space comprising multiple sites and laboratory nodes, it aims at solving the challenge of traditional campus management in complex space-time resource allocation, environmental interference prediction and facility operation and maintenance.This method constructs a digital twin network model, and obtains the dynamic topology mapping data of resource occupation and environmental evolution of each to-be-allocated site node in real time.The data includes the spatial interference evolution law characterized by spatial repulsion gradient, and the physical wear evolution law characterized by virtual node wear value.The system performs concurrent execution rehearsal on time-overlapping high-risk task combinations, performs spatial interference conflict checking, generates a first interference avoidance scheduling strategy or a second buffer scheduling strategy, and reduces the temporary capacity of the node.The application improves the campus resource utilization efficiency, task scheduling accuracy, campus facility operation and maintenance level and service life.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to an intelligent campus integrated service management method and system. Background Technology

[0002] With the rapid advancement of information technology in higher education, the management efficiency of comprehensive campus services directly impacts the smooth operation of teaching and research. Modern campuses typically contain numerous physical spaces and laboratories, which bear diverse and varying workloads in scientific research experiments and comprehensive tasks. Currently, most mainstream campus space and service management systems employ a static rule-based reservation and scheduling mechanism. Users submit reservation requests through the front-end system, and the back-end system simply allocates spaces based on their availability (first-come, first-served) or fixed priorities.

[0003] However, as the tasks undertaken by campus physical spaces become increasingly complex, existing technologies are gradually revealing numerous limitations. Traditional management methods treat campus spaces merely as independent, static time-block resources, ignoring the complex spatial interference effects generated during the execution of different tasks and the implicit physical wear and tear on equipment. For example, sudden changes in local energy consumption caused by high-intensity operations, acoustic field disturbances or abnormal carbon emissions from adjacent laboratories, can all generate cross-node interference through the topological relationships of physical spaces. Existing technologies lack the ability to deeply integrate multimodal IoT sensing data and extract environmental fingerprints, making it impossible to dynamically track the evolution of these physical states. Traditional systems cannot perform risk-resistant time-domain simulation and global performance evaluation, leading to severe resource fragmentation in spatiotemporal allocation schemes, limited global scheduling success rates, and low efficiency and error-prone manual decision-making in the face of massive amounts of sensing data and complex task scheduling requirements. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent campus integrated service management method and system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An intelligent campus integrated service management method is applied to a campus physical space that includes multiple venues and laboratory nodes. Specific steps include: S1: Obtain dynamic topology mapping data of resource occupancy and environmental evolution of the digital twin network model of the site nodes to be assigned within a preset time window. The dynamic topology mapping data is configured to characterize the spatial interference evolution law represented by the spatial repulsion gradient and the evolution law of physical loss represented by the virtual node wear value over time in the process of each site node to be assigned carrying different tasks. S2: Based on the dynamic topology mapping data obtained in step S1, an optimal scheduling strategy is automatically generated and matched for a group of tasks to be assigned, specifically including: S21: Perform concurrent execution rehearsals for high-risk task combinations with temporal overlap to perform spatial interference conflict verification and generate the first interference avoidance strategy, specifically including: S2101: Extract the spatial conduction attenuation factor from the j-th source node to the target node f, and calculate the actual influence value of the j-th source node on the target node f by multiplying the spatial conduction attenuation factor with the spatial repulsion gradient G(t) of the j-th source node. S2102: Extract the virtual node wear value of the f-th target node and compare it with the preset health status benchmark threshold. If the virtual node wear value is higher than the health status benchmark threshold, it is determined that the target node is in a high fatigue state after the task, and the second buffer scheduling strategy is triggered.

[0006] An intelligent campus integrated service management system includes: The data sensing submodule is configured to acquire multimodal IoT sensing data online through multimodal IoT sensing devices deployed at each site node to be assigned. The sensing time window is set within a preset time window before the task executor enters the site node to be assigned. The multimodal sensing data includes environmental time-series sensing data and spatial energy consumption monitoring data. The feature extraction submodule is configured to perform window-level segmentation and feature extraction on the environmental and energy consumption sequences in the online multimodal sensing data through a long short-term memory autoencoder, and output an environmental fingerprint fusion feature vector representing the current physical state of each site node to be assigned. The evolution parameter inference submodule is configured to use a graph neural network, with the environmental fingerprint fusion feature vector as the initial feature of the node, and to use the evolution mechanism of how different experimental tasks interfere in space extracted from the campus scientific research map to construct the constraints of information propagation, to perform iterative inference on the spatial logical adjacency graph, and output an interactive evolution parameter set that can characterize the virtual loss evolution rate α and interference evolution rate β of each site node to be assigned during task execution. The dynamic topology generation submodule is configured to use the interactive evolution parameter set as the dynamic boundary condition of the digital twin network model to generate dynamic topology mapping data. The concurrent execution pre-drill module is configured to perform concurrent execution pre-drills on high-risk task combinations with temporal overlap based on state topology mapping data, in order to perform spatial interference conflict verification and generate a first interference avoidance strategy; and extract the virtual node wear value of the f-th target node and compare it with a preset health status benchmark threshold. If the virtual node wear value is higher than the health status benchmark threshold, it is determined that the target node is in a high fatigue state after the task, and the second buffer scheduling strategy is triggered.

[0007] Compared with the prior art, the beneficial effects of the present invention are: This invention constructs a digital twin network model to acquire real-time dynamic topology mapping data on resource occupancy and environmental evolution of site nodes, and automatically generates and matches scheduling strategies based on this data. By performing concurrent execution rehearsals on high-risk task combinations with overlapping timeframes and conducting spatial interference conflict verification, the system can generate interference avoidance scheduling strategies, thereby effectively avoiding resource conflicts and efficiency losses. Simultaneously, by comparing virtual node wear values ​​with health status benchmark thresholds, the system can predict node fatigue states and trigger buffer scheduling, optimizing the allocation and use of campus site and laboratory resources while ensuring equipment and site health.

[0008] Traditional methods have limitations in assessing task interference in complex physical spaces. This invention acquires environmental fingerprint fusion feature vectors through multimodal IoT sensing devices and utilizes graph neural networks to infer the virtual loss evolution rate α and the interference evolution rate, thus meticulously characterizing the evolution mechanism of spatial interference. By extracting the spatial conduction attenuation factor (considering distance, material medium, and structural path attenuation) and multiplying it with the spatial repulsion gradient, the system calculates the actual impact of the source node on the target node, quantifying the impact of spatial interference. This quantitative assessment mechanism enables the system to more accurately identify potential conflicts and generate predictive first-order interference avoidance strategies.

