Intelligent exhibition hall control system based on Internet of Things

Abstract

The invention relates to the technical field of intelligent exhibition hall control, and discloses an intelligent exhibition hall control system based on the Internet of Things. The system synchronizes the operation state of a physical exhibition hall through a digital twin model, and performs operation situation slicing and feature extraction on a virtual mirror image to obtain a situation feature vector. And constructing an exhibition hall operation situation evolution graph based on historical data, mapping the real-time feature vectors to graph nodes, and performing multi-hop state deduction along the edges, thereby generating a future situation prediction path and identifying potential abnormal nodes. And finally, simulation regulation and control are performed in the digital twinborn model according to a deduction result, and a preventive regulation and control instruction is generated. According to the scheme, a closed loop from real-time state sensing to future evolution prediction and active regulation and control is realized, and the predictability and management and control perspectiveness of the operation risk of the complex exhibition hall are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent exhibition hall control technology, specifically to an intelligent exhibition hall control system based on the Internet of Things. Background Technology

[0002] Current smart exhibition halls mostly employ IoT and digital twin technologies to achieve data collection and 3D visualization monitoring. Existing solutions can map physical states, but their control logic relies on preset rules or single parameter threshold responses. Their drawback lies in the fact that multi-source data streams are merely presented in parallel or simply aggregated, failing to extract structured features that characterize the overall operational status of the exhibition hall from the temporally interwoven whole. The data remains in a raw or coarse-grained form, making it difficult to depict the dynamic relationships between subsystems, resulting in superficial assessments of the operational status and an inability to form a deep understanding.

[0003] Existing predictive technologies mostly target single-device fault warnings or parameter trend extrapolation, severing the inherent networked connections within the system. Due to the lack of modeling of evolutionary patterns and transition paths in historical state sequences, these methods cannot describe "how" the system state evolves or "why" different states are related. This results in a lack of foresight in management decisions, with regulatory actions becoming reactive remedies for already occurring anomalies, making it difficult to identify and intervene in the early stages of risk evolution.

[0004] Existing technologies are insufficient to meet the needs of in-depth understanding, dynamic simulation, and proactive control of complex exhibition hall operations. A method is needed that can construct structured situational representations in real time and model state evolution networks based on these representations, enabling proactive control from perceiving the current state to predicting evolutionary paths and ultimately supporting simulation verification. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent exhibition hall control system based on the Internet of Things to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent exhibition hall control system based on the Internet of Things, the system comprising: The digital twin synchronization module is used to construct a digital twin model of the exhibition hall and collect the physical space operation flow of the target exhibition hall in real time through the Internet of Things sensor network. The physical space operation flow is synchronously injected into the digital twin model of the exhibition hall to form a dynamically updated virtual exhibition hall operation image. The operational status slicing module is used to perform operational status slicing processing on the virtual exhibition hall's operational image, generate a continuous time slice sequence, and perform status feature distillation extraction on each time slice to obtain a status feature vector that reflects the multi-dimensional operational correlation features within the time slice. The situation evolution map construction module is used to learn the evolution pattern of situation feature vector sequences based on historical operation data and construct the situation evolution map of the exhibition hall. The situation simulation and prediction module is used to map the real-time generated situation feature vector to the exhibition hall operation situation evolution map for node positioning, and perform multi-hop state simulation along the map edge to generate situation prediction paths and potential abnormal nodes for multiple future moments. The preventive control simulation module is used to perform parameter intervention simulation in the digital twin model of the exhibition hall based on the predicted situation path and potential abnormal nodes, and generate a set of preventive control instructions for the physical exhibition hall equipment.

[0007] Preferably, the step of synchronously injecting the physical space operation flow into the digital twin model of the exhibition hall to form a dynamically updated virtual exhibition hall operation mirror includes: The digital twin model of the exhibition hall is used to mirror the physical equipment layout, spatial topology, and environmental sensor network of the target exhibition hall; The physical space operation flow includes heterogeneous device operation signal flow, multi-dimensional environmental perception flow, and visitor activity trajectory flow; The heterogeneous device operation signal stream is parsed and normalized to generate a standardized device status event stream, wherein each device status event includes a device identifier, timestamp, operation type and set of operating parameters. The multidimensional environmental perception stream is spatiotemporally aligned and data is completed to generate a complete gridded environmental state field. The gridded environmental state field discretizes the exhibition space into a three-dimensional grid, and each grid stores the environmental parameter value corresponding to the location of the grid. The visitor activity trajectory flow is cleaned and segmented into behavior segments to generate a behavior sequence composed of continuous behavior units; The standardized device status event stream is mapped to the corresponding virtual device nodes in the exhibition hall digital twin model, and the real-time status attributes of the virtual device nodes are updated. The gridded environmental state field is superimposed onto the spatial grid of the digital twin model of the exhibition hall to update the environmental state attributes of the virtual space; Each behavioral unit in the behavioral sequence is bound to a virtual visitor agent, and the virtual visitor agent is driven to perform the corresponding behavior in the space of the digital twin model of the exhibition hall, and the virtual space impact of the behavior is recorded. Through an event-driven engine, based on device status events, environmental status updates, and virtual visitor agent behavior, the virtual interactions and influences between these events are calculated, enabling the overall state of the digital twin model of the exhibition hall to dynamically evolve over time, forming a virtual exhibition hall operation mirror.

[0008] Preferably, the step of performing operational status slicing processing on the virtual exhibition hall's running image to generate a continuous time slice sequence includes: Each time slice contains a snapshot of all equipment, environment, and visitor status within its corresponding time period; Set a fixed time slice length and sliding step size, and continuously extract a fixed duration of the running state interval on the timeline of the virtual exhibition hall running mirror with the sliding step size as the interval; For each captured running state interval, extract the state change sequence of all virtual device nodes, the environmental parameter change curve of all spatial grids, and the behavioral interaction records of all virtual visitor agents within the running state interval; The sequence of device state changes, the curve of environmental parameter changes, and the records of behavioral interactions within the operating state interval are encapsulated into a structured data packet in chronological order to form a time slice. In the structured data packet, the device state change sequence, environmental parameter change curve and behavior interaction record are indexed by a unified time axis to ensure that data from different dimensions at the same moment can be aligned. All generated time slices are sorted according to their corresponding start timestamps to form the continuous time slice sequence.

[0009] Preferably, the step of performing situation feature distillation extraction on each time slice to obtain a situation feature vector reflecting multi-dimensional operational correlation features within the time slice includes: State transition pattern mining is performed on the equipment state change sequence in the time slice to extract the pattern features of equipment operation. The pattern features include periodic patterns, burst patterns and collaborative patterns between related equipment. Spatiotemporal correlation features are extracted from the environmental parameter change curves in the time slice to extract the spatiotemporal features of environmental evolution. The spatiotemporal features include the spatial distribution of parameter gradients, anomaly propagation paths, and the correlation strength with device location. Clustering of behavioral interaction records in time slices is performed to extract group behavior patterns and group characteristics of visitor behavior. These group characteristics include gathering hotspots, movement patterns, and the frequency and type of interaction with devices. Construct a ternary association graph among device pattern features, environmental spatiotemporal features, and behavioral group features. In the graph, nodes represent different types of features, and edge weights represent the co-occurrence strength and causal association strength among features within the time slice. Graph embedding learning is performed on the ternary association graph to encode the association information of all nodes and edges in the graph into a low-dimensional vector of fixed dimension. The low-dimensional vector is the situation feature vector, which contains the comprehensive association state of the device, environment and behavior in the time slice.

