Intelligent management system applied to exhibition center

Abstract

The application discloses a smart management system applied to a convention and exhibition center, relates to the technical field of convention and exhibition smart management, realizes unified and high-precision data input through protocol conversion and multi-modal denoising, supports second-level alarm and topology reconstruction through edge-side cooperation of anomaly detection and adaptive management, improves the comprehensive interpretation accuracy of passenger flow and environmental situation by using a multi-modal fusion algorithm, then maps the sensing information into a virtual venue situation by means of digital twinning and a self-learning mechanism, realizes continuous optimization of sensor calibration and algorithm weight, finally introduces multi-Agent complex network dynamics to simulate the interaction evolution of large-scale crowds in emergency scenarios, and provides closed-loop management of active prediction and emergency layout for smart convention and exhibition, so as to realize leap-forward improvement in multiple aspects such as monitoring accuracy, safety disposal and resource scheduling, and cope with convention and exhibition layout adjustment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent management of exhibitions and conferences, in particular to an intelligent management system applied to an exhibition and conference center. BACKGROUND

[0002] In super-large exhibition centers, stadiums and smart parks, hundreds to thousands of heterogeneous sensing devices, including high-definition cameras, infrared thermal imagers, RFID / Bluetooth positioning probes, environmental monitors (temperature and humidity, air quality, noise), Wi-Fi probes, face and license plate recognition terminals, etc., are deployed simultaneously and connected to access control, fire control, security robots, building automation and energy consumption management systems. These devices are distributed in exhibition halls, corridors, public areas and back offices, and are often blocked and have blind spots due to exhibition stand reorganization, temporary partitions, weather changes or night lighting adjustments; the sudden increase in crowd density during visitor peak times causes video congestion, metal structure multipath fading and wireless signal interference; and inconsistent device models and communication protocols result in a variety of data formats and scattered interfaces. The management party expects to obtain fine-grained crowd heat maps, security situations and environmental indicators within seconds to support intelligent scheduling and predictive operation and maintenance, in order to ensure safety, improve experience and optimize energy consumption.

[0003] According to the search, a kind of intelligent management system applied to exhibition center is disclosed in Chinese invention patent with application publication No.CN112202857A. Including system server group and with the system server group connected with several access control equipment, several identity recognition equipment, several self-service machines, several self-service invoice printers, several self-service inquiry machines, client, security network module. The present application unifies the management system by the security data server and the subserver separately connected with it, including exhibition center, host party, exhibitor, purchaser, exhibition related service provider, audience, realizes data intercommunication, unified management. From the architecture level, it is integrated design, style uniform operation is simple, convenient for later maintenance, improves management efficiency, reduces labor cost, increases enterprise benefit.

[0004] However, the existing intelligent venue system generally relies on scattered subsystems:

[0005] CCTV video is monitored manually, and passenger flow statistics and environmental monitoring are independent, lacking a unified data aggregation and accurate fusion mechanism. The core technical problem is: how to realize real-time data alignment and adaptive calibration across protocols and modalities under the conditions of dynamic changes in venue layout and multi-source heterogeneous data environment, so as to ensure the accurate consistency of multi-end sensing data in time, space and numerical dimensions, and provide reliable input for subsequent AI fusion analysis and edge anomaly detection? This problem directly restricts the response speed of the system to sudden safety incidents, the accurate interpretation of high-density crowds and the fine control of complex environmental parameters, and is a key bottleneck that needs to be broken through for the fine management of current intelligent exhibition venues.

[0006] To this end, the application provides a smart management system applied to an exhibition center. SUMMARY

[0007] (I) Technical problems solved

[0008] In view of the deficiencies in the prior art, the application provides a smart management system applied to an exhibition center, which realizes unified and high-precision data input through protocol conversion and multi-modal denoising; supports second-level alarm and topology reconstruction through edge-side cooperation of anomaly detection and adaptive management; and improves the comprehensive interpretation accuracy of passenger flow and environmental situation through a multi-modal fusion algorithm; then maps the perception information into a virtual venue situation by means of digital twinning and self-learning mechanism, realizes continuous optimization of sensor calibration and algorithm weight; finally introduces multi-Agent complex network dynamics to simulate the interaction evolution of large-scale crowds in emergency scenarios, provides closed-loop management of proactive prediction and emergency layout for smart exhibition, so as to realize leap-forward improvement in monitoring accuracy, safety disposal, resource scheduling and other aspects, and solve the technical problems recorded in the background art.

[0009] (II) Technical solutions

[0010] To achieve the above purpose, the application realizes the technical solutions as follows: a smart management system applied to an exhibition center, comprising: when a large amount of heterogeneous sensor data is gathered simultaneously, a multi-source perception network utilizes protocol conversion and time sequence alignment algorithm to perform format unification and noise filtering on original readings, and transforms them into a multi-source data set;

[0011] After the preprocessed multi-source data set is time-corrected, a multi-modal fusion module performs cross-modal aggregation on each channel feature vector through a convolution network and dynamic weight scheduling, and generates a multi-modal fusion vector;

[0012] When the multi-modal fusion vector is sent to an edge node, an anomaly detection and edge adaptive engine utilizes anomaly scoring and perception topology reconstruction mechanism to perform real-time alarm and device rapid calibration on suspicious passenger flow or extreme environmental data, and generates an adaptive perception configuration;

[0013] When the multi-source fusion data is preliminarily calibrated, a digital twinning and self-learning module integrates the multi-modal fusion vector, anomaly score and adaptive perception configuration into a situation vector of a virtual venue through a situation mapping function and a self-learning feedback function, and dynamically updates the sensor weight;

[0014] Through multi-Agent modeling and introduction of network dynamics mechanism in the update rule, the virtual crowd distribution is time-sequenced and crowded effect is deduced, and simulation error is introduced between simulation results and actual monitoring data to optimize emergency evacuation scheme and layout configuration.