[0009] This invention introduces virtual node wear values ​​to characterize the evolution of physical wear and tear, and combines this with the overall aging coefficient and load stress factor to predict the fatigue state of nodes. By comparing the virtual node wear values ​​with a health status benchmark threshold, the system can promptly identify nodes in a high-fatigue state and trigger a second buffer scheduling strategy to prevent them from continuing to carry out tasks in an excessively fatigued state. This operation and maintenance management method based on physical wear and tear prediction helps extend the lifespan of equipment and facilities, reduce unplanned downtime and maintenance needs, and promote the stable operation of campus facilities. Attached Figure Description

[0010] Figure 1 This is a schematic diagram illustrating the technical route of the campus physical space and its digital twin network model applied in this invention. Figure 2This is a schematic diagram illustrating the execution steps of the method of the present invention; Figure 3 This is a schematic diagram of the system module block diagram and execution flow of the present invention. Detailed Implementation

[0011] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0012] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below. Example 1:

[0013] Please see Figures 1 to 3 The present invention provides a technical solution: An intelligent campus integrated service management method is applied to a campus physical space that includes multiple venues and laboratory nodes. Specific steps include: S1: Obtain dynamic topology mapping data of resource occupancy and environmental evolution of the digital twin network model of the site nodes to be assigned within a preset time window. The dynamic topology mapping data is configured to characterize the spatial interference evolution law represented by the spatial repulsion gradient and the evolution law of physical loss represented by the virtual node wear value over time in the process of each site node to be assigned carrying different tasks. This invention protects the physical space of a campus, comprising multiple sites and laboratory nodes, and its corresponding digital twin network mapping system. In modern smart campus management, different types of site nodes to be allocated (such as high-throughput computing centers, chemical synthesis laboratories, and large lecture halls) generate complex spatiotemporal resource occupancy and physical environment coupling when undertaking scientific research or teaching tasks. Unlike traditional campus scheduling based on static capacity allocation, the core advantage of this invention lies in its "dynamic twin mapping," that is, the system treats the physical sites within the campus as a multi-dimensional evolving nonlinear dynamic graph network. Different tasks in the space not only consume basic water and electricity resources but also generate heat island effects, acoustic interference, and sewage loads; these factors are abstracted as "coherent interference" and "virtual losses." Through the deep integration of graph neural networks and multimodal sensing technology, this invention can, like high-precision weather forecasting, proactively calculate the evolution of physical losses of each node under specific tasks, thereby providing absolutely reliable underlying data support for the dynamic allocation of campus sites and cross-node collaborative scheduling. This multi-dimensional management requirement also presents challenges to site allocation: when multiple high-load scientific research tasks occur concurrently in adjacent physical spaces, the environmental disturbances they generate, such as sound, light, heat, and electricity, are extremely complex and highly overlap with the temporal fluctuations generated by regular teaching activities. The threshold alarm principle of traditional management systems is difficult to achieve early evolution prediction and precise scheduling.

[0014] The application scenario of this embodiment is deployed in the "Interdisciplinary Research Center Building" of a comprehensive university. This physical space consists of multiple site nodes with complex physical interaction attributes to be assigned. For ease of explanation, three core associated nodes are selected: Site Node V1 (Basement Level): Heavy Machinery and Materials Forging Laboratory. It houses a large high-frequency vibration test bench and a high-power industrial heating furnace. This node has extremely high peak energy consumption and radiates intense mechanical shock waves and thermodynamic diffusion in all directions when operating under high loads.

[0015] Site node V2 (directly above the ground floor): High-precision quantum optics and laser interferometry laboratory. It is equipped with nanoscale measurement devices that are extremely sensitive to environmental vibrations and temperature drift.

[0016] Site node V3 (second floor, side): Biological cell constant temperature culture room. It has stringent requirements regarding the concentration of carbon dioxide in the air and the constant temperature environment.

[0017] Within a preset time window before the task personnel enter the assigned site node (in this embodiment, this is set to 30 minutes before entry until the entire task execution process), the system continuously acquires multimodal IoT sensing data online. The hardware deployment list includes: an environmental time-series sensor array: a triaxial piezoelectric high-frequency vibration sensor (sampling rate 1000Hz), a high-precision sound level meter (accuracy 0.1dB), and an NDIR non-dispersive infrared carbon dioxide sensor (accuracy 1ppm). A space energy consumption monitoring data terminal: three-phase smart meters deployed in the distribution boxes of each node (providing real-time power and voltage fluctuation monitoring, sampling rate 1Hz).

[0018] To ensure the accuracy of subsequent site allocation and coordinated scheduling, Implementation Example 1 focuses on analyzing how to construct a dynamic topology model reflecting the actual physical operating state of various sites on campus within an extremely short sensing time window, and output accurate evolution parameters. Specifically, this includes the following sub-steps: S11. The campus sites awaiting allocation are not isolated systems, but rather interconnected topologies exhibiting significant physical space thermodynamic transfer and interference effects. The system first constructs an interconnected topology model for the entire campus physical space. The system acquires multimodal IoT sensing data online through multimodal IoT sensing devices deployed at each site awaiting allocation node. The sensing time window is set within a preset time window before the task executor enters the site awaiting allocation node. The multimodal sensing data includes environmental time-series sensing data and spatial energy consumption monitoring data. S12. Extract the feature parameters of each site node to be assigned within the perception time window. Using a long short-term memory autoencoder, perform window-level segmentation and feature extraction on the environmental and energy consumption sequences in the online multimodal perception data to construct an environmental fingerprint fusion feature vector for each site node within its perception time window. The specific feature extraction and calculation formulas are as follows: Root mean square of load fluctuation P rms : Obtain the instantaneous power consumption value P of each sampling point within the sensing time window. i The square of each instantaneous power consumption value is calculated, and the arithmetic mean of all the calculated square values ​​is obtained. The square root of the mean square value is then calculated by taking the square root of the mean square value. Where N is the total number of sampling points; the root mean square (RMS) of load fluctuation represents the energy throughput intensity of the node to be assigned within the sensing window. Unlike average power, RMS is more sensitive to peak values ​​and can effectively reflect the high energy consumption impact caused by the instantaneous startup of heavy equipment (such as pulsed lasers and centrifuges). It is the core basis for judging whether a node is a "high-energy-consumption hotspot"; the specific unit is kilowatt (kW). Specific examples: Ordinary classrooms / offices: 2.0~10.0kW; Biological constant temperature laboratory: 20.0~50.0kW (air conditioning always on); Heavy machinery / material forging laboratory: 100.0~500.0kW (high energy consumption peaks exist); Acoustic run complexity A jump : Based on the preset environmental baseline decibel value Db base To determine the threshold, the time-series decibel sampling points Db within the sensing time window are traversed. i When two adjacent sampling points are both greater than or equal to the judgment threshold, it is recorded as a state transition; the total number of state transitions within this sensing time window is counted, and this total number is A. jump Acoustic run complexity A jump Characterizing the "disorder" or "impulse characteristics" of a site's acoustic environment. A device (such as a fan) that continuously emits a constant high noise level. jump It may be very low; while for impact equipment that frequently starts and stops (such as stamping presses, drop test benches) A jump It will be high. This is used to distinguish between "steady-state noise" and "transient disturbance"; the unit of measurement is superposition. Specific examples: Quiet environment (quantum lab): 0-5 times (extremely low disturbance). General teaching area: 10-30 times (personnel movement). Heavy machinery / material forging lab: 100-200 times (high-frequency impact). Judgment threshold: environmental baseline decibel value (db). base Set as the baseline for ambient background noise, with a preferred value of 50dB or 60dB.