[0010] Preferably, the construction of the exhibition hall operation status evolution map includes: Collect all time slices generated within the historical operating cycle and their corresponding situation feature vectors; Each situation feature vector is used as a state node in the evolution map of the exhibition hall's operational status. Calculate the situation evolution similarity between any two state nodes. The situation evolution similarity is calculated based on the distance between the two situation feature vectors in the feature space and the proximity of the time slices they represent on the time axis. In the evolution map of the exhibition hall's operational status, a directed edge is established for a pair of state nodes whose status evolution similarity exceeds a preset connection threshold. The direction of the edge points from the earlier node in time to the later node in time, and the weight of the edge is positively correlated with the status evolution similarity. Community discovery is performed on the evolution map of the exhibition hall's operation status, and multiple state nodes with similar evolution patterns are clustered into the same community, with each community representing a typical exhibition hall operation mode. A pattern label is extracted for each community, and transfer conditions are marked for the transfer paths between communities, forming an evolution map of the exhibition hall's operational status with a hierarchical community structure.

[0011] Preferably, the step of mapping the real-time generated situation feature vector to the exhibition hall's operational situation evolution map for node localization, and performing multi-hop state deduction along the map edges to generate situation prediction paths and potential abnormal nodes for multiple future time points, includes: Calculate the feature similarity between the real-time generated situation feature vector and the features of each state node in the exhibition hall operation situation evolution map; The real-time situation feature vector is mapped to the state node with the highest feature similarity, which serves as the current location node in the map. Starting from the positioning node, traverse all the next-hop nodes connected by its outgoing edges, and calculate the transition probability to each next-hop node based on the weight of the edge. Based on the aforementioned transition probabilities, a random walk algorithm is used to perform multi-step state deduction starting from the location node, generating multiple possible evolution paths starting from the current state. Each path consists of a series of state nodes connected in chronological order. Analyze the terminal nodes of each evolution path, identify nodes that belong to abnormal operation mode communities or are highly similar to historical abnormal nodes, and mark these nodes as potential abnormal nodes. The evolution paths containing the potential abnormal nodes are selected as the situation prediction paths that require special attention.

[0012] Preferably, the step of performing parameter intervention simulation in the digital twin model of the exhibition hall to generate a set of preventative control instructions for the physical exhibition hall equipment includes the following steps: S1: Select a situation prediction path that needs to be focused on, and simulate the operation of the exhibition hall along the situation prediction path in the digital twin model of the exhibition hall, starting from the current virtual state. S2: When the simulation reaches the potential abnormal node, pause the simulation and analyze the key equipment parameters and environmental conditions that caused the operating state to enter the potential abnormal node; S3: In the digital twin model of the exhibition hall, multiple adjustment schemes are preset for the key equipment parameters, and the corresponding parameters of the virtual equipment nodes are modified; S4: Restart the simulation from the state after modifying the parameters, and observe whether the subsequent operation in the digital twin model of the exhibition hall can avoid the potential abnormal nodes or switch to a better operating mode. S5: Evaluate the effectiveness of each adjustment plan in avoiding anomalies or transitioning to a better mode, as well as the equipment adjustment costs required to implement the plan; S6: Considering both effectiveness and control costs, select the optimal adjustment scheme and convert the equipment parameters to be adjusted and their target values ​​in the adjustment scheme into specific, executable equipment control instructions; Traverse all the situation prediction paths that require key attention, repeat steps S1 to S6 for each path, and summarize the control commands generated by all paths to obtain the set of preventive control commands.

[0013] Preferably, the evaluation of the effectiveness of each adjustment scheme in avoiding anomalies or transitioning to a better mode, and the equipment adjustment costs required for implementing the scheme, includes: After implementing the adjustment plan in the digital twin model of the exhibition hall, the simulation continues for a period of time in the future, and the sequence of situation feature vectors generated by the simulation is extracted. Map the situation feature vector sequence back to the exhibition hall operation situation evolution map, and observe whether its evolution path completely avoids the potential abnormal nodes, or whether its final state node belongs to a better operation mode community. If the abnormal node is successfully avoided, record the minimum graph distance between it and the abnormal node as a safety margin; if it is switched to a better community, record the performance improvement index of the new community mode compared with the original path end community mode. The safety margin and performance improvement indicators are summarized, and the effectiveness score of the adjustment scheme is calculated through a preset utility function; The range and frequency of equipment parameter changes required by the adjustment plan are analyzed. Combined with the equipment energy consumption model and equipment wear model, the additional energy consumption and equipment wear caused by implementing the adjustment plan are estimated. The estimated additional energy consumption and equipment wear are quantified into a control cost score using a preset cost function; The effectiveness score and the regulation cost score are input into a multi-objective decision model to obtain a comprehensive evaluation result of the adjustment scheme.

[0014] Preferably, the construction of the digital twin model of the exhibition hall includes: Collect architectural information model data, equipment asset list and physical topology connection relationship of the target physical exhibition hall, and construct a static three-dimensional geometric model of equipment-space association; Based on the equipment asset list, a virtual device agent model is constructed for each type of physical equipment, which includes multi-level attributes and behavioral rules. The multi-level attributes include at least equipment identification, model specifications, performance parameters and communication protocols. The behavioral rules define the dynamic response relationship of equipment state as control commands, operating parameters and external environment change. In the static three-dimensional geometric model, based on the physical topology connection relationship and the actual spatial coordinates of the devices, the corresponding virtual device proxy model is instantiated and deployed to form a virtual device node network; An integrated simulation engine for physical rules of the exhibition hall environment is used to simulate the propagation and interaction of light, temperature, humidity and air flow in a multi-dimensional spatial grid. A virtual sensor network and a virtual actuator network are embedded in the static three-dimensional geometric model. The virtual sensor network is used to perceive the state of the mesh environment calculated by the environmental physical rules simulation engine during the simulation. The virtual actuator network is used to receive control commands and drive the corresponding virtual device agent model to perform actions. Establish data interaction and causal logic connections between the virtual device node network, virtual sensor network, virtual actuator network and environmental physical rule simulation engine to form a high-fidelity simulation system that can dynamically respond to and evolve the state of equipment control commands, environmental disturbances and visitor interactions, namely the digital twin model of the exhibition hall.

[0015] Preferably, the system further includes: an instruction issuance and closed-loop verification module, comprising: Each instruction in the preventive control instruction set is subject to executability verification and conflict detection to ensure that the equipment involved in the instruction is online and its status is controllable, and that there are no resource or timing conflicts between the instructions. The verified instructions are sorted according to their preset execution time windows and encapsulated into a data frame format recognized by the device control protocol. When the preset triggering conditions or execution time of the instruction are met, the corresponding control instruction data frame is sent to the target device control terminal through the Internet of Things communication link; Receive command response signals and equipment status feedback returned by the equipment control terminal in real time to confirm the command execution results; The execution results of the instructions and the feedback of the device status after execution are synchronously injected into the digital twin model of the exhibition hall to update the virtual exhibition hall operation image and trigger a new round of situational awareness, prediction and control simulation cycle.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By performing operational status slicing on the dynamically updated virtual exhibition hall's operational image, a continuous time-slice sequence is generated. For each time-slice, status feature distillation is performed to extract status feature vectors reflecting multi-dimensional operational correlations. This transforms real-time, heterogeneous IoT data streams into a series of structured, information-condensed state representation units. Noise is filtered from massive asynchronous data, retaining core information that characterizes the overall system behavior and internal relationships. This allows for precise quantification of the comprehensive operational status, providing a unified and semantically clear data foundation for in-depth analysis.