[0015] Further, the multi-source perception network is configured with a heterogeneous protocol conversion module at the gateway side, to map the communication protocols of various sensors to a unified interface protocol based on sensor node identification and network access identification, and to maintain an access dynamic mapping table at the network convergence end to fix the correspondence between sensor node identification and perception channel.

[0016] Further, the original data from the perception channels are aligned in time, a data alignment transformation operation is defined based on a timing fitting method, and the timing misalignment is dynamically corrected during the alignment process; the aligned data are denoised by a weighted spectral reorganization function to obtain a multi-source data set.

[0017] Further, a learnable feature mapping operator is constructed for each perception channel to map the multi-source data set collected from the perception channel to a feature vector; the feature vectors of all channels are spliced or adaptively fused to obtain a multi-channel fusion feature vector.

[0018] Further, after introducing a set of learnable or adaptive weights, a dynamic weight is calculated by a weight scheduling function to represent the weight occupied by the perception channel; the weighted feature vector is subjected to a vector element level weighting operation according to the obtained dynamic weight to obtain a multi-modal fusion vector with dynamic cooperative characteristics, and the weight scheduling function updates the weight when the layout of the exhibition hall or the state of the crowd changes.

[0019] Further, an anomaly detection function is used to measure the deviation degree of the multi-modal fusion vector in a high-dimensional space, and an anomaly score is output; if the obtained anomaly score exceeds a preset dynamic threshold function, an abnormal event is generated, triggering edge adaptive perception and rapid calibration; the corresponding timestamp and anomaly level are written into an edge alarm queue, wherein the abnormal event includes abnormal intensity, timestamp, and associated channel information.

[0020] Further, a decision function for edge adaptive perception is called to reassign and activate part of the standby sensors, and the key sensors are rapidly calibrated to generate a perception network reconstruction decision set and a sensor incremental calibration parameter, and the updated calibration parameter matrix.

[0021] Further, a virtual entity corresponding to a key entity is created in a digital twin space, a twin node is defined using sensor identification and network mapping information, and a node set of the digital twin space is obtained;

[0022] A situation mapping function is called to project the multi-modal fusion vector, the anomaly score, and the perception network reconstruction decision set into the digital twin to obtain a global situation vector; when the anomaly score is calculated, the anomaly score is projected as a risk intensity display of each node in the virtual exhibition hall by the situation mapping function.

[0023] Further, given the global situation vector and internal simulation parameters at the current moment, a predictive scenario simulation function is introduced to perform scenario calculation on the future period to obtain twin state prediction.

[0024] The twin state prediction is displayed in the form of scenario animation or trend chart in the digital twin visualization interface, and outputs predictive indicators, including passenger flow peak period, safety alarm probability and environmental load.

[0025] Further, the self-learning feedback function uses abnormal events, reconstructed decision set and situation vector to iteratively update sensor weights, calibration parameters and fusion algorithm, wherein: the self-learning feedback function automatically collects the difference between the actual scene and the twin simulation in each learning period, and incorporates it into the learning loss function, drives the iterative correction of the calibration parameter matrix set and dynamic weight of the sensor, and updates the state of the virtual entity node.

[0026] Further, a plurality of Agent sets are added in the digital twin space, and a state vector of each Agent is constructed, which is used to describe the position, speed, target interest point, emergency behavior tendency and other state parameters of the Agent in the digital twin environment, and the interaction rule between people and the scene is defined in the digital twin space.

[0027] Further, for each Agent, an update rule is defined in the digital twin space to simulate the evolution of its behavior, the adjacency relationship is defined in the Agent set, and the network dynamics mechanism is introduced in the update rule; the states of all Agents are updated in parallel in the digital twin space to generate the overall crowd distribution at the next moment.

[0028] Further, an error measurement operator is constructed to compare the obtained virtual crowd distribution with the actual monitoring data, and if the simulation error obtained is continuously higher than expected, the parameters are adjusted in the digital twin space;

[0029] If the simulation accuracy gradually improves, the simulation result is used for venue layout adjustment, emergency drill plan or optimization of security and evacuation strategy, an evaluation result about the accuracy of crowd simulation and the effectiveness of the strategy is generated, and key improvement suggestions or parameter correction are fed back to the digital twin and self-learning engine.

[0030] (Three) beneficial effects

[0031] The application provides a smart management system applied to an exhibition center, and has the following beneficial effects:

[0032] Step one realizes the alignment and cross-protocol unification of the multi-source data set X clean (t), which ensures that the multi-source sensor data remains highly available and low noise in subsequent processes; the output not only filters redundant signals, but also lays a solid foundation for multi-modal fusion.

[0033] Step two, under the cooperation of deep learning and dynamic weight distribution, maps different channel data into a multi-modal fusion vector Y(t), which can flexibly cope with the sudden changes in exhibition hall layout and maximize information complementarity and accuracy using dynamic cooperative algorithms, increasing the unified judgment ability of multi-dimensional elements such as people flow, security and environment.

[0034] Step three, with the help of abnormal scoring δ(t) and adaptive perception network, the edge side realizes second-level disposal, closely combines distributed AI with physical environment requirements, quickly detects abnormal events and emergency linkage, shares necessary information summary between edge nodes and cloud, not only reduces bandwidth load, but also keeps critical decisions within controllable range, forming a pattern of local efficiency and global insight.

[0035] Step four, inject the situation vector U(t) into the digital twin model to establish a realistic mapping of the overall situation of the venue; and continuously absorb information such as abnormal scoring δ(t) with the self-learning module to dynamically update sensor parameters and fusion weights, so that the digital twin is no longer limited to visualization, but through self-learning and actual system to form a cycle of evolution, promoting the venue management to move towards active prediction and iterative optimization of deep intelligence.