[0019] Carbon footprint deviation C dev : Obtain the average actual carbon dioxide concentration C within the first preset time period. actual And obtain the baseline carbon dioxide concentration value C under no-task load conditions. base Calculate the absolute difference and divide by the result to obtain the carbon footprint deviation using the following formula. Carbon footprint deviation C dev Characterizing the degree of deterioration of air quality within a site relative to its vacant state, it not only reflects personnel density but also captures specific task characteristics such as chemical reaction emissions and biological respiration. The unit of measurement is dimensionless, representing a relative ratio. Specific example: No task / Good ventilation: 0.0~0.1. High personnel density / Routine experiments: 0.2~0.5. Specific gas emission tasks (such as bio-fermentation): 0.8~2.0 (i.e., concentration exceeding the standard by more than double); Energy consumption time-series autocorrelation product E autoObtain a time window sequence P(t) containing multiple consecutive instantaneous power consumption values, and generate a comparison sequence P(t−1) lagged by one time step; multiply each power consumption value by its corresponding power consumption value in the comparison sequence and sum them up, then calculate the time series autocorrelation product E of energy consumption using the following formula. auto : The time-series autocorrelation product of energy consumption characterizes the "memory" or "inertia" of equipment operating modes. A high autocorrelation product means that the current energy consumption state is highly correlated with the previous time step, indicating that the equipment is in a continuous and stable operating state (such as a constant temperature chamber); a low value means strong randomness; the unit of measurement is kilowatt square (kW). 2 ); The system combines the data from the above four dimensions in a predetermined order to generate an environmental fingerprint fusion feature vector F=[P rms A jump C dev E auto As shown in Table 1.

[0020] Table 1: Multimodal Environmental Fingerprint Feature Extraction Table for Site Nodes to be Assigned in Campus Physical Space

[0021] Table 1 demonstrates the system's ability to represent complex environments in a dimensionality reduction manner. V1's A jump Extremely high, reflecting mechanical impact characteristics; while V3's C dev The data is high, accurately capturing specific gas emission deviations in biological culture chambers. These nonlinear characteristics provide high-dimensional basis for subsequent graph neural network classification; S10001. For each preset node load state type c (c∈{1,2,...,M}), obtain the standard feature centroid vector C that uniquely corresponds to the current load state type from the pre-trained model parameter library. c ; S10002, Calculate the environmental fingerprint fusion feature vector F and the centroid vector C of each standard feature respectively. c The multidimensional Euclidean distance between them yields the distance scalar D corresponding to each load condition type. c ; S10003. For all distance scalars, to eliminate the influence of dimensions, a preset smoothing constant ϵ is added to each distance scalar, and the reciprocal is taken to obtain the preliminary similarity score S for the corresponding load state type. c : ; S10004. Traverse all preset node load state types and repeat steps S10001 to S10003 to generate a score vector containing the preliminary similarity scores of each load state type. S1232, For all preliminary similarity scores S in the score vector c The total denominator value is obtained by performing a summation operation. Each preliminary similarity score is divided by the total denominator to achieve normalization. The quotient is the final classification confidence vector P for the corresponding load state type. c :

[0022] S13, the step of inputting the environmental fingerprint fusion feature vector F into the long short-term memory autoencoder and outputting the classification confidence vector representing each state type specifically includes: S131. Considering that each site node to be assigned contains multiple different types of equipment, the system assigns weights w to each device based on its rated power or heat / vibration generation capacity. id (And the sum of the weights of all devices is 1). The system iterates through N nodes in the area to be assigned. d The overall aging coefficient K of the core equipment is calculated by weighted summation. age :

[0023] Among them, T serve,id With T life,id These represent the service life and design life of the id-th device, respectively; 0.1 is the preset linear aging damping constant; representing the "health degradation state" of the node hardware. It is an amplifier for virtual loss calculation. The larger the coefficient, the closer the device is to its end-of-life. Example: Brand new equipment: close to 1.0. Halfway through service: around 1.05. Overdue service: >1.1 (at this point, the loss rate accelerates significantly); S132. Calculate the spatial permeability coefficient by extracting the door and window ventilation openings and wall material properties of the nodes to be assigned to the site. :

[0024] Among them, V ratio M is the proportion of the area of ​​openings such as door and window vents to the total interior surface area; trans The wall medium transmittance; 0.7 and 0.3 are preset influence weights; spatial permeability coefficient. This value characterizes the "leakage capability" of a physical space to interference signals such as light, heat, and sound. A higher value indicates a more "leaky" room, meaning the interference generated is more likely to affect neighboring nodes. Examples include: Electromagnetic shielding room / soundproof room: 0.05~0.2 (high isolation); Ordinary laboratory / classroom: 0.4~0.6; Open-ventilated laboratory: 0.7~0.9 (high transparency). S133, Based on the final classification confidence vector P of each node c The system activates a pre-constructed spatial logical adjacency graph (nodes are V1, V2, and V3, and edges represent physical distance and thermodynamic transfer relationships). The system extracts the evolutionary mechanism constraints from the campus scientific research map: "The high-frequency mechanical vibration of V1 will generate strong acoustic and vibration interference along the load-bearing wall of the building to V2 directly above, and the heat island effect generated by the high-power heating furnace will transfer heat to V3 to the side and above, affecting its constant temperature environment." Under this constraint, the overall aging coefficient K calculated by the system in conjunction with the aforementioned steps is... age With spatial permeability coefficient O node The following computational logic outputs the interactive evolution parameter set for each site node to be assigned, specifically including the virtual loss evolution rate α and the interference evolution rate β: The virtual loss evolution rate α is obtained by pre-setting a set of physical wear modulus vectors M corresponding to the node load state types. wear =[μ1,μ2,...,μ M ], where μ c Characterize the baseline physical loss caused to the site by the c-th load condition type per unit time; combine the final classification confidence vector Pc output from step S1232 with the physical wear modulus vector M wear Perform a weighted dot product operation and compare the result with the overall aging coefficient K calculated in step S131. age By performing nonlinear coupling, the virtual loss evolution rate α is calculated using the following formula:

[0025] Where η is a preset material fatigue correction factor (taken as 1.05 in this embodiment), P c [c] represents vector P. c The confidence value of the c-th component. This formula characterizes the age of the equipment (K). age Larger) and bearing high-wear tasks (μ c The larger the proportion of (the higher the value), the faster the virtual loss evolution rate α; the virtual loss evolution rate α represents the basic rate at which the physical health of this node decreases under the current comprehensive operating conditions per unit time. It is a comprehensive reflection of equipment aging and task type intensity; Dimension / Unit: Loss Unit / h; The interference evolution rate β is obtained by pre-setting a set of feature interference transformation weight vectors W corresponding to the dimensions of the environmental fingerprint fusion feature vector FF. int =[ω P ,ω A ,ω C ,ω EThe values ​​represent the interference conversion capabilities of the root mean square of load fluctuation, acoustic run complexity, carbon footprint deviation, and energy consumption time-series autocorrelation product, respectively. The weighted logarithmic norm of the environmental fingerprint fusion feature vector F is calculated and compared with the spatial permeability coefficient O obtained in step S132. node Modulation is performed, and the calculation formula is as follows:

[0026] Where F[k] is the k-th eigenvalue of the eigenvector F, ω k Here, e represents the corresponding weighting coefficient, and e is the base of the natural logarithm (used to ensure the truth value is greater than 1). This formula characterizes the higher the spatial permeability of the site (O... node The larger the node's internal environment disturbance characteristics and the more severe the disturbance, the higher its interference evolution rate β radiated to the external space. The interference evolution rate β characterizes the gain coefficient of the node as an interference source, radiating interference energy to the external space. It compresses the complex internal multimodal environmental fingerprint (noise, heat, etc.) into a scalar, with a dimensionless unit (gain factor). Example: Silent node: 0.5~1.0. Strong interference source: 5.0~15.0; S14. Based on the acquired interactive evolution parameter set of each site node to be assigned, generate dynamic topology mapping data for each site node to be assigned. The core steps are as follows: S141. Obtain the time series T and the task load intensity sequence L describing the process of the task carried by the site node to be allocated. The sequence contains a series of discrete time points and the measured or preset load intensity value at each time point. S142. Set the initial state of all site nodes to be assigned at time t=0, where the initial spatial repulsion gradient is 0, the initial virtual node wear value is 0, and the initial load intensity is extracted from the sequence and denoted as L. initial ; S143. To refine the tracking of evolution, the precise current load intensity L0 of any target time point t is calculated using linear interpolation: In the time series, find the two closest consecutive time points in time that include the target time point t and their corresponding load intensity values, and calculate L0 by interpolation. S144. Obtain the virtual loss evolution rate α obtained in S133, and calculate the current load intensity L0 relative to the initial load intensity L. initial The normalized increment is calculated, and the nonlinear growth characteristic term is extracted by performing a natural logarithmic operation on the load increment ratio. Finally, the three factors are multiplied together, and the nonlinear fatigue characteristics of the normalized increment are extracted using an exponential function to generate the load stress factor. The virtual node wear value W(t) at any time point t is predicted based on the current load state using the following formula:

[0027] S145. Obtain the interferometric evolution rate β obtained in S133, and calculate the current load intensity L0 and the initial load intensity L. initial The square of the difference is used to characterize the load fluctuation energy; the load fluctuation energy is multiplied by the interference evolution rate β and divided by the preset diffusion damping coefficient γ, which reflects the spatial physical barrier effect, so that the spatial repulsion gradient G(t) at any time point t can be calculated by the following formula:

[0028] The preset diffusion damping coefficient γ is a constant, its value being related to the wall thickness and material properties of the physical space, used to quantify the space's attenuation capability for interference signal propagation. For simplicity and to provide a concrete example, the diffusion damping coefficient γ is calculated using the following empirical formula: γ = γbase + Kwall × Dwall; where: γbase is the base damping coefficient, set to 0.5 in this embodiment; Kwall is the wall thickness influence factor, representing the damping increment provided per centimeter of wall thickness, set to 0.25 in this embodiment; and Dwall is the effective wall thickness in centimeters (cm). Specifically, when the wall thickness is 5cm, the diffusion damping coefficient is: γ = 0.5 + 0.25 × 5 = 1.75. When the wall thickness is 10cm, the diffusion damping coefficient is: γ = 0.5 + 0.25 × 10 = 3.00. When the wall thickness is 15cm, the diffusion damping coefficient is: γ = 0.5 + 0.25 × 15 = 4.25. The preset diffusion damping coefficient γ is a dimensionless constant that quantifies the attenuation or blocking ability of physical space to propagate interference signals. In this embodiment, the value of γ directly depends on the wall thickness of the physical space. The thicker the wall, the better the damping performance, and the larger the γ value, the stronger the blocking effect; conversely, the smaller the γ value, the weaker the blocking effect.

[0029] The final interactive evolution parameters for each node are output, as shown in Table 2.

[0030] Table 2: Example Table of Interactive Evolution Parameters and Dynamic Topology Mapping Data for Each Site Node to be Assigned in the Campus Physical Space

[0031] For V1 (Heavy Electromechanical Laboratory): Due to its extremely high energy consumption autocorrelation product and acoustic jump (refer to Table 1), and its high spatial transparency (O node=0.85), and the interference evolution rate β calculated by the formula is as high as 14.20. This leads to an extremely high value (284.00) on the spatial repulsion gradient G(t). In the associated topology model, this means that a huge "interference repulsion field" is formed around V1, and the system will force a separation between the execution time window or physical distance of vibration-sensitive tasks (such as the quantum experiment of V2) and V1 during subsequent scheduling. For V2 (the quantum optics laboratory): as a node that requires absolute silence, its equipment is often highly enclosed and isolated (O node =0.12), therefore its outward radiation interference evolution rate β is only 0.75 (i.e. it does not cause external interference). At the same time, its extremely low virtual loss evolution rate α results in extremely low W(t), indicating that its state is biased towards stability.

[0032] It should be noted that the digital twin network model in this invention is the core of realizing intelligent campus integrated service management. Its construction process aims to map the physical spaces and laboratory nodes of the campus to the digital world with high fidelity, reflecting in real time their dynamic resource occupancy, environmental evolution, and physical wear and tear and interference patterns. The construction steps mainly include the following stages: In the first phase, corresponding to S11, multimodal IoT sensing devices are deployed at various unassigned site nodes on campus (such as laboratories, classrooms, and conference rooms). These devices include, but are not limited to, environmental sensors (temperature, humidity, light intensity, CO2 concentration, VOCs, etc.), acoustic sensors, energy consumption monitoring modules (electricity, water, gas, etc.), and possible personnel positioning or asset tracking sensors. These devices acquire multimodal IoT sensing data online in real time. The data collection time window is typically set within a preset time window before the task personnel enter the site to capture the baseline characteristics of the site in a non-task or preparation state. The collected raw data will undergo preliminary cleaning, noise reduction, and time synchronization processing to lay the foundation for subsequent analysis.

[0033] In the second stage, corresponding to S12, to extract meaningful features from the massive amount of raw sensing data, this invention employs a Long Short-Term Memory Autoencoder (LSTM-AE) to process the preprocessed multimodal sensing data. Specifically, for the environmental time-series sensing data and spatial energy consumption monitoring data sequences, LSTM-AE performs window-level segmentation and performs feature extraction within each time window. These features include, but are not limited to: root mean square of load fluctuation (characterizing energy consumption stability), acoustic run complexity (characterizing noise environment dynamics), carbon footprint deviation (characterizing the indirect correlation between air quality and population density), and energy consumption time-series autocorrelation product (characterizing the periodicity of energy consumption patterns). These extracted features are combined in a predetermined order to form a high-dimensional "environmental fingerprint fusion feature vector" (F), which can comprehensively and concisely represent the current physical state of each site node to be assigned.

[0034] In the third stage, corresponding to S10001-S1232, after generating the environmental fingerprint fusion feature vector F, the system further classifies its load status. A pre-trained model parameter library is pre-established, which stores the standard feature centroid vector C for different node load status types (such as "idle", "light load", "medium load", "heavy load", etc.). c The system computes the fingerprint fusion feature vector F and the centroid vector C of each standard feature. c The multidimensional Euclidean distance between them yields the distance scalar D. c To eliminate the influence of dimensions, D... c The initial similarity score S is obtained by performing smoothing and reciprocal processing. c Then, all preliminary similarity scores are normalized to generate a final classification confidence vector P corresponding to various load state types. c This vector reflects the probability distribution of the current site condition belonging to each load type.

[0035] In the fourth stage, a graph neural network (GNN) is used, with the environmental fingerprint fusion feature vector F as the initial feature of each node. This is combined with the evolutionary mechanism of how different experimental tasks interfere in space, extracted from the campus research map, to construct constraints for information propagation. The GNN performs iterative reasoning on the "spatial logical adjacency graph" representing the topology of the campus physical space. During this process, the system also integrates other key information: the overall aging coefficient K. age The aging degree of the core equipment within the site is calculated by weighted summation. Spatial permeability coefficient O node The system extracts information such as door and window vents and wall material properties to calculate the physical permeability of the space. This information is then combined with the final classification confidence vector P. c The GNN outputs an "interactive evolution parameter set" for each site node to be assigned, which mainly includes: The virtual loss evolution rate α is calculated by taking a weighted dot product of Pc and a preset physical wear modulus vector, and then multiplying it by K. age Nonlinear coupling calculations were performed to determine the rate at which site physical losses accumulate over time.