[0017] An evolutionary map of operational status is constructed based on historical sequence learning patterns, and real-time feature vectors are mapped to map nodes, performing multi-hop state deduction along the edges. The state transition rules are modeled as a dynamic topological network. The deduction process generates possible future state evolution paths based on network connection logic and probability, which can intuitively reveal the transmission chain of state changes and automatically identify abnormal nodes with weak connections or low probability in the path, realizing the transformation from predicting a single result to anticipating the complete evolution process and its potential bottlenecks.

[0018] The projected future trajectory and potential anomaly nodes are fed back into the digital twin model for parameter intervention simulation, forming a closed loop from analysis and prediction to decision verification. For risk nodes in the path, the impact of different control strategies on the subsequent evolutionary trajectory is simulated in virtual space. By comparison and selection, the optimal strategy that effectively avoids risks and guides the system towards the desired state is identified, thereby generating precise preventative control commands. This ensures that control actions are based on the simulation and verification of the future evolutionary trajectory and are executed before physical anomalies manifest, thus transforming system management from a passive response to a proactive, forward-looking guidance based on evolutionary prediction. Attached Figure Description

[0019] Figure 1 This is a timing diagram of the IoT-based intelligent exhibition hall control system described in this invention. Figure 2 A flowchart for generating the operational status slice sequence; Figure 3 A flowchart for constructing an evolution map of the exhibition hall's operational status; Figure 4A two-bar chart comparing the edge weights and transition probabilities of the location node N72315 to the next hop node; Figure 5 A path diagram illustrating the evolution of the two-dimensional feature space for the operation of an intelligent exhibition hall. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 This invention provides an IoT-based intelligent exhibition hall control system. The system includes: constructing a virtual model synchronized with the physical exhibition hall using digital twin technology, and performing perception, analysis, prediction, and control of the exhibition hall's operational status based on this model. The overall implementation scheme of the system is as follows: A digital twin synchronization module constructs a digital twin model of the exhibition hall and collects the physical space operation flow of the target exhibition hall in real time through an IoT sensor network. This physical space operation flow is synchronously injected into the digital twin model of the exhibition hall, forming a dynamically updated virtual exhibition hall operation image. An operation status slicing module performs operation status slicing processing on the virtual exhibition hall operation image, generating a continuous time slice sequence. For each time slice, status feature distillation extraction is performed to obtain a status feature vector reflecting the multi-dimensional operational correlation characteristics within that time slice. A status evolution map construction module learns the evolution pattern of the status feature vector sequence based on historical operational data, constructing an exhibition hall operation status evolution map. A status inference and prediction module maps the real-time generated status feature vector to the exhibition hall operation status evolution map for node positioning and performs multi-hop state inference along the map edges, generating status prediction paths and potential abnormal nodes for multiple future time points. Based on the predicted path and potential abnormal nodes, the preventive control simulation module performs parameter intervention simulation in the digital twin model of the exhibition hall, generating a set of preventive control instructions for the physical exhibition hall equipment.

[0022] In one embodiment of the present invention, the construction of the digital twin model of the exhibition hall is achieved by collecting architectural information model data, equipment asset list, and physical topology connections of the target physical exhibition hall to construct a static three-dimensional geometric model of equipment-space association. Based on the equipment asset list, a corresponding virtual equipment proxy model containing multi-level attributes and behavioral rules is constructed for each type of physical equipment. The multi-level attributes include at least equipment identification, model specifications, performance parameters, and communication protocols. The behavioral rules define the dynamic response relationship of equipment state to changes in control commands, operating parameters, and external environment. In this static three-dimensional geometric model, the corresponding virtual equipment proxy model is instantiated and deployed according to the physical topology connections and the actual spatial coordinates of the equipment, forming a virtual equipment node network. An integrated exhibition hall environment physical rule simulation engine is used to simulate the propagation and interaction of light, temperature, humidity, and airflow in a multi-dimensional spatial grid. A virtual sensor network and a virtual actuator network are embedded in the static three-dimensional geometric model. The virtual sensor network is used to perceive the grid environment state calculated by the environment physical rule simulation engine in the simulation, and the virtual actuator network is used to receive control commands and drive the corresponding virtual equipment proxy model to perform actions.

[0023] A data interaction and causal logic connection is established between the virtual device node network, virtual sensor network, virtual actuator network, and environmental physical rule simulation engine to form a high-fidelity simulation system capable of dynamically responding to and evolving the state of device control commands, environmental disturbances, and visitor interactions—this is the digital twin model of the exhibition hall. When the physical space operation flow is synchronously injected into this digital twin model, it mirrors the physical device layout, spatial topology, and environmental sensor network of the target exhibition hall. This physical space operation flow includes heterogeneous device operation signal flow, multi-dimensional environmental perception flow, and visitor activity trajectory flow. Protocol parsing and signal normalization are performed on the heterogeneous device operation signal flow to generate a standardized device state event flow, where each device state event includes a device identifier, timestamp, operation type, and set of operating parameters. Spatiotemporal alignment and data completion are performed on the multi-dimensional environmental perception flow to generate a complete gridded environmental state field. This gridded environmental state field discretizes the exhibition hall space into a three-dimensional grid, with each grid storing the environmental parameter values ​​corresponding to that grid location. The visitor's activity trajectory flow is cleaned and segmented into behavioral fragments to generate a behavioral sequence composed of continuous behavioral units. This standardized device state event flow is mapped to the corresponding virtual device nodes in the digital twin model of the exhibition hall, updating the real-time state attributes of the virtual device nodes. The gridded environmental state field is overlaid onto the spatial grid of the digital twin model of the exhibition hall, updating the environmental state attributes of the virtual space. Each behavioral unit in the behavioral sequence is bound to a virtual visitor agent, and the virtual visitor agent is driven to execute the corresponding behavior in the space of the digital twin model of the exhibition hall, recording the virtual space impact of the behavior. Through an event-driven engine, based on device state events, environmental state updates, and virtual visitor agent behavior, the virtual interactions and influences between these elements are calculated, allowing the overall state of the digital twin model of the exhibition hall to dynamically evolve over time, forming a virtual mirror image of the exhibition hall's operation.