[0036] Step five, on this basis, superimpose crowd multi-agent simulation, make the situation vector U(t) interface with complex network dynamics to simulate the evolution of people flow in special scenarios; and compare the simulation results with the actual multi-modal fusion vector Y(t) to correct the crowd model and layout strategy. This composite linkage of large-scale simulation, digital twin and edge adaptive can predict congestion and resource bottlenecks in advance and achieve accurate decision-making. Through self-learning, the virtual and real are connected, making it possible to quickly iterate people flow safety, environmental control and resource scheduling, generating overall efficiency with greater depth and foresight than isolated technologies. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The figure is a process schematic diagram of the intelligent management system of the exhibition center. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0039] Please refer to Figure 1 The present application provides an intelligent management system applied to an exhibition center, comprising,

[0040] Step one, when the camera, infrared and RFID data in the super-large exhibition center through the unified gateway centralized reporting, multi-source perception network synchronous call protocol conversion module and high-level denoising algorithm, the original observation value X raw Performing communication alignment, format uniformity and redundancy filtering operations, reconstructing it into a unified and time-synchronized multi-source data set X clean (t);

[0041] The step one includes the following contents:

[0042] Step 101, the composition of multi-source perception network and heterogeneous protocol conversion

[0043] Define sensor node identification And network access identification To uniquely identify different sensor devices and their corresponding data stream sources; at the network level, by configuring protocol conversion modules at the main switch or edge gateway, the proprietary communication protocols of various sensors are mapped to a unified interface protocol, for example, the data of Zigbee, LoRaWAN or custom TCP / IP protocol are encapsulated into a unified format, and then indexed with sensor node identification Management.

[0044] For different data types (video stream, infrared signal stream, RFID reading, environmental monitoring value, etc.), multi-source perception channels are established respectively (denoted as , etc.), and access dynamic mapping tables (MappingTable) are maintained at the network convergence end to fix the correspondence between sensor node identification And the corresponding perception channel So that the next step can directly reference these identification information for data processing.

[0045] In use, through the management mode of sensor node identification And network access identification , new types of sensors can be quickly accessed when the venue scale is expanded or deployed, without the need for large-scale changes to the network structure; dynamic mapping tables (MappingTable) are used to manage the mapping relationship between sensor node identification And perception channel , so that the network can quickly remap when the venue layout changes temporarily or the equipment is updated, which can improve the overall adaptation ability.

[0046] Step 102, data alignment and advanced denoising preprocessing

[0047] Because different perception channels ( The data sampling frequency of the original data (e.g., video, infrared, radar, etc.) is different from the data distribution delay, and the original data needs to be uniformly time-aligned, and a data alignment transformation operation Γ is defined, as follows:

[0048] X align (t) = Γ(X raw (t-τ1), X raw (t-τ2), …, X raw (t-τ n n); β

[0049] In the formula, X raw (t-τ n n) represents the original data of the sensor n at time t-τ n n; τ n n is the data delay estimation of the nth perception channel.

[0050] β is an internal reference parameter of the alignment operator, which is used to adjust the weight distribution of the historical window in the alignment operation. For example, when interpolating or compensating between multi-frame rate video streams and low-frequency infrared data, β can adjust the smoothness of the interpolation function.

[0051] The Γ operation can be based on wavelet filtering, singular spectrum reconstruction (SVD-based), or other advanced time sequence fitting methods, and dynamically corrects the time sequence misalignment in the alignment process.

[0052] In order to ensure the accuracy of subsequent multi-modal data fusion, the aligned data needs to be denoised. In order to enhance the anti-interference ability and take into account the characteristics of multi-modal, a weighted spectrum reorganization function Θ can be defined as follows:

[0053] X denoise (t) = Θ(X align (t), λ, γ)

[0054] In the formula, X denoise (t) is the multi-source data vector after alignment; λ is an adaptive filtering threshold adjustment factor, which is used to dynamically determine the attenuation coefficient of each frequency band; γ is a smoothness control factor of the denoising algorithm, and the value range is determined according to the requirements.

[0055] The weighted spectrum reorganization function Θ can not only attenuate high-frequency noise by using hybrid wavelet transform, but also retain core feature signals at the low-frequency end. Through frequency domain weighting, various irregular interferences in the environment, such as local occlusion or sudden electromagnetic interference, can be effectively filtered.

[0056] The sensor node identifier and the channel identifier established in step 101 are inherited to the result processed in this step, and the aligned and clean multi-source data set X clean (t) is obtained.

[0057] In use, the delay difference of multi-source data is corrected by the Gamma operator, more accurate timing alignment can be achieved, and confusion of person trajectory or environment measurement caused by data misplacement during subsequent fusion can be avoided. The weighted spectrum regularization function Theta is used for weighted spectrum regularization processing, and differential attenuation of noise characteristics of each channel can be realized between different perception channels. Finally, the multi-source data set X clean (t) is obtained. By introducing an advanced timing reconstruction algorithm in the alignment operation, the data mismatch rate in a high dynamic scene can be significantly reduced.

[0058] Step two, when the preprocessed multi-source data set X clean (t) is completed, the timing correction is completed, and the multi-modal fusion module can be deeply mined. Through the convolution network and the dynamic weight scheduling, the embedding mapping and adaptive aggregation of each channel feature vector F x (t) are performed to generate a multi-modal fusion vector Y(t), which can finely depict the flow and environment situation while eliminating single-source blind areas.