[0036] The interferometric evolution rate β is calculated by taking the weighted log-norm of F and comparing it with O. node Modulation calculations are performed to determine the propagation and accumulation speed of potential interference signals in space.

[0037] The fifth stage, dynamic topology mapping data generation (corresponding to S14): Finally, the interactive evolutionary parameter set (α and β) obtained from the above reasoning is used as the dynamic boundary conditions of the digital twin network model. Combined with the time series T of the task-bearing process and the task load intensity sequence L, the final "dynamic topology mapping data" is generated. This stage will further calculate: Virtual node wear value W(t): Based on α, current load intensity L0 and initial load intensity Linitial, the nonlinear fatigue characteristics of the load increment are extracted using an exponential function to generate the load stress factor, and then W(t) at any time point t is predicted.

[0038] Spatial repulsion gradient G(t): Based on the difference between β, L0 and Linitial (characterizing load fluctuation energy), and divided by the preset diffusion damping coefficient γ (reflecting the spatial physical barrier effect), G(t) is calculated at any time point t.

[0039] Through the detailed steps described above, a digital twin network model is constructed. It not only includes the static structural information of the campus physical space, but more importantly, it dynamically reflects the resource consumption, environmental evolution, physical wear and tear, and spatial interference patterns of each node in real time, providing a foundation for subsequent intelligent scheduling and management.

[0040] In this embodiment, by deploying multimodal IoT sensing devices at each site node to be assigned, the system can continuously acquire environmental time-series sensing data and spatial energy consumption monitoring data online. This data is collected within a preset time window before the task personnel enter, ensuring the capture of the site baseline state and dynamic changes before the task commences. Step S12 uses a long short-term memory autoencoder to perform window-level segmentation and feature extraction on these raw multimodal data, constructing an "environmental fingerprint fusion feature vector F". This vector includes key indicators such as the root mean square of load fluctuation, acoustic run complexity, carbon footprint deviation, and energy consumption time-series autocorrelation product. These indicators provide a comprehensive and concise quantitative expression of the current physical state of the site node from multiple dimensions, including energy intensity, acoustic disturbance, air quality, and operational mode inertia. For example, the root mean square of load fluctuations is sensitive to the high energy consumption impact of sudden starts of heavy equipment; acoustic run complexity can effectively distinguish between steady-state noise and transient impact interference; carbon footprint deviation not only reflects personnel density but also captures the gas emission characteristics produced by specific chemical reactions or biological cultivation; and the time-series autocorrelation product of energy consumption reveals the memory or randomness of equipment operating modes. The combination of these features enables the system to represent complex environments in a reduced-dimensional way, providing a high-dimensional basis for subsequent intelligent analysis.

[0041] Building upon this, steps S10001 to S1232 further classify the environmental fingerprint fusion feature vector F according to its load status. By calculating the Euclidean distance between F and the standard feature centroid vector Cc in the pre-trained model parameter library and performing normalization, a "final classification confidence vector Pc" is generated. Pc reflects the probability distribution of the current site status belonging to various preset load types (such as idle, light load, heavy load, etc.), providing a quantitative basis for assessing the site's carrying capacity and potential risks.

[0042] Step S13 is crucial for the dynamic evolution parameter inference of the digital twin model. The system utilizes a graph neural network, using the environmental fingerprint fusion feature vector F as the initial feature of each node. It also incorporates evolutionary mechanisms extracted from the campus research map regarding how different experimental tasks interfere in space, constructing information propagation constraints and iteratively inferring the spatial logical adjacency graph. Simultaneously, the system integrates the "overall aging coefficient K". age (This is achieved by weighted summation of the ratio of the service life to the design life of each core component, reflecting the health and degradation status of the hardware facilities) and "spatial permeability coefficient O". node (This is calculated using the properties of door and window vents and wall materials to characterize the physical space's ability to transmit interference signals.) Based on P c K age and O node The system outputs two core interactive evolution parameters: one is the "virtual loss evolution rate α", which is determined by the relationship between Pc and the physical wear modulus vector M. wear The weighted dot product, and with K age Nonlinear coupling calculations yielded two values: one representing the basic rate of decline in site physical health under current comprehensive operating conditions; the other representing the "interference evolution rate β," which is derived through the weighted logarithmic norm of F and O. node The modulation calculations yielded a gain coefficient characterizing the energy radiated from a node as an interference source into external space. These parameters compress the complex internal environmental fingerprint into quantifiable scalars, providing a basis for predicting physical losses and spatial interference.

[0043] Step S14 uses these interactive evolutionary parameters (α and β) as dynamic boundary conditions for the digital twin network model, combining them with the time series T of the task-bearing process and the task load intensity sequence L to generate the final "dynamic topology mapping data". This includes calculating the precise current load intensity L0 at any target time point using linear interpolation, predicting the "virtual node wear value W(t)" at any time point t based on α, and calculating the "spatial repulsion gradient G(t)" at any time point t based on β. W(t) reflects the physical loss over time, while G(t) quantifies the intensity of the node's influence on the surrounding space as a source of interference. For example, the V1 heavy electromechanical laboratory, due to its high energy consumption, high vibration characteristics, and high spatial transparency, exhibits high values ​​for both its interference evolution rate β and spatial repulsion gradient G(t), indicating that it forms a strong "interference repulsion field" during operation. In contrast, the V2 quantum optics laboratory, due to its high isolation, has low values ​​for both β and G(t), indicating that it does not generate significant external interference.

[0044] This embodiment utilizes dynamic twin mapping to predict the physical state and potential interference of each site node. This allows the system to consider not only time conflicts but also spatial interference and physical wear during scheduling, achieving optimal resource allocation and avoiding resource waste and efficiency reduction caused by blind scheduling. By combining multimodal sensing data, graph neural network inference, and refined parameter calculation, the system can quantify the mutual interference of different tasks in complex physical spaces and predict their evolution trends, providing a solid data foundation for developing proactive interference avoidance strategies. By predicting virtual node wear values, the system can assess the health status of sites and equipment in real time, identifying potential fatigue and wear in advance. This supports predictive maintenance and intelligent buffer scheduling, thereby extending the service life of facilities and reducing operation and maintenance costs. Example 2:

[0045] This embodiment builds upon the dynamic topology mapping data generated in Embodiment 1 (as shown in Table 2) to demonstrate how the system automatically generates and matches the optimal scheduling strategy for a group of tasks to be assigned based on direct threshold comparison.

[0046] Initial task requests pending processing: Task A: To be carried out in V1 (Heavy Electromechanical Laboratory), time: Wednesday morning 9:00-11:00.

[0047] Task B: To be carried out in V2 (Quantum Optics Laboratory), Wednesday morning 9:00-12:00.