[0024] In practical implementation, the construction of the exhibition hall's digital twin model involves collecting architectural information model data, equipment asset lists, and physical topology connections of the target physical exhibition hall. This creates a static 3D geometric model linking equipment and space. The architectural information model data includes precise 3D geometric information about the exhibition hall's internal walls, floor structure, and exhibition area divisions. The equipment asset list details the unique identifiers, models, specifications, and installation locations of all lighting, air conditioning, and display equipment. The physical topology connections describe the power supply lines and communication network connection paths between the equipment. Based on the equipment asset list, a corresponding virtual device proxy model is constructed for each type of physical equipment, containing multi-level attributes and behavioral rules. These multi-level attributes include at least equipment identification, model specifications, performance parameters, and communication protocols. The behavioral rules define the dynamic response relationship of the equipment state to changes in control commands, operating parameters, and the external environment. For example, the performance parameters of the virtual air conditioning equipment proxy model include cooling power and fan speed range, and the behavioral rules include calculating response delays and energy consumption changes based on the current load and ambient temperature after receiving a temperature adjustment command.

[0025] In the static 3D geometric model, based on the physical topology connections and the actual spatial coordinates of the devices, corresponding virtual device proxy models are instantiated and deployed to form a virtual device node network. Each node in the virtual device node network corresponds one-to-one with a physical device in the physical exhibition hall. An integrated exhibition hall environment physical rule simulation engine is used to simulate the propagation and interaction of light, temperature, humidity, and airflow in a multi-dimensional spatial grid. This engine calculates the diffusion and attenuation of environmental parameters between grids based on computational fluid dynamics and heat conduction equations. Virtual sensor networks and virtual actuator networks are embedded in the static 3D geometric model. The virtual sensor network is used to perceive the grid environment state calculated by the exhibition hall environment physical rule simulation engine during simulation. The virtual actuator network receives control commands and drives the corresponding virtual device proxy models to perform actions. The node positions of the virtual sensor network are consistent with the positions of the real environmental sensors deployed in the physical exhibition hall. Establish data interaction and causal logic connections between the virtual device node network, virtual sensor network, virtual actuator network and the exhibition hall environment physical rule simulation engine to form a high-fidelity simulation system that can dynamically respond to and evolve the state of equipment control commands, environmental disturbances and visitor interactions, namely the exhibition hall digital twin model.

[0026] In practical implementation, when the physical space operation flow is synchronously injected into the digital twin model of the exhibition hall, the digital twin model is used to mirror the physical equipment layout, spatial topology, and environmental sensor network of the target exhibition hall. The physical equipment layout includes the installation coordinates of lighting equipment at the top of the exhibition area and the installation coordinates of air conditioning equipment on the corridor side. The physical space operation flow includes heterogeneous equipment operation signal flow, multi-dimensional environmental perception flow, and visitor activity trajectory flow. The heterogeneous equipment operation signal flow comes from device controllers from different manufacturers, the multi-dimensional environmental perception flow comes from distributed temperature and humidity sensors and carbon dioxide concentration sensors, and the visitor activity trajectory flow comes from ultra-wideband positioning base stations. Protocol parsing and signal normalization are performed on the heterogeneous equipment operation signal flow to generate a standardized device status event flow. Protocol parsing identifies register read / write operations in the Modbus protocol and topic publication messages in the MQTT protocol. Signal normalization converts all parsed data into a sequence of JSON objects containing device identifiers, timestamps, operation types, and sets of operating parameters. The multi-dimensional environmental perception flow undergoes spatiotemporal alignment and data completion to generate a complete gridded environmental state field. Spatiotemporal alignment unifies sensor data from different sampling frequencies to the same time base, while data completion fills in missing values ​​caused by sensor failures or communication interruptions. The gridded environmental state field discretizes the exhibition space into a three-dimensional grid, with each grid storing environmental parameter values ​​corresponding to its location, including temperature, humidity, and light intensity. Visitor activity trajectories are cleaned and segmented into behavioral segments to generate behavioral sequences composed of continuous behavioral units. Trajectory cleaning filters out jump points and drift points in the positioning signals, and behavioral segmentation divides the trajectory into movement, stillness, and interactive behavioral units based on speed changes and dwell time.

[0027] Standardized device status event streams are mapped to corresponding virtual device nodes in the digital twin model of the exhibition hall, updating the real-time status attributes of the virtual device nodes. For example, the brightness parameter of a lighting device in a device status event is written into the current brightness attribute of the virtual lighting device node. A gridded environmental state field is overlaid onto the spatial grid of the digital twin model of the exhibition hall, updating the environmental state attributes of the virtual space. For example, the temperature value of each grid in the gridded environmental state field is assigned to the temperature attribute of the corresponding spatial grid in the digital twin model of the exhibition hall. Each behavioral unit in the behavioral sequence is bound to a virtual visitor agent, and the virtual visitor agent is driven to perform the corresponding behavior in the space of the digital twin model of the exhibition hall. The virtual spatial impact of the behavior is recorded. For example, the behavior unit of a virtual visitor agent stopping in front of an exhibit triggers a virtual sensor network to record an increase in the flow of people in that area. Through the event-driven engine, based on device status events, environmental status updates, and virtual visitor agent behavior, the virtual interactions and influences between these events are calculated, enabling the overall state of the digital twin model of the exhibition hall to dynamically evolve over time, forming a virtual exhibition hall operation mirror. The event-driven engine processes events in the event queue in chronological order and calls the exhibition hall environment physical rule simulation engine to update the grid state.

[0028] It is understandable that generating the gridded environmental state field involves calculating the completion of missing sensor data, which occurs during sensor coverage blind spots or communication failures. Optionally, for a grid in the gridded environmental state field that lacks a certain environmental parameter value, a weighted interpolation method based on spatial distance is used for completion. The completion formula is expressed as: in: Representing spatial coordinates The environmental parameter values ​​after mesh completion. Indicates the number of effective sensor grids around. Indicates the first Environmental parameter measurements of an effective sensor grid, Indicates the first Interpolation weights for each effective sensor grid, interpolation weights From grid to coordinates The Euclidean distance is inversely proportional to the square of the distance. In practice, the Euclidean distance is calculated through the difference in three-dimensional spatial coordinates. The effective sensor grid refers to the grid containing sensors that have reliable measurement values ​​at the same time. Data comparison shows that in the area of ​​the east corridor of the physics exhibition hall where temperature data is missing due to sensor failure, the completion formula generates a gridded environmental state field based on the temperature measurement values ​​of the effective sensor grids on the west and north sides. The average absolute error between the completed temperature distribution and the measured temperature data after subsequent sensor repair is less than 0.5 degrees Celsius. However, the missing areas that are not completed remain blank in the virtual environment, resulting in a discontinuous environmental state.

[0029] Understandably, the event-driven engine's computation process ensures the synchronization of virtual interactions and impacts. Optionally, the event-driven engine defines event types including device status events, environmental status update events, and virtual visitor agent behavior events. Each event contains a timestamp and event data. The event-driven engine processes events in ascending order of timestamps and executes the status update function. In specific implementations, the status update function calls the behavior rules of the virtual device agent model to calculate device status changes, calls the exhibition environment physical rule simulation engine to calculate environmental parameter diffusion, and calls the virtual visitor agent model to adjust behavioral logic. Data comparison shows that when the air conditioning equipment in the physical exhibition hall raises the set temperature at 2:00 PM, the heterogeneous equipment operation signal stream captures this event at 2:00:00 PM. The standardized equipment status event stream is generated and injected into the digital twin model of the exhibition hall within 100 milliseconds at 2:00:00 PM. The event-driven engine updates the virtual air conditioning equipment node status and triggers the exhibition hall environment physical rule simulation engine to recalculate the grid temperature distribution within 200 milliseconds at 2:00:00 PM. The virtual exhibition hall running image shows that the grid temperature in the corresponding area begins to rise at 2:00:01 PM, which is consistent with the temperature rise trend recorded by the actual temperature sensor in the physical exhibition hall. The time delay is controlled within the millisecond level.