[0059] The step two includes the following contents:

[0060] Step 201, feature extraction and latent space mapping

[0061] For each perception channel A learnable feature mapping operator Phi is defined x The clean data collected from the perception channel at time t is represented as The feature vector of the perception channel in the latent space is denoted as F x (t):

[0062]

[0063] In the formula: denotes the de-noised data of the perception channel at time t; eta x is a feature extraction hyperparameter of the perception channel , which is used to control the setting of the feature mapping operator Phi x in terms of convolution kernel size, attention gate structure or auto-encoding depth, and different channels can adopt different hyperparameters eta x ; Phi x uses a multi-layer deep network or other advanced embedding functions, such as a multi-head attention network, to extract high-order mode information of the mode.

[0064] To obtain cross-modal correlation information, the feature vectors F x (t) of all channels are spliced or adaptively fused to obtain a unified latent representation F all (t), and the aggregation operation is defined as Stacking and alignment of different feature vectors:

[0065]

[0066] where F all (t) is the preliminary multi-channel fusion feature vector, still in the high-dimensional latent space, without any weight or dynamic synergy strategy yet.

[0067] In use, by mapping operator Φ x High-order nonlinear mapping is performed on each channel data, which can discover the potential correlation between different modalities such as vision and infrared that cannot be captured by a single channel; aggregation of features can preserve the relative independence of each channel, and each channel data is processed by a dedicated mapping operator Φ x The hyperparameter η x , which customizes the optimal extraction method for each modality before fusion, can achieve more fine-grained differentiated processing, rather than blindly processing all modalities with the same model. When aggregating in the latent space, instead of directly linearly splicing, high-dimensional attention or adaptive mechanism can be combined to make different channel features retain the maximum plasticity in subsequent dynamic synergy.

[0068] Step 202, dynamic weight allocation and synergy output

[0069] To achieve adaptive fusion of multiple channels, a set of learnable or adaptive weights {ω x (t)} is introduced, where ω x (t) represents the weight of the perception channel at time t, satisfying: 0≤ω x (t)≤1, and

[0070] The value of dynamic weight ω x (t) can be dynamically generated by comprehensive evaluation of the real-time state of the venue (such as current crowd distribution, emergency events, and availability of each sensor), or calculated by a weight scheduling function Ψ:

[0071] ω x (t)=Ψ(F all (t),ξ)

[0072] where: ξ represents the hyperparameter set inside the weight scheduling function, such as attention threshold, local density threshold or adaptive threshold, etc.; Ψ can be based on the multi-head attention mechanism of Transformer or graph network structure, to learn the importance distribution of each modality from the multi-channel fusion feature vector F all (t);

[0073] According to the obtained dynamic weight ω x(t), the final multi-modal fusion output Y(t) can be defined as:

[0074]

[0075] where: is a vector element level weighting operation; F x (t) is a single-channel latent feature vector;

[0076] Through this weighted sum, a multi-modal fusion vector Y(t) with dynamic coordination characteristics can be obtained. When the layout of the exhibition hall or the state of the crowd changes, the weight scheduling function Ψ will update the weight ω x (t), so that the features of important channels account for a larger proportion in the fusion result, thereby realizing adaptive scheduling across modalities.

[0077] When some sensor data in the venue causes the single-channel latent feature F x (t) to abnormally deviate due to occlusion or failure, the weight scheduling function Ψ can automatically reduce its dynamic weight ω x (t), avoiding interference of this modality with the overall result, based on the multi-channel fusion feature vector F all (t) and the dynamic weight ω x (t), a multi-modal fusion vector Y(t) can be quickly generated, without the need for repeated backtracking of the original data for large-scale retraining, meeting the real-time requirements of the convention and exhibition center;

[0078] The introduction of the weight scheduling function Ψ performs context-aware dynamic analysis on the multi-channel fusion feature vector F all (t), realizing importance allocation across modalities, rather than fixing the proportion of a certain channel based on prior experience; through the form of element-level weighted sum, the contribution of single-channel features in multi-modal fusion can be fine-tuned to the dimension level, allowing independent gain or attenuation of different dimension feature components.

[0079] Step three, when the multi-modal fusion vector Y(t) is sent to the edge node and needs to be identified in real time for security and crowd anomalies, the anomaly detection and edge adaptive engine uses the high-dimensional anomaly score δ(t) and the local perception topology reconstruction algorithm to generate a perception network reconstruction decision set R(t) for suspicious crowd or extreme environmental data for immediate alarm and rapid calibration of equipment;

[0080] The step three includes the following contents:

[0081] Step 301, real-time anomaly detection

[0082] Since the multi-modal fusion vector Y(t) contains multi-source information such as vision, infrared, RFID, and environmental sensors across modalities, this vector can more comprehensively reflect the current venue situation;

[0083] To measure the degree of deviation and fluctuation of the multi-modal fusion vector Y(t) in the high-dimensional space from time t-Δt to time t, the following integral type anomaly detection function Ω is defined, which outputs the anomaly score δ(t):

[0084]

[0085] In the formula: δ(t) is the anomaly score, the larger the value means that there is a stronger abnormal signal or severe fluctuation in the time window [t-Δt, t]; Δt is the size of the backtracking time window;

[0086] Y(s) is the multi-modal fusion vector at time s, which contains the information after the fusion of visual, infrared, RFID and other multi-source features;

[0087] H is a symmetric positive semi-definite matrix (or a positive definite matrix), which is used to weight the dimensions of Y(s);

[0088] is the first derivative (gradient) of the multi-modal fusion vector Y(s) with respect to time s, which describes the rate of change of Y(s) in the time domain, represents its Euclidean norm (size);

[0089] β is the fluctuation sensitivity coefficient, which is used to balance the numerical deviation and the time change rate in the high-dimensional space;

[0090] If the obtained anomaly score δ(t) exceeds the preset dynamic threshold function Δv thd , it is considered that an abnormal event ε(t) occurs at time t; at this time, the subsequent (step 302) edge adaptive perception and rapid calibration can be triggered; Δv thd is the internal hyperparameter under this step and the venue dynamic environment, which is used to dynamically update the specific value of Δ(·) according to the current crowd density, sensor health, and historical normal data distribution and other factors;

[0091] If the abnormal event ε(t) is generated, the corresponding timestamp and abnormal level are written into the edge alarm queue, and the abnormal event ε(t) contains the abnormal intensity (represented by the anomaly score δ(t)), the timestamp, the associated channel information, etc.