[0048] Tasks C, D, and E: Multiple small, unrelated tasks, with application times spread across Wednesday throughout the day; When processing initial task requests, the system performs a series of automated pre-rendering and verification processes. It determines whether to activate the corresponding scheduling strategy by comparing the evolution parameters predicted in Example 1 with thresholds in the system's built-in "Safety and Efficiency Rule Base." Based on the comparison of preset thresholds, candidate spatiotemporal allocation adjustment schemes are generated, with specific steps including: S21. First, conduct a concurrent execution rehearsal for high-risk task combinations (task A and task B) that overlap in time to perform spatial interference conflict verification and generate the first interference avoidance strategy, which specifically includes: S2101. Extract the energy transfer efficiency from the j-th source node to the target node f, calculate and obtain the spatial conduction attenuation factor, and combine the spatial conduction attenuation factor with the spatial repulsion gradient G(t) of the j-th source node to calculate and obtain the actual influence value of the j-th source node on the target node f. The spatial propagation attenuation factor from the j-th source node to the target node f is obtained as follows: S210101. Based on the inverse square law of energy diffusion, calculate the distance attenuation factor, denoted as F. dist The formula is as follows:

[0049] d jk Let be the shortest effective propagation path length between source node j and target node f, and let d be a preset reference attenuation distance. ref This is used to adjust the steepness of the attenuation curve, set to a typical room size of the building, for example, d. ref =5 meters; When d jk When Fdist = 0, Fdist = 1, and there is no distance decay.

[0050] When d jk =d ref When Fdist=0.5, the energy is reduced by half.

[0051] S210102. Extract the energy transfer efficiency from the j-th source node to the target node f, calculate the cumulative attenuation of energy as it continuously passes through different building materials in the Y-layer propagation path, and calculate the material and medium attenuation factor F. mat It is the product of the "transmittance" of each layer of material along the path, and the formula is as follows:

[0052] in, This represents the energy transmittance through the y-th layer. Among them, the energy transmittance of the reinforced concrete floor slab (250mm thick) is 0.80 (high conduction efficiency and low attenuation); the energy transmittance of the solid brick wall (240mm thick) is 0.45 (with some insulation effect); the energy transmittance of the lightweight gypsum board partition wall (100mm thick, including keel) is 0.15 (good insulation effect); and the energy transmittance of the professional-grade vibration-damping floating floor slab is 0.05 (high insulation efficiency).

[0053] S210103. Extract the structural connection type from the j-th source node to the target node f, and calculate the structural path correction factor F. path This is used to indicate the "rigidity" of a structural connection. Rigid connections (such as monolithically cast load-bearing structures) are "highways" for vibration and structural noise, while non-rigid connections (such as connections via flexible gaskets) produce additional attenuation. If the j-th source node to the target node f is a direct load-bearing structural connection, then the structural path correction factor F path It is 1.0. If the j-th source node and the target node f are connected by adjacent non-load-bearing walls, then the structural path correction factor F path It is 0.7; If the j-th source node to the target node f is a non-direct connection across floors, then the structural path correction factor F path It is 0.4; Direct load-bearing structural connection 1.0 (e.g., between V1 and V2, connected through the same load-bearing column and floor slab, without additional attenuation); The connection between adjacent non-load-bearing walls is 0.7 (e.g., between two adjacent offices, energy must pass through the floor slab and then through the partition wall, reducing path efficiency). Cross-floor indirect connections 0.4 (e.g., V1-V3, with complex paths involving multiple reflections and structural interruptions) S210104, Distance attenuation factor F dist Calculate the material and dielectric attenuation factor F mat and structural path correction factor F path Multiply the two to obtain the spatial transmission attenuation factor from the j-th source node to the target node f, and combine the spatial transmission attenuation factor with the spatial repulsion gradient G(t) of the j-th source node to calculate the actual influence value of the j-th source node on the target node f. A preset interference safety threshold is set. When the actual impact value of the j-th source node on the target node f is higher than the interference safety threshold, it indicates that there is a spatial interference conflict between the j-th source node and the target node f, triggering the first interference avoidance strategy, which includes: keeping the execution nodes (j and f) of the two tasks unchanged, and on the system scheduling timeline, shifting the start time of one of the tasks (usually a task with lower priority or fewer dependencies) backward for at least the duration of the other task, and adding a safety interval, which is set to 5 to 10 minutes. All candidate nodes that possess the necessary resources and environmental conditions to perform the task are selected. Then, the interference calculation in S210104 is re-executed for each candidate node, and the node with a calculated actual impact value lower than the interference safety threshold Th is selected. resholdf The node f is used as the new execution location.

[0054] S2102. Extract the virtual node wear value of the f-th target node and compare it with the preset health status benchmark threshold. If the virtual node wear value is higher than the health status benchmark threshold, it is determined that the target node is in a high fatigue state after the task, and the second buffer scheduling strategy is triggered, including: calculating the percentage of virtual node wear value that is higher than the health status benchmark threshold to obtain temporary capacity; and extracting the estimated load value of each task to be scheduled. Then the scheduler only accepts new tasks whose estimated load value is less than or equal to the current temporary capacity. The system temporarily lowers the node's "scheduling capacity limit" by the same amount as the "over-threshold percentage" calculated in the previous step. The node's full capacity is considered 100%. When a node is fatigued, its ability to handle new tasks decreases accordingly. If node f's over-threshold percentage is 30%, its new scheduling capacity limit is temporarily set to 100% - 30% = 70%. During the next fixed "buffer period" (e.g., 2 hours), the system scheduler will follow this new capacity limit when assigning new tasks to node f.

[0055] Each task to be scheduled has an estimated "load value" (e.g., 80% load for high-intensity tasks, 40% for medium-intensity tasks, and 10% for low-intensity tasks). The scheduler will only accept new tasks with a load value less than or equal to the current temporary capacity limit (70%). After a fixed "buffer period" ends, the system will automatically restore the scheduling capacity limit of node f to 100%.

[0056] Example: A new task request with 80% load will be rejected because it exceeds the temporary 70% limit.

[0057] A new task request with a load of 40% will be accepted.

[0058] A new task request with a load of 10% will also be accepted.

[0059] In this embodiment, the dynamic topology mapping data generated in Embodiment 1 is used to perform concurrent execution rehearsals on combinations of high-risk tasks that overlap in time. Combined with precise spatial interference conflict verification and virtual node wear status assessment, an optimal scheduling strategy is automatically generated and matched. The core of this approach lies in step S21, which calculates the actual impact value of the source node on the target node in S2101 and compares it with a preset interference safety threshold to determine whether spatial interference conflict exists. Specifically, steps S210101 to S210103 use a multi-dimensional quantification method to comprehensively calculate the "spatial transmission attenuation factor" from the source node to the target node. This includes a "distance attenuation factor Fdist" based on the inverse square law of energy diffusion, which considers the attenuation of energy propagation due to physical distance; a "material and medium attenuation factor Fmat," which quantifies the cumulative attenuation of energy as it passes through different building materials, such as the high transmittance of piano concrete slabs and the low transmittance of professional-grade vibration-damping floating slabs; and a "structural path correction factor Fpath," which adjusts the transmission efficiency according to the structural connection type (such as direct load-bearing, adjacent non-load-bearing walls, and indirect connections across floors), reflecting the "rigidity" of the structure in the face of vibration and noise propagation. Multiplying Fdist, Fmat, and Fpath yields the final spatial transmission attenuation factor, which, when multiplied by the spatial repulsion gradient G(t) of the source node, accurately calculates the "actual impact value" of the source node on the target node. When this actual impact value exceeds a preset interference safety threshold, the system determines that a spatial interference conflict exists and triggers a "first interference avoidance strategy," which may resolve the conflict through time or location changes. In addition, step S2102 introduces a health status assessment based on virtual node wear values. By comparing the virtual node wear value W(t) of the target node with a preset health status benchmark threshold, the system can determine whether the node is in a high-fatigue state. If W(t) exceeds the threshold, a "second buffer scheduling strategy" is triggered, which temporarily reduces the scheduling capacity limit of the node (the reduction is related to the percentage exceeding the threshold), and within a fixed buffer period, only tasks with estimated load values ​​less than or equal to the new capacity limit are accepted. This ensures that fatigued nodes get sufficient rest and avoids equipment damage or performance degradation caused by overuse.