[0030] In one embodiment of the present invention, see [reference] Figure 2When performing operational status slicing on the virtual exhibition hall's running image, each time slice contains a snapshot of the status of all devices, the environment, and visitors within its corresponding time period. A fixed time slice length and sliding step size are set, and fixed-length operational status intervals are continuously extracted from the timeline of the virtual exhibition hall's running image at intervals based on this sliding step size. For each extracted operational status interval, the state change sequences of all virtual device nodes, the environmental parameter change curves of all spatial grids, and the behavioral interaction records of all virtual visitor agents within that interval are extracted. The device state change sequences, environmental parameter change curves, and behavioral interaction records within this operational status interval are encapsulated into a structured data packet in chronological order, forming a time slice. In this structured data packet, the device state change sequences, environmental parameter change curves, and behavioral interaction records are indexed by a unified time axis to ensure that data from different dimensions at the same moment are aligned. All generated time slices are sorted according to their corresponding start timestamps to form this continuous time slice sequence. When performing situational feature distillation extraction for each time slice, state transition pattern mining is performed on the equipment state change sequence in the time slice to extract the pattern features of equipment operation, including periodic patterns, sudden patterns, and collaborative patterns between related equipment. Spatiotemporal correlation feature extraction is performed on the environmental parameter change curves in the time slice to extract the spatiotemporal features of environmental evolution, including the spatial distribution of parameter gradients, anomaly propagation paths, and the correlation strength with equipment location. Group behavior pattern clustering is performed on the behavioral interaction records in the time slice to extract the group features of visitor behavior, including gathering hotspots, movement patterns, and the frequency and type of interaction with equipment. A ternary association graph is constructed between equipment pattern features, environmental spatiotemporal features, and behavioral group features. Nodes in the graph represent different types of features, and edge weights represent the co-occurrence strength and causal correlation strength between features within the time slice. Graph embedding learning is performed on this ternary association graph to encode the association information of all nodes and edges in the graph into a fixed-dimensional low-dimensional vector. This low-dimensional vector is the situational feature vector, which contains the comprehensive association state of equipment, environment, and behavior within the time slice.

[0031] In practical implementation, when performing operational status slicing on the virtual exhibition hall's operational image, each time slice contains a snapshot of the status of all devices, the environment, and visitors within its corresponding time period. The example scenario processes the operational image of a medium-sized science and technology museum from 10:00 AM to 12:00 PM on June 15, 2024. A fixed time slice length and sliding step size are set. Using the sliding step size as the interval, fixed-length operational status intervals are continuously extracted from the timeline of the virtual exhibition hall's operational image. The time slice length is set to five minutes, and the sliding step size is set to one minute. The first extraction begins at 10:00 AM, capturing the operational status interval from 10:00 AM to 10:05 AM. The next extraction begins at 10:01 AM, capturing the operational status interval from 10:01 AM to 10:06 AM. For each captured operating state interval, extract the state change sequence of all virtual device nodes, the environmental parameter change curve of all spatial grids, and the behavioral interaction records of all virtual visitor agents within the operating state interval. In the operating state interval from 10:00:00 to 10:05:00, extract the sequence of the illuminance value of virtual lighting device node L001 linearly decreasing from 500 lux to 300 lux, extract the sequence of the supply air temperature of virtual air conditioning device node A003 fluctuating from 24 degrees Celsius to 25 degrees Celsius, extract the curve of the temperature of spatial grid G-201 rising from 23.5 degrees Celsius to 24.2 degrees Celsius, and extract the behavioral interaction record of virtual visitor agent P032 staying in front of exhibit E15 for four minutes and triggering two interactions. The sequence of device status changes, environmental parameter change curves, and behavioral interaction records within the operating state interval are encapsulated into a structured data packet in chronological order, forming a time slice. This structured data packet is stored in binary format and includes a unified timestamp index, an array of device status change sequences, a matrix of environmental parameter change curves, and a linked list of behavioral interaction records. Within the structured data packet, the device status change sequences, environmental parameter change curves, and behavioral interaction records are indexed using a unified timeline to ensure alignment of data from different dimensions at the same moment. The timeline uses milliseconds as the smallest unit. At 10:02:30, the index can simultaneously retrieve the illuminance value of lighting device node L001, the temperature value of grid G-201, and the interaction actions of visitor agent P032.

[0032] In the specific implementation, when performing situational feature distillation extraction for each time slice, state transition pattern mining is performed on the equipment state change sequence in the time slice to extract the pattern features of equipment operation. The pattern features include periodic patterns, sudden patterns, and collaborative patterns between related equipment. In the time slice with a start time of 10:30:00, state transition pattern mining identifies the periodic pattern of virtual projection equipment node PJ101 automatically calibrating once every two minutes, the sudden pattern of virtual ventilation equipment node VF205 suddenly increasing its speed within ten minutes, and the collaborative pattern between virtual lighting equipment node L012 reducing brightness and virtual curtain equipment node C008 closing. Spatiotemporal correlation features were extracted from the environmental parameter change curves in the time slices to extract the spatiotemporal features of environmental evolution. These features include the spatial distribution of parameter gradients, anomaly propagation paths, and the correlation strength with equipment locations. In the same time slice, the spatiotemporal correlation feature extraction calculated the spatial distribution of temperature gradients in the eastern area of ​​the exhibition hall being significantly higher than that in the western area. The path of a high-temperature anomaly point propagating from the air conditioning vent to the exhibition area was tracked, and the correlation strength between the local carbon dioxide concentration increase and the working location of the virtual fresh air equipment node FA301 was calculated.

[0033] Group behavior pattern clustering was performed on the behavioral interaction records in the time slices to extract the group characteristics of visitor behavior. The group characteristics include gathering hotspots, movement patterns, and the frequency and type of interaction with devices. In the same time slice, the group behavior pattern clustering found that visitors formed gathering hotspots in the No. 3 immersive experience area, and identified the unidirectional movement pattern of visitors from the entrance to the core exhibition area. The frequency of visitor interaction with virtual interactive device node ID005 was counted as 15 times per minute, mainly through touch interaction. A ternary association graph was constructed between device pattern characteristics, environmental spatiotemporal characteristics, and behavioral group characteristics. In the graph, nodes represent different types of features, and edge weights represent the co-occurrence strength and causal relationship strength between features within the time slice. The nodes include "periodic dimming of lighting", "high temperature gradient in the east area", and "crowd gathering in the immersive exhibition area". Between the two feature nodes "periodic dimming of lighting" and "high temperature gradient in the east area", the edge weights were calculated based on the correlation coefficient of their synchronous changes within the time window and the significance level of the Granger causality test. Graph embedding learning is performed on the ternary association graph, encoding the association information of all nodes and edges in the graph into a fixed-dimensional low-dimensional vector. The low-dimensional vector is the situation feature vector, which contains the comprehensive association state of devices, environment, and behavior within the time slice. The fixed dimension is set to 128 dimensions. The graph embedding learning adopts the Node2Vec algorithm based on random walk, mapping each feature node to a vector, and using the average of the vectors of all nodes in the graph as the situation feature vector of the time slice.