[0092] When used, it can still ensure the sensitive capture of abnormalities under different exhibition hall layouts, crowd densities and sensor state changes, avoid the interference of simple noise in traditional methods, and dynamically update the threshold function Δv thresholdThe system can dynamically adjust in real time according to historical data and current context of the venue, enhancing the detection accuracy of unexpected events (such as short-term surge in passenger flow). Using methods such as manifold comparison or local topological measure, it can more flexibly identify rare patterns or extreme data in high-dimensional fusion feature space. By incorporating adaptive decision threshold into the hyperparameter v for online learning, the anomaly detection algorithm itself can continuously evolve with the operation of the venue.

[0093] Step 302, edge adaptive perception and rapid calibration

[0094] Upon receiving the abnormal event ε(t), the multi-modal fusion vector Y(t) is obtained, as well as the dynamic mapping table of the sensors (used to locate specific devices and perception channels

[0095] Define the decision function Π for edge adaptive perception, according to the current detected abnormal situation, sensor availability and venue physical conditions, reassign and activate part of the standby sensors, and perform rapid calibration on the key sensors;

[0096] The core execution logic is:

[0097] R(t) = Π(ε(t), Y(t), ζ)

[0098] Where: R(t) is the perception network reconstruction decision set, recording the sensor identification that needs to be activated or disabled and its new sampling frequency, network routing, edge computing node allocation, etc.

[0099] ζ is a hyperparameter, which may involve maximum allowed network load, minimum sampling period limit, temporary layout change information, etc.

[0100] The decision function Π can make dynamic perception decisions based on the load evaluation model on the edge side, physical distribution constraints, and the judgment of abnormal types (such as safety type or environmental type).

[0101] For sensors that may have accuracy deviations, or specific perception channels with sudden noise increases in abnormal environments, rapid calibration operations can be performed:

[0102]

[0103] Where: represents the perception channel the previous calibration parameter matrix (such as the imaging geometric transformation matrix of the camera in the visual channel, the sensitivity factor of the environmental sensor, etc.);

[0104] κ is the incremental calibration value automatically calculated according to the current abnormal type and sensor health status;

[0105] This represents a nonlinear merging with the original parameter matrix, which can be solved iteratively based on comparative data from benchmark measurement points or nearby sensors.

[0106] This process is performed on the edge side, which can greatly speed up the correction of sudden failures or abnormal drift.

[0107] Step 302 finally produces the perception network reconstruction decision set R(t) and the calibrated calibration parameter matrix. Synchronous updates enable subsequent multi-source data acquisition and fusion to be performed under the new configuration.

[0108] In practice, upon detecting an anomaly, edge nodes do not need to wait for long-term cloud-based decisions. They can quickly reconstruct the local perception network based on the anomaly event ε(t) and the multimodal fusion vector Y(t), significantly reducing latency in responding to sudden events. Through a rapid calibration mechanism, sensor parameters can be adjusted promptly in case of sudden changes in the local scene (e.g., the sudden addition of large equipment to an exhibition area causing infrared sensor saturation), maintaining subsequent data quality. By tightly linking the anomaly event ε(t) with the rapid calibration operation, sensor topology and parameters can be reconstructed in the shortest possible time, rather than relying on manual intervention or a centralized update process.

[0109] Step 4: When the edge has generated the perception network reconstruction decision set R(t) and completed the initial calibration of the multi-source fusion data, the digital twin and self-learning module uses the situation mapping function Φ DT With self-learning feedback function Γ Learn The multimodal fusion vector Y(t), the anomaly score δ(t), and the perception network reconstruction decision set R(t) are integrated into the virtual venue's situation vector U(t) and the sensor weights are dynamically updated to form a scalable twin model.

[0110] Step four includes the following:

[0111] Step 401: Digital Twin Model Construction and Real-time Situation Mapping

[0112] To connect with various key entities in the physical venue (venue areas, sensor devices, sensing channels) (etc.) correspond one-to-one in the digital twin space Create virtual entities corresponding to key entities in the middle, and use sensor identifiers. Using network mapping information to define twin nodes, the digital twin space is then... The node set can be expressed as

[0113]

[0114] in represents the twin node corresponding to a certain sensor entity, d Zone may represent a virtual representation of an exhibition area or a specific functional area, etc. represents the twin node corresponding to a certain sensor entity, represents the twin node corresponding to a certain perception channel in the digital twin space;

[0115] In order to project the multi-modal fusion vector Y(t) and the abnormal score δ(t) and other information obtained in the previous step into the digital twin, a situation mapping function Φ DT is constructed, which has the following specific form:

[0116] U(t) = Φ DT (Y(t), R(t), δ(t), χ)

[0117] U(t) is the global situation vector in the digital twin space, which records the state update of each twin node (such as temperature and humidity conditions, security risk levels, and people flow density); χ is a set of defined internal hyperparameters or rules, such as the coordinate system of the twin space, the visualization style of each node, and the connected topology between exhibition areas, etc.

[0118] When the abnormal score δ(t) is calculated, the digital twin platform projects this score as a risk intensity display of each node in the virtual venue through the situation mapping function Φ DT When the perception network reconstruction decision set R(t) and the risk intensity display are received, the parameters of the corresponding nodes in the virtual model are updated synchronously, and the changes in sensor topology and risk level are displayed in real time.