[0060] The beneficial effect of this embodiment is that, by combining refined quantification of spatial interference and predictive assessment of node fatigue state, the system can automatically identify and resolve complex spatiotemporal conflicts, generate more adaptive and forward-looking scheduling strategies, and reduce the need for manual intervention. The first interference avoidance strategy ensures that environmentally sensitive experimental tasks are protected from external interference, improving the accuracy and reliability of experimental data; the second buffer scheduling strategy effectively extends the service life of equipment and sites and reduces the risk of unplanned maintenance by avoiding overload of fatigued nodes.

[0061] Figure 1 The left side shows an isometric hidden-line diagram, depicting the campus physical space and its digital twin network model applied in this invention; the right side shows a technology roadmap, presenting the core logical flow of the method of this invention. The isometric hidden-line diagram on the left depicts a modern smart campus environment, including multiple site nodes such as teaching buildings, laboratories, and libraries. These nodes are connected through abstract data flows and networks, collectively forming a digital twin network model. This model reflects the resource occupancy and environmental evolution status of each site node in real time, which is the foundation for the intelligent management of this invention. The technology roadmap on the right elaborates on the core method steps of this invention: The first flowchart, "Acquire Dynamic Topology Mapping Data," corresponds to step S1. This step forms the data foundation of this method. It uses a digital twin network model to acquire real-time data on resource occupancy and environmental evolution of each node to be assigned within a preset time window. This data specifically characterizes the spatial interference evolution law reflected by the spatial repulsion gradient, and the evolution law of physical loss over time reflected by the virtual node wear value. The second flowchart, "Generate and Match the Optimal Scheduling Strategy," corresponds to step S2, which is the overall objective. Based on the dynamic topology mapping data acquired in step S1, the system automatically generates and matches the optimal scheduling strategy for a group of tasks to be assigned, demonstrating the intelligent decision-making capability of this invention in complex resource allocation. The third flowchart, "Execute Spatial Interference Conflict Verification," corresponds to step S21. This step is performed on high-risk task combinations with temporal overlap. The system concurrently executes a pre-simulation and performs refined spatial interference conflict verification, specifically extracting the spatial transmission attenuation factor from the source node to the target node through step S2101 and calculating the actual impact value in conjunction with the spatial repulsion gradient to determine whether a conflict exists. If a conflict exists, a first interference avoidance strategy is generated. The fourth flowchart, "Triggering Buffer Scheduling Strategy," corresponds to step S2102. This step focuses on the health status of the target node. The system extracts the virtual node wear value of the target node and compares it with a preset health status benchmark threshold. If the wear value is higher than the threshold, the target node is determined to be in a high-fatigue state, thereby triggering the second buffer scheduling strategy. This strategy avoids overuse and ensures the health of the equipment and site by reducing the node's temporary capacity.

[0062] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, min-max-normalization and Z-score standardization. The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (e.g., using the widely used pandas library in Python) configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (e.g., a DataFrame data structure). Subsequent algorithm steps will directly query and retrieve the required configuration parameters from this in-memory data structure.

[0063] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent campus integrated service management method, applied to a campus physical space containing multiple sites and laboratory nodes, characterized in that: The specific steps include: S1: Obtain dynamic topology mapping data of resource occupancy and environmental evolution of the digital twin network model of the site nodes to be assigned within a preset time window. The dynamic topology mapping data is configured to characterize the spatial interference evolution law represented by the spatial repulsion gradient and the evolution law of physical loss represented by the virtual node wear value over time in the process of each site node to be assigned carrying different tasks. S2: Based on the dynamic topology mapping data obtained in step S1, an optimal scheduling strategy is automatically generated and matched for a group of tasks to be assigned, specifically including: S21: Perform concurrent execution rehearsals for high-risk task combinations with temporal overlap to perform spatial interference conflict verification and generate the first interference avoidance strategy, specifically including: S2101: Extract the spatial conduction attenuation factor from the j-th source node to the target node f, and calculate the actual influence value of the j-th source node on the target node f by multiplying the spatial conduction attenuation factor with the spatial repulsion gradient G(t) of the j-th source node. S2102: Extract the virtual node wear value of the f-th target node and compare it with the preset health status benchmark threshold. If the virtual node wear value is higher than the health status benchmark threshold, it is determined that the target node is in a high fatigue state after the task, and the second buffer scheduling strategy is triggered.

2. The intelligent campus integrated service management method according to claim 1, characterized in that: Step S1 includes: S11. Multimodal IoT sensing devices deployed at each site node to be assigned are used to acquire multimodal IoT sensing data online. The sensing time window is set within a preset time window before the task personnel enter the site node to be assigned. The multimodal sensing data includes environmental time-series sensing data and spatial energy consumption monitoring data. S12. Using a long short-term memory autoencoder, window-level segmentation and feature extraction are performed on the environmental and energy consumption sequences in the online multimodal sensing data, and an environmental fingerprint fusion feature vector representing the current physical state of each site node to be assigned is output. S13. Using a graph neural network, the environmental fingerprint fusion feature vector is used as the initial feature of the node. The evolution mechanism of how different experimental tasks interfere in space is extracted from the campus scientific research map to construct the constraint of information propagation. Iterative reasoning is performed on the spatial logical adjacency graph to output an interactive evolution parameter set that can characterize the virtual loss evolution rate α and interference evolution rate β of each site node to be assigned during the task execution process. S14. Use the interactive evolution parameter set as the dynamic boundary condition of the digital twin network model to generate dynamic topology mapping data.

3. The intelligent campus integrated service management method according to claim 2, characterized in that: The feature extraction steps in S12 include: constructing the root mean square of load fluctuation, acoustic run complexity, carbon footprint deviation, and energy consumption time-series autocorrelation product within the perception time window for each site node to be assigned, and combining them in a predetermined order to generate an environmental fingerprint fusion feature vector; the calculation method for the root mean square of load fluctuation is as follows: obtain the instantaneous power consumption value of each sampling point within the perception time window, calculate the square of each instantaneous power consumption value, perform an arithmetic mean operation on all the calculated square values ​​to obtain the mean square value, and perform a square root operation on the mean square value to obtain the result as the root mean square of load fluctuation; the calculation method for acoustic run complexity is as follows: using a preset environmental baseline decibel value as a judgment threshold, traverse the time-series decibel sampling points within the perception time window; when two adjacent sampling points are such that the previous sampling point is less than the judgment threshold and the subsequent sampling point is greater than or equal to the judgment threshold, record... For each state transition, the total number of state transitions within the sensing time window is counted, and this total number is the acoustic run complexity. The carbon footprint deviation is calculated as follows: the average actual carbon dioxide concentration within the first preset time period is obtained, and the baseline carbon dioxide concentration value under no-task load is obtained; the absolute difference between the average actual carbon dioxide concentration and the baseline carbon dioxide concentration value is calculated, and the absolute difference is divided by the baseline carbon dioxide concentration value, and the quotient is the carbon footprint deviation. The energy consumption time-series autocorrelation product is calculated as follows: a time window sequence containing multiple consecutive instantaneous power consumption values ​​is obtained, and a comparison sequence is generated that lags the sequence by one time step; each power consumption value in the time window sequence is multiplied by the power consumption value at its corresponding position in the comparison sequence, and all product terms are summed, and the sum is the energy consumption time-series autocorrelation product.