[0034] In some embodiments, state transition pattern mining uses a hidden Markov model to identify patterns in device state sequences, spatiotemporal correlation feature extraction uses the spatial autocorrelation Moran index to calculate the clustering of parameter gradients, group behavior pattern clustering uses the DBSCAN density clustering algorithm to discover clustering hotspots, the edge weights of the ternary association graph are calculated by the weighted sum of co-occurrence strength and causal association strength, and graph embedding learning uses a graph neural network model for encoding. In practical implementation, state transition pattern mining employs Hidden Markov Models (HMMs) to identify patterns in device state sequences. The HMM uses the state change sequences of virtual device nodes extracted within a time slice as the input observation sequence. By estimating model parameters, including initial state probabilities, state transition probabilities, and the observation probability matrix, it learns the hidden transition patterns between device states. After model training, the state sequences are decoded to identify recurring periodic patterns in device operation. These periodic patterns are characterized by states appearing cyclically at fixed time intervals, such as the brightness value of a lighting device node changing regularly during specific daily periods. Simultaneously, by analyzing anomalous jumps in the state transition probability matrix, sudden patterns are detected. These sudden patterns are characterized by drastic fluctuations in device states within a short period, such as a sudden increase in air supply temperature due to a sudden increase in load on an air conditioning device node. Furthermore, by comparing the decoding results of state sequences from multiple associated device nodes, collaborative patterns are discovered. These collaborative patterns are characterized by highly synchronized state transitions between different virtual device nodes, such as the simultaneous opening of a projection device node and the closing of a curtain device node.

[0035] It is understandable that calculating the edge weights of a ternary association graph requires quantifying the association relationships between feature nodes. Optionally, the edge weight calculation comprehensively considers both feature co-occurrence strength and causal association strength, and the edge weight calculation uses the following formula for quantification: in: Represents feature nodes With feature nodes Edge weights between them and These are the preset coefficients that balance the co-occurrence strength term and the causal association strength term. Indicates the total duration of the time slice Internal, characteristics With features The length of time that is active at the same time. This represents the characteristics calculated based on the Granger causality test. Features The causal influence strength value, Representation of features Features The causal influence strength value. In specific implementation, a preset coefficient is used. Set it to 0.6. Set to 0.4, total duration of time slice The duration is 300 seconds. An active state is defined as a feature value exceeding a preset threshold. The causal influence strength value is obtained by normalizing the Granger causality test F-statistic for the time series data. Data comparison shows that in the 10:30:00 time slice, the co-occurrence duration of the features "periodic calibration of projection equipment" and "periodic fluctuation of sound pressure level in the exhibition area" is... The causal influence strength value of feature m on feature n over a period of 280 seconds. The causal influence strength value of feature n on feature m is 0.85. The value is 0.10. Substituting this value into the formula, we obtain the edge weight. The value is 0.73, while another feature with weaker correlation, "constant lighting brightness" and "high corridor wind speed," has a co-occurrence time of only ten seconds. The causal influence strength values ​​are close to zero, and the calculated edge weights are... With a value of 0.02, the edge weight calculation result effectively distinguishes between feature pairs with strong and weak associations.

[0036] It is understandable that the graph embedding learning process encodes graph structural information into low-dimensional vectors. Optionally, the Node2Vec algorithm is used to perform graph embedding learning on ternary association graphs. This algorithm generates node sequences by performing biased random walks on the graph, and then inputs the node sequences into a Skip-gram model to learn the node vector representations. In specific implementations, the depth of the biased random walk is set to ten, the number of walk sequences is set to eighty, and the window size of the Skip-gram model is set to five. Finally, a thirty-two-dimensional vector is generated for each feature node. The arithmetic mean of all feature node vectors is calculated to obtain a one-hundred-twenty-eight-dimensional situation feature vector. The one-hundred-twenty-eight-dimensional vector is calculated by multiplying the thirty-two feature nodes by the dimension of each node vector, which is four dimensions. Data comparison shows that when ternary association graphs were constructed and graph embedding learning was performed on time slices reflecting the "normal and stable operation" state and time slices reflecting the "local abnormal congestion" state, the cosine similarity of the two situation feature vectors in the vector space was 0.15, showing a significant difference. However, the cosine similarity of the situation feature vectors of two time slices that belong to the "normal and stable operation" mode but are at different times was 0.89, indicating that the situation feature vectors generated by graph embedding learning can effectively characterize and distinguish different comprehensive operation states of the exhibition hall.

[0037] In one embodiment of the present invention, see [reference] Figure 3When constructing the exhibition hall's operational status evolution map, all time slices and their corresponding status feature vectors generated within the historical operation cycle are collected. Each status feature vector is used as a state node in the exhibition hall's operational status evolution map. The status evolution similarity between any two state nodes is calculated, based on the distance between the two status feature vectors in the feature space and the proximity of the time slices they represent on the time axis. In the exhibition hall's operational status evolution map, a directed edge is established for a pair of state nodes whose status evolution similarity exceeds a preset connection threshold. The direction of the edge is from the earlier node to the later node, and the weight of the edge is positively correlated with the status evolution similarity. Community detection is performed on the exhibition hall's operational status evolution map, clustering multiple state nodes with similar evolution patterns into the same community. Each community represents a typical exhibition hall operation pattern. A pattern label is extracted for each community, and transition conditions are labeled for the transition paths between communities, forming an exhibition hall's operational status evolution map with a hierarchical community structure.

[0038] In the specific implementation, when constructing the exhibition hall's operational status evolution map, all time slices generated within the historical operational period and their corresponding status feature vectors are collected. In the example scenario, the historical operational period is set to the entire month of June 2024, resulting in the collection of 86,400 time slices and their corresponding 128-dimensional status feature vectors. Each time slice is five minutes long. Each status feature vector serves as a state node in the exhibition hall's operational status evolution map, which contains a total of 86,400 state nodes. Each state node stores a 128-dimensional status feature vector and its corresponding time slice's start timestamp. The status evolution similarity between any two state nodes is calculated. This similarity is based on the distance between the two status feature vectors in the feature space and the proximity of the time slices they represent on the time axis. For state node A and state node B, the cosine similarity of their status feature vectors in the feature space is calculated, along with the time difference between the start timestamps of the two time slices. In the specific implementation, the status evolution similarity... It is calculated using a function that combines feature similarity and temporal proximity; the specific formula is as follows: in: This represents the similarity in situational evolution between state node A and state node B. This is a preset coefficient used to balance the weights of feature similarity and temporal proximity; its value is between 0 and 1. The situation feature vector of state node A The situation feature vector of state node B Cosine similarity between them It is the time decay coefficient. and These are the start timestamps of the time slices corresponding to state node A and state node B, respectively. This refers to the absolute time difference between the two. In the evolution map of the exhibition hall's operational status, a directed edge is established for a pair of state nodes whose status evolution similarity exceeds a preset connection threshold. The direction of the edge points from the earlier node in time to the later node in time, and the weight of the edge is positively correlated with the status evolution similarity. For example, if the preset connection threshold is set to 0.85, and the status evolution similarity between state node A and state node B is... The calculated value is 0.92. Since 0.92 is greater than 0.85, a directed edge is established in the exhibition hall's operational status evolution graph, pointing from node A to node B, and the weight of this edge is set to 0.92. Data comparison shows that the similarity between node A and the immediately following node B is 0.92, while the similarity with node C three hours later is only 0.15. Therefore, the exhibition hall's operational status evolution graph only establishes a directed edge between node A and node B, and does not establish a connection with node C. This reflects the graph's characteristic of capturing short-term state evolution.