[0119] When in use, the global situation vector U(t) presents people flow heat, dangerous hotspots, and security trends in a graphical manner, providing a global view for managers or upper-level business modules; the associated information of the sensor topology (i.e., the perception network reconstruction decision set R(t), the multi-modal fusion vector Y(t), and the abnormal score δ(t)) is also stored in the twin model, breaking through the limitations of traditional digital twins that are often only static visualizations or single-dimensional parameter mappings, and forming a dynamic interconnected venue ecological model.

[0120] Step 402, predictive scenario simulation and strategy deduction

[0121] In the digital twin space , a predictive scenario simulation function Ψ Sim is introduced on the global situation vector U(t) of the current time t, and the predictive scenario simulation function Ψ Sim performs scenario calculation on the future time period [t, t+T]:

[0122]

[0123] where U future (s) represents the twin state prediction at time s (s≤t+T);

[0124] internal simulation parameters, such as passenger flow growth model, exhibition hall layout change hypothesis, and probability distribution of possible emergencies, etc.

[0125] Predictive scenario simulation function Ψ Sim Based on differential equations, reinforcement learning simulation, or high-order methods such as multi-agent system simulation, non-simple linear extrapolation;

[0126] When making predictions, different strategy hypotheses (such as emergency evacuation plan A, air conditioning load balancing plan B, security personnel scheduling plan C, etc.) can be imported simultaneously, their effects can be simulated in parallel in the digital twin model, and several groups of Combined with the objective function (such as energy consumption, safety risk, and personnel congestion), the manager can be provided with optimization decision suggestions.

[0127] Twin state prediction U future (s) can be displayed in the form of scenario animation or trend chart in the digital twin visualization interface, and the predictive indicators (such as passenger flow peak period, safety alarm probability, and environmental load) can be reported to the upper business platform in specific numerical values or icons;

[0128] When used, before an event occurs in a real venue, the impact of the event can be quickly evaluated through multi-scenario deduction of a virtual scene; the decision-making time for temporary layout adjustment or emergency disposal is greatly shortened. With multi-modal fusion data Y(t) and abnormal information δ(t), the twin model has more accurate input basis, improving the simulation accuracy and reliability. Through parallel multi-strategy simulation, different layouts or emergency plans can be switched in the digital twin with one key, the impact on passenger flow, safety, and environment can be quantified in real time, which surpasses the traditional rough simulation or single-path prediction.

[0129] Step 403, self-learning closed loop and sensor weight iteration

[0130] Abnormal events δ(t), reconstructed decision set R(t), and situation vector U(t) are comprehensively utilized to iteratively update sensor weights, calibrate parameters, and fusion algorithms, and a self-learning feedback function Γ Learn is introduced.

[0131]

[0132] where: W newNew optimal version of calibration parameters or other sensor configurations used in the previous step of anomaly detection and edge adaptive processing, which can cover up to the processing weight, offset, etc. of each sensor in multi-modal fusion ;

[0133] Optimization result of dynamic weight ω of each perception channel defined in step two dynamic coordination algorithm, which realizes more accurate cross-modal fusion after several rounds of learning; x ; Hyperparameters, including learning rate, history memory decay factor, alignment strategy of twin model and physical data, etc.

[0134] Self-learning feedback function Γ Learn Collects the difference between the actual scene and the twin simulation in each learning cycle (or when an abnormal event occurs), and incorporates it into the learning loss function to drive the iterative correction of the calibration parameter matrix set W of the sensor and the dynamic weight ω x . The situation vector U(t) also updates the state of the virtual entity node after each iteration to ensure that the model and the real situation gradually converge; when the updated calibration parameter set W new and the optimized dynamic weight ω are determined, these updated parameters can be used in the next data collection and fusion process to complete the truly accurate use.

[0135] When used, the deviation between real observation and twin prediction is incorporated into the closed-loop learning process to realize dynamic correction of sensor offset, algorithm weight, etc. For exhibition centers with frequent layout changes, this self-learning mechanism can continuously track venue structure adjustment to ensure that sensor coverage and fusion weight evolve accordingly; the digital twin prediction error (the difference between the physical reality of the venue and the twin simulation) is directly incorporated into the learning feedback link, combined with multiple information such as abnormal information and edge adaptive decision R(t) for multiple constraint optimization, that is, both the deviation W of the sensor itself and the dynamic weight ω of the cross-modal fusion algorithm can be dynamically corrected and real-time optimized x , realizing double self-learning of parameters and weights.

[0136] Step five, after obtaining the iteratable situation vector U(t), the large-scale crowd simulation component performs time-series calculation and crowd effect deduction on the virtual crowd distribution S(t) through multi-Agent modeling and introducing network dynamics mechanism in the update rule Θ crowd , and introduces simulation error ∈(t) between the simulation result and the actual monitoring data to optimize the emergency evacuation plan and layout configuration.

[0137] The step five includes the following contents:

[0138] Step 501, Multi-Agent model construction and digital twin scene expansion

[0139] In the digital twin space Add a multi-agent set representing individuals (or group units) of the crowd Denoted as:

[0140]

[0141] Where N is the size of the virtual crowd, Represents the i-th Agent, usually corresponding to one or a group of visitors with similar behavior characteristics; define the state vector s i (t) of each Agent, which describes its position, speed, target interest point, emergency behavior tendency, etc. in the digital twin environment, which may include:

[0142]

[0143] Where x i (t), y i (t) represent the coordinate position of the Agent, v i (t) can represent the speed or motion direction related quantity, and p i (t) can reflect the degree of emergency or crowd sensitivity, etc.