4. The intelligent campus integrated service management method according to claim 1, characterized in that: Step S12 also includes: S10001. For each preset node load state type c, obtain the standard feature centroid vector C that uniquely corresponds to the current load state type from the pre-trained model parameter library. c ; S10002, Calculate the environmental fingerprint fusion feature vector F and the centroid vector C of each standard feature respectively. c The multidimensional Euclidean distance between them yields the distance scalar D corresponding to each load condition type. c ; S10003. To eliminate the influence of dimensions, a preset smoothing constant is added to each distance scalar for all distance scalars. Taking the reciprocal, we obtain the preliminary similarity score S for the corresponding load state type. c : S10004. Traverse all preset node load state types and repeat steps S10001 to S10003 to generate a score vector containing the preliminary similarity scores of each load state type. S1232, For all preliminary similarity scores S in the score vector c The total denominator value is obtained by performing a summation operation. Divide each preliminary similarity score by the total denominator to achieve normalization; the quotient is the final classification confidence vector P for the corresponding load state type. c .

5. The intelligent campus integrated service management method according to claim 1, characterized in that: The S13 steps specifically include: S131, Traverse N within the nodes to be assigned. d The overall aging coefficient K of the core equipment is calculated by weighted summation. age ; S132. Calculate the spatial permeability coefficient O by extracting the door and window ventilation openings and wall material properties of the nodes to be assigned to the site. node ; S133, The final classification confidence vector P based on each node c Activate the pre-built spatial logical adjacency graph; combine with the overall aging coefficient K age With spatial permeability coefficient O node It outputs the interactive evolution parameter set for each site node to be assigned, specifically including the virtual loss evolution rate α and the interference evolution rate β.

6. The intelligent campus integrated service management method according to claim 5, characterized in that: The virtual loss evolution rate α is obtained by: pre-setting a set of physical wear modulus vectors corresponding to the node load state types, representing the baseline physical loss caused to the site by the c-th load state type per unit time; and then using the final classification confidence vector P... c Perform a weighted dot product operation with the physical wear modulus vector, and then combine the result with the overall aging coefficient K. age Nonlinear coupling is performed to calculate and obtain the virtual loss evolution rate.

7. The intelligent campus integrated service management method according to claim 5, characterized in that... The interference evolution rate β is obtained as follows: a set of feature interference transformation weight vectors corresponding to the dimensions of the environmental fingerprint fusion feature vector F are preset, which correspond to the interference transformation capabilities of the root mean square of load fluctuation, acoustic run complexity, carbon footprint deviation, and energy consumption time series autocorrelation product, respectively; the weighted logarithmic norm of the environmental fingerprint fusion feature vector F is calculated and compared with the spatial permeability coefficient O. node Modulation is performed, and the interference evolution rate β is calculated.

8. The intelligent campus integrated service management method according to claim 2, characterized in that: Step S14 includes: S141. Obtain the time series T and the task load intensity sequence L describing the process of the task carried by the site node to be allocated. The sequence contains a series of discrete time points and the measured or preset load intensity value at each time point. S142. Set the initial state of all site nodes to be assigned at time t=0, where the initial spatial repulsion gradient is 0, the initial virtual node wear value is 0, and the initial load intensity is extracted from the sequence and denoted as L. initial ; S143. Calculate the precise current load intensity L0 at any target time point t using linear interpolation; S144. Obtain the virtual loss evolution rate α, and calculate the current load intensity L0 relative to the initial load intensity L. initial The normalized increment is obtained by performing a natural logarithmic operation on the load increment ratio to extract its nonlinear growth characteristic term. Finally, the three are multiplied together, and the nonlinear fatigue characteristics of the normalized increment are extracted using an exponential function to generate the load stress factor. Based on the current load state, the virtual node wear value W(t) at any time point t is predicted. S145. Obtain the interferometric evolution rate β, and calculate the current load intensity L0 and the initial load intensity L. initial The difference is multiplied by the load fluctuation energy; the load fluctuation energy is multiplied by the interference evolution rate β and divided by the preset diffusion damping coefficient γ, which reflects the spatial physical barrier effect, so as to calculate and obtain the spatial repulsion gradient G(t) at any time point t.

9. The intelligent campus integrated service management method according to claim 1, characterized in that: Step S2101 specifically includes: S210101. Based on the inverse square law of energy diffusion, calculate and obtain the distance attenuation factor; S210102. Extract the energy transfer efficiency from the j-th source node to the target node f, calculate the cumulative attenuation of energy as it passes through different building materials in the Y layer along the propagation path, and calculate the material and medium attenuation factors. S210103. Extract the structural connection type from the j-th source node to the target node f, and calculate the structural path correction factor; where: if the j-th source node to the target node f is a direct load-bearing structural connection, the structural path correction factor is 1.0; if the j-th source node to the target node f is an adjacent non-load-bearing wall connection, the structural path correction factor is 0.7; if the j-th source node to the target node f is a cross-floor non-direct connection, the structural path correction factor is 0.

4. S210104. Multiply the distance attenuation factor, material and medium attenuation factor and structural path correction factor to obtain the spatial conduction attenuation factor from the j-th source node to the target node f. Then, multiply the spatial conduction attenuation factor with the spatial repulsion gradient G(t) of the j-th source node to calculate the actual influence value of the j-th source node on the target node f.

10. An intelligent campus integrated service management system, applied to the intelligent campus integrated service management method according to any one of claims 1-9, comprising: The data sensing submodule is configured to acquire multimodal IoT sensing data online through multimodal IoT sensing devices deployed at each site node to be assigned. The sensing time window is set within a preset time window before the task executor enters the site node to be assigned. The multimodal sensing data includes environmental time-series sensing data and spatial energy consumption monitoring data. The feature extraction submodule is configured to perform window-level segmentation and feature extraction on the environmental and energy consumption sequences in the online multimodal sensing data through a long short-term memory autoencoder, and output an environmental fingerprint fusion feature vector representing the current physical state of each site node to be assigned. The evolution parameter inference submodule is configured to use a graph neural network, with the environmental fingerprint fusion feature vector as the initial feature of the node, and to use the evolution mechanism of how different experimental tasks interfere in space extracted from the campus scientific research map to construct the constraints of information propagation, to perform iterative inference on the spatial logical adjacency graph, and output an interactive evolution parameter set that can characterize the virtual loss evolution rate α and interference evolution rate β of each site node to be assigned during task execution. The dynamic topology generation submodule is configured to use the interactive evolution parameter set as the dynamic boundary condition of the digital twin network model to generate dynamic topology mapping data. The concurrent execution pre-drill module is configured to perform concurrent execution pre-drills on high-risk task combinations with temporal overlap based on state topology mapping data, in order to perform spatial interference conflict verification and generate a first interference avoidance strategy; and extract the virtual node wear value of the f-th target node and compare it with a preset health status benchmark threshold. If the virtual node wear value is higher than the health status benchmark threshold, it is determined that the target node is in a high fatigue state after the task, and the second buffer scheduling strategy is triggered.

Images