[0039] In practice, community discovery is performed on the evolution map of the exhibition hall's operation status. Multiple state nodes with similar evolution patterns are clustered into the same community, with each community representing a typical exhibition hall operation mode. Community discovery employs the Louvain algorithm based on modularity optimization. Taking the weights of all state nodes and directed edges between them in the exhibition hall's operation status evolution map as input, the algorithm iteratively merges nodes into communities to maximize the modularity of the entire map. In the example, 86,400 state nodes are clustered into five main communities: Community 1 contains approximately 20,000 state nodes appearing during midday; Community 2 contains approximately 30,000 state nodes appearing during off-peak hours in the morning and evening; Community 3 contains approximately 15,000 state nodes appearing during large group visits; Community 4 contains approximately 10,000 state nodes appearing during nighttime cleaning; and Community 5 contains approximately 11,000 state nodes appearing during routine equipment maintenance or minor equipment malfunctions. A pattern label is extracted for each community, and transition conditions are labeled for the transition paths between communities, forming a hierarchical community structure of the exhibition hall's operation status evolution map. The pattern labels for communities are extracted based on the mean characteristics of the situational feature vectors of all state nodes within the community. For example, nodes in community 1 generally have characteristics of high equipment load, medium to high environmental parameters, and high visitor density, so the extracted pattern label is "midday high load operation mode". The transfer conditions for the transfer paths between communities are determined by analyzing the differences in features or context between the starting node and the target node on the edges between two communities connected by a large number of directed edges. For example, the transfer condition for the path from the "off-peak steady state mode" of community 2 to the "midday high load operation mode" of community 1 is labeled as "visitor entrance traffic exceeds two people per second for five consecutive minutes and the ambient temperature rise rate in the core exhibition area is greater than 0.1 degrees Celsius per minute".

[0040] Understandably, the community discovery process categorizes a large set of state nodes into meaningful groups of operational patterns. In its implementation, the edge weight matrix of the exhibition hall's operational status evolution graph is input into the Louvain algorithm. After multiple iterations, the algorithm converges, dividing the 86,400 nodes into five communities. This community partitioning increases the overall modularity of the graph from an initial 0.12 to 0.65, indicating that connections within communities are tight while connections between communities are relatively sparse. Data comparison shows that the average edge weight between nodes within community 1 is 0.88, while the average edge weight between nodes in community 1 and community 2 is 0.23, and the average edge weight between nodes in community 1 and community 5 is 0.10. This verifies the effectiveness of the community partitioning, which groups together nodes with highly similar evolutionary relationships.

[0041] It is understandable that extracting pattern labels requires summarizing the common characteristics of nodes within a community. In practice, the process of extracting pattern labels for a community involves calculating the mean of the situational feature vectors of all state nodes within each community to obtain a community center vector. Then, the 128 dimensions of the community center vector are ranked by feature importance, and the original features corresponding to the top five dimensions with the highest weights are selected for interpretation. For example, for the center vector of community 5, its top five high-weight dimensions correspond to features such as "high frequency of abnormal shutdowns of air conditioning equipment," "abnormal temperature gradients in local areas," "many visitor complaint interactions," "sudden increase in the load of fresh air equipment," and "frequent dimming attempts of lighting equipment." Combining these features, the pattern label "equipment abnormality and regulation disorder pattern" is extracted as the pattern label for community 5. Labeling the transfer conditions for the transfer paths between communities involves analyzing all directed edges originating from nodes in community X and pointing to nodes in community Y, and extracting the common feature changes of the starting nodes on these edges in the short period before the transfer occurs. For example, analyzing the starting nodes of all such transition edges from the "nighttime closed mode" of Community 4 to the "off-peak steady-state mode" of Community 2, we found that they generally have the characteristics of "all main lighting equipment status changing from closed to open" and "entrance access control equipment status changing from locked to unlocked". Therefore, the transition condition is marked as "main lighting system starts and main entrance is unlocked".

[0042] In one embodiment of the present invention, when mapping the real-time generated situation feature vector to the exhibition hall operation situation evolution map for node localization, the feature similarity between the real-time generated situation feature vector and each state node in the exhibition hall operation situation evolution map is calculated. The real-time situation feature vector is mapped to the state node with the highest feature similarity, which is taken as the current localized node in the map. Starting from the localized node, all outgoing edges connecting to the next-hop node are traversed, and the transition probability to each next-hop node is calculated based on the edge weight. Based on the transition probability, a random walk algorithm is used to perform multi-step state deduction starting from the localized node, generating multiple evolution paths starting from the current state. Each path consists of a series of state nodes connected in chronological order. The terminal nodes of each evolution path are analyzed to identify nodes belonging to abnormal operation mode communities or those highly similar to historical abnormal nodes, and these nodes are marked as potential abnormal nodes. The evolution paths containing these potential abnormal nodes are selected as the situation prediction paths that require key attention.

[0043] In practical implementation, when mapping the real-time generated situation feature vector to the exhibition hall operation status evolution map for node localization, the feature similarity between the real-time generated situation feature vector and each state node in the exhibition hall operation status evolution map is calculated. In the example scenario, the real-time generated situation feature vector corresponds to the time slice of 14:30:00 on July 1, 2024, and the exhibition hall operation status evolution map contains 86,400 state nodes constructed based on historical data from June. The cosine similarity between the real-time generated situation feature vector and the situation feature vector of each state node in the map is calculated, resulting in 86,400 similarity values. The real-time situation feature vector is mapped to the state node with the highest feature similarity, which is used as the current localization node in the map. The calculation shows that the real-time generated situation feature vector has the highest feature similarity with state node N72315 in the map, with a value of 0.96. Therefore, the real-time state is located at state node N72315. Data comparison shows that the real-time generated situation feature vector has a similarity of 0.88 with the second most similar node, and the similarity with the hundredth most similar node has dropped to 0.71, indicating that the location node N72315 has a significant similarity advantage.

[0044] Starting from the locating node, traverse all the next-hop nodes connected by its outgoing edges, and calculate the transition probability to each next-hop node based on the edge weights. Locating node N72315 has four directed outgoing edges in the exhibition hall's operational evolution graph, connecting to nodes N72316, N72320, N72345, and N72380, respectively. The weights of these four edges are 0.94, 0.91, 0.88, and 0.60, respectively. The transition probability is calculated by normalizing the weights of all outgoing edges originating from the same node, using the following formula: in: Indicates starting from the current location node Move to the next hop node The probability, Indicates from node Pointing to node The weight of the directed edge. Represents a node The set of next-hop nodes pointed to by all outgoing edges. Represents a node The sum of the weights of all outgoing edges. In the specific implementation, the sum of the weights of the four outgoing edges of the location node N72315 is 0.94 + 0.91 + 0.88 + 0.60 = 3.33. According to the formula, the probability of moving to node N72316 is 0.94 / 3.33 ≈ 0.282, the probability of moving to node N72320 is 0.91 / 3.33 ≈ 0.273, the probability of moving to node N72345 is 0.88 / 3.33 ≈ 0.264, and the probability of moving to node N72380 is 0.60 / 3.33 ≈ 0.181.