[0144] Based on the venue coordinate system and situation vector U(t) in the digital twin model of the fourth step, define the interaction rules between people and the scene (such as traffic constraints, queuing rules, danger area avoidance, etc.) in the digital twin space So that the multi-agent system is consistent with the actual physical layout in the virtual space;

[0145] When used, by defining the Agent set and the state vector, the differences in behavior of different visitors in space movement, interest point selection, emergency evacuation path, etc. can be simulated, and through scene expansion operation, the real-time situation information (such as abnormal position, security situation) contained in the situation vector U(t) can directly affect the Agent behavior, ensuring the dynamic consistency of physical and virtual crowds; by including the advanced behavior characteristics p i (t) such as emergency degree and crowd sensitivity in the Agent state vector, a more realistic crowd interaction model can be exhibited in the scene of sudden events or high-density exhibition areas.

[0146] Step 502, complex network dynamics simulation and multi-agent collaborative calculation

[0147] For each In the digital twin space Define a set of high-dimensional dynamic equations or update rules Θ crowd, which can be formalized as:

[0148] s i (t+Δt)=Θ crowd (s i (t),U(t),α i ,ζ)

[0149] where s i (t) is the state vector at time t;

[0150] U(t) is derived from the digital twin state output of the fourth step (including venue layout, security hotspots, historical passenger flow density, etc.), providing external environmental reference for Agent behavior;

[0151] α i represents unique behavior preferences or attributes (such as whether there is a VIP channel, whether it is extremely sensitive to congestion, etc.);

[0152] ζ is a set of simulation parameters independent of this step, such as step size Δt, avoidance strength, following distance, etc.

[0153] To consider the mutual influence between the crowd, the adjacency relationship (such as communication, proximity, or social relationship) can be defined in the Agent set, and the network dynamics mechanism (such as the attraction / repulsion force of neighbor behavior, group synchronization, and traffic bottleneck effect) can be introduced in the update rule Θ crowd , which can be represented as an adjacency matrix M (dimension N x N), where M ij >0 represents the interaction strength between In the digital twin space , the states of all Agents are updated in parallel, and the overall crowd distribution S(t+Δt) at the next time is recorded. Numerical simulation or multi-threaded GPU parallel can be used to accelerate this process to meet the computing needs of large-scale scenarios (N can reach tens of thousands or even more).

[0154] The environmental indices and abnormal alarm information in the state vector U(t) can directly affect the path or speed chosen by the Agent, reflecting the two-way influence relationship between the crowd and the environment in the real scene. By embedding the state vector U(t) into the core equation of crowd dynamics, the real-time coupling between the environment and the crowd is realized, and the adjacency matrix M is used to strengthen the group coordination characteristics in large-scale parallel computing, which is more suitable for dynamic calculation of tens of thousands of Agents in terms of algorithm.

[0155] Step 503, simulation and actual monitoring data comparison and strategy feedback

[0156] ​​​The virtual crowd distribution S(t) obtained after the digital twin simulation is run is compared with the actual monitoring data to construct an error measurement operator Π comp , which is as follows:

[0157] ∈(t) = Π comp (S(t), Y(t), γ)

[0158] where S(t) is the virtual crowd state; Y(t) includes the number of passenger flow and thermal distribution fused in the previous step and other multi-modal observations; γ is a comparison hyperparameter (such as a spatial position matching threshold, a time tolerance, a clustering radius, etc.); ∈(t) reflects the difference between the simulation result and the actual venue passenger flow or crowd distribution;

[0159] If the simulation error ∈(t) obtained is continuously high, it indicates that the simulation model or parameter Θ crowd does not match the reality, and further parameter adjustment needs to be made in the digital twin space ; if the simulation accuracy gradually improves, the simulation result can be used for venue layout adjustment, emergency drill plan, or security and evacuation strategy optimization, forming a linkage with the fourth step self-learning module.

[0160] Finally, an evaluation result about the crowd simulation accuracy and strategy effectiveness is generated, and key improvement suggestions or parameter corrections (such as the update of the adjacency matrix M and the improvement of the potential function in the update rule Θ crowd ) are fed back to the digital twin and self-learning engine to further improve the system's control over real crowd dynamics.

[0161] When in use, the large-scale simulation in the digital twin is no longer an isolated script-based deduction, but can be compared with the actual sensor observation Y(t) for differences, and continuously converges to a model that better fits the real scene. Through dynamic tracking of the simulation error ∈(t), it can be accurately evaluated how certain exhibition area layout or evacuation scheme works in the real environment, and the scheme can be iterated in time to minimize safety risks or reduce congestion. The use of multi-modal fusion output Y(t) to verify the accuracy of large-scale crowd simulation can improve the level of description of real crowd distribution and hierarchical behavior.

[0162] Step 501 incorporates Agent entities into the digital twin space to expand the fine description of crowd behavior; step 502 combines the environmental situation U(t) and the network interaction between Agents to complete the time-series crowd dynamics calculation; step 503: compares the simulation result with the real data Y(t), measures the simulation error ∈(t), and feeds back to the digital twin and self-learning model in time to iteratively optimize the venue layout and emergency plan.

[0163] Through the comprehensive mechanism of virtual-real contrast + multi-agent cooperation + complex network dynamics, the intelligent exhibition system can more forwardly handle large-scale crowd management problems, from emergency evacuation under emergencies to congestion control in daily exhibitions, and can more scientifically and effectively dispatch resources. In the mutual echo and data intercommunication with the first four steps, the crowd simulation system proposed in this step not only plays a role alone, but also produces global gain in overall cooperation.

[0164] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0165] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0166] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be realized by other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0167] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected to achieve the purpose of the embodiment according to actual needs.

[0168] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in 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 management system applied to convention and exhibition centers, characterized by: include, When massive amounts of heterogeneous sensor data are simultaneously aggregated, the multi-source sensing network uses protocol conversion and time-series alignment algorithms to perform format unification and noise filtering on the raw readings, transforming them into a multi-source dataset. After the preprocessed multi-source dataset has undergone temporal correction, the multimodal fusion module performs cross-modal aggregation on the feature vectors of each channel through a convolutional network and dynamic weight scheduling to generate a multimodal fusion vector. After the multimodal fusion vector is sent to the edge node, the anomaly detection and edge adaptive engine uses anomaly scoring and perception topology reconstruction mechanism to provide real-time alarms and rapid device calibration for suspicious pedestrian flow or extreme environment data, and generate adaptive perception configuration. When the initial calibration of the multi-source fusion data is completed, the digital twin and self-learning module integrates the multi-modal fusion vector, anomaly score and adaptive perception configuration into the virtual venue's situation vector and dynamically updates the sensor weights through the situation mapping function and self-learning feedback function. By using multi-agent modeling and introducing network dynamics mechanisms into update rules, we can perform time-series calculations on the distribution of virtual crowds and extrapolate congestion effects. Furthermore, we can introduce simulation errors between simulation results and actual monitoring data to optimize emergency evacuation plans and layout configurations.

2. The intelligent management system for convention and exhibition centers according to claim 1, characterized in that: The multi-source sensing network is configured with a heterogeneous protocol conversion module on the gateway side. Using sensor node identifiers and network access identifiers, it maps the communication protocols of various sensors to a unified interface protocol. At the network aggregation end, it maintains an access dynamic mapping table to fix the correspondence between sensor node identifiers and sensing channels.

3. The intelligent management system for convention and exhibition centers according to claim 2, characterized in that: The raw data from the sensing channel is uniformly time-stamped and aligned. The data alignment transformation operation is defined based on the time series fitting method, and the time series misalignment is dynamically corrected during the alignment process. The weighted spectral renormalization function is used to denoise the aligned data to obtain a multi-source dataset.

4. The intelligent management system for convention and exhibition centers according to claim 3, characterized in that: For each sensor channel, a learnable feature mapping operator is constructed to map the multi-source dataset collected from the sensor channels into feature vectors. The feature vectors of all channels are concatenated or adaptively fused to obtain a multi-channel fused feature vector.

5. The intelligent management system for convention and exhibition centers according to claim 4, characterized in that: After introducing a learnable or adaptive set of weights, dynamic weights are calculated by a weight scheduling function to represent the weights of the sensing channels: The weighted feature vector is weighted at the vector element level based on the obtained dynamic weights to obtain a multimodal fusion vector with dynamic collaborative characteristics. When the layout of the exhibition hall or the flow of people changes, the weight scheduling function updates the weights.

6. The intelligent management system for convention and exhibition centers according to claim 5, characterized in that: An anomaly detection function is used to measure the degree of deviation of the multimodal fusion vector in high-dimensional space and output an anomaly score; If the obtained anomaly score exceeds the preset dynamic threshold function, an anomaly event is generated, triggering edge adaptive perception and rapid calibration, and the corresponding timestamp and anomaly level are written to the edge alarm queue. The anomaly event includes anomaly intensity, timestamp, and associated channel information.

7. The intelligent management system for convention and exhibition centers according to claim 6, characterized in that: The decision function for edge adaptive sensing is invoked to reallocate and activate some spare sensors, and key sensors are quickly calibrated to generate a perception network reconstruction decision set and sensor incremental calibration parameters, and the calibration parameter matrix is ​​updated synchronously.

8. The intelligent management system for convention and exhibition centers according to claim 7, characterized in that: Create virtual entities corresponding to key entities in the digital twin space, define twin nodes using sensor identification and network mapping information, and obtain the node set of the digital twin space; The situation mapping function is called to project the multimodal fusion vector, anomaly score, and perception network reconstruction decision set into the digital twin to obtain the global situation vector; after the anomaly score is calculated, the situation mapping function projects the anomaly score into the risk intensity display of each node in the virtual venue.

9. The intelligent management system for convention and exhibition centers according to claim 8, characterized in that: Given the global situation vector and internal simulation parameters at the current moment, a predictive scenario simulation function is introduced to perform scenario calculations for future time periods and obtain twin state predictions. The twin status prediction is displayed in the digital twin visualization interface in the form of scene animation or trend chart, and outputs predictive indicators, including peak passenger flow periods, security alarm probability, and environmental load.

10. The intelligent management system for convention and exhibition centers according to claim 9, characterized in that: The self-learning feedback function uses anomalous events, reconstructed decision sets, and situation vectors to iteratively update sensor weights, calibration parameters, and the fusion algorithm, where: The self-learning feedback function automatically collects the differences between the actual scene and the twin simulation in each learning cycle, incorporates them into the learning loss function, drives the iterative correction of the sensor calibration parameter matrix set and dynamic weights, and updates the state of the virtual entity nodes.

11. The intelligent management system for convention and exhibition centers according to claim 10, characterized in that: In the digital twin space, a set of multiple agents is added, and a state vector is constructed for each agent. The state vector is used to describe its position, speed, target interest points, and emergency behavior tendencies in the digital twin environment. The interaction rules between humans and the scene are defined in the digital twin space.

12. The intelligent management system for convention and exhibition centers according to claim 11, characterized in that: For each Agent, update rules are defined in the digital twin space to simulate its behavioral evolution. Adjacency relationships are defined in the Agent set, and network dynamics mechanisms are introduced into the update rules. The states of all Agents are updated in parallel within the digital twin space to generate the overall population distribution at the next time step.

13. The intelligent management system for convention and exhibition centers according to claim 12, characterized in that: An error measurement operator is constructed to compare the obtained virtual population distribution with the actual monitoring data. If the obtained simulation error continues to be higher than expected, the parameters are adjusted in the digital twin space. If the simulation accuracy gradually improves, the simulation results can be used to adjust the venue layout, optimize emergency drill plans or security and evacuation strategies, generate evaluation results on the accuracy of crowd simulation and the effectiveness of strategies, and feed back key improvement suggestions or parameter corrections to the digital twin and self-learning engine.

Images