[0045] Based on transition probabilities, a random walk algorithm is used to perform multi-step state deduction starting from the current node, generating multiple evolution paths from the current state. Each path consists of a series of state nodes connected in chronological order. The deduction steps are set to five steps, predicting the state evolution of five future time slices. The random walk algorithm randomly selects the next node based on the transition probabilities calculated at each step. This process is repeated one thousand times, generating one thousand five-step evolution paths. It can be understood that due to the randomness of the transition probabilities, each walk generates different paths, but high-probability transition edges are selected more frequently. See Table 1.

[0046] Table 1: A five-step random walk deduced from node N72315 Deduction Steps Current node Selected next hop node Edge weights used in this transfer This transition probability Step 1 N72315 N72316 0.94 0.282 Step 2 N72316 N72401 0.95 0.310 Step 3 N72401 N72485 0.90 0.283 Step 4 N72485 N72560 0.93 0.295 Step 5 N72560 N72612 0.89 0.280 Analyzing the terminal nodes of each evolution path, nodes belonging to abnormal operation mode communities or highly similar to historical abnormal nodes were identified and marked as potential abnormal nodes. In the community structure of the exhibition hall operation status evolution map, community 5 was marked as the "equipment anomaly and regulation disorder mode" community. Analyzing the generated one thousand five-step deduction paths, the community affiliation of the terminal nodes of all paths was statistically analyzed. It was found that the terminal nodes of 920 paths belonged to community 2 or community 1, the terminal nodes of 70 paths belonged to community 3, but the terminal nodes of 10 paths belonged to community 5. These ten terminal nodes belonging to community 5, such as nodes N80125 and N80230, were marked as potential abnormal nodes.

[0047] Evolution paths containing potentially anomalous nodes were selected as key situation prediction paths requiring focused attention. Among the ten paths identified in the previous step whose endpoints were potentially anomalous nodes, one path sequence was N72315->N72316->N72401->N72485->N72560->N80125. This path shows that the system deduced that starting from the current state, after four "normal" evolutions, it entered node N80125, representing an anomalous state, in the fifth step. This complete path was selected as a key situation prediction path requiring focused attention. From the generated one thousand paths, ten such situation prediction paths were selected, all ending at a node in community 5. These paths revealed multiple trajectories from the current state to the anomalous state.

[0048] It is understandable that the number of random walks affects the comprehensiveness of path coverage and computational cost. Optionally, the number of random walks can be dynamically adjusted based on the availability of real-time computing resources and the required prediction confidence. In practical implementation, for scenarios with high real-time requirements, the number of random walks can be set to 500 to quickly generate predictions; for offline deep analysis scenarios, the number of random walks can be set to 10,000 to obtain a more stable path probability distribution. Data comparison shows that when 500 random walks are performed, six paths containing potentially anomalous nodes are selected; when 1,000 random walks are performed, ten paths containing potentially anomalous nodes are selected; and when 5,000 random walks are performed, fifty-two paths containing potentially anomalous nodes are selected, but forty-five of these paths are repetitions or minor variations of the first ten paths, and the main evolution pattern was already discovered in earlier iterations.

[0049] It is understandable that the accuracy of node localization directly affects the starting point of subsequent inferences. In practice, cosine similarity is used for feature similarity calculation because it is sensitive to directional differences in high-dimensional vectors. When the highest feature similarity is below a certain threshold, it indicates that the current real-time state has no highly similar historical states in the graph. At this time, node localization and subsequent inference will not be performed. Instead, a state anomaly warning will be triggered and the new state will be recorded for later use in graph updates.

[0050] See Figure 4This is a dual-bar chart comparing the edge weights and transition probabilities of the next-hop node from location node N72315, showcasing relevant data for the next-hop node of "location node N72315" in the intelligent exhibition hall situation simulation. Edge weights and transition probabilities are positively correlated; the higher the edge weight, the greater the transition probability. Node N72380 has the lowest edge weight (0.60) and transition probability (0.181), indicating that it represents the least likely evolution direction for this location node. This chart belongs to the analysis stage of the intelligent exhibition hall operation situation simulation, used to visually display the "probability distribution of the evolution of the current state node to other nodes," helping to identify high-priority subsequent states and serving as an auxiliary visualization tool for situation prediction and anomaly early warning.

[0051] In one embodiment of the present invention, when performing parameter intervention simulation in a digital twin model of an exhibition hall, a situation prediction path requiring focused attention is selected. Starting from the current virtual state, the exhibition hall's operation is simulated along this situation prediction path within the digital twin model. When the simulation reaches a potential anomaly node, it is paused, and the key equipment parameters and environmental conditions that caused the operating state to enter this potential anomaly node are analyzed. In the digital twin model, multiple adjustment schemes are preset for these key equipment parameters, and the corresponding parameters of the virtual equipment nodes are modified. The simulation is restarted from the state after parameter modification, and it is observed whether subsequent operation in the digital twin model can avoid the potential anomaly node or transition to a better operating mode. The effectiveness of each adjustment scheme in avoiding anomalies or transitioning to a better mode, as well as the equipment control costs required for implementing the schemes, are evaluated. When evaluating the effectiveness and control costs of each adjustment scheme, after executing the adjustment scheme in the digital twin model, the simulation continues for a period of time, and the situation feature vector sequence generated by the simulation is extracted. Map the sequence of situational feature vectors back to the evolution map of the exhibition hall's operational status, and observe whether its evolution path completely avoids the potential abnormal node, or whether its final state node belongs to a better operational mode community. If the abnormal node is successfully avoided, record the minimum map distance between it and the abnormal node as a safety margin; if it transitions to a better community, record the performance improvement index of the new community mode compared to the original path's final community mode. Summarize the safety margin and performance improvement index, and calculate the effectiveness score of the adjustment plan using a preset utility function. Analyze the range and frequency of equipment parameter changes required by the adjustment plan, and calculate the additional energy consumption estimate and equipment wear estimate caused by implementing the adjustment plan by combining the equipment energy consumption model and the equipment loss model. Quantify the additional energy consumption estimate and equipment wear estimate into a control cost score using a preset cost function. Input the effectiveness score and the control cost score into a multi-objective decision model to obtain a comprehensive evaluation result of the adjustment plan. Combining effectiveness and control cost, select the optimal adjustment plan, and transform the equipment parameters to be adjusted and their target values ​​in the adjustment plan into specific, executable equipment control instructions.

[0052] The system iterates through all situational prediction paths requiring close monitoring, repeating the above steps for each path. All control commands generated by these paths are then aggregated to obtain the preventative control command set. The command issuance and closed-loop verification module performs executability verification and conflict detection on each command in this preventative control command set, ensuring that the devices involved in the command are online and in a controllable state, and that there are no resource or timing conflicts between commands. Verified commands are sorted according to their preset execution time windows and encapsulated into data frame formats recognized by the device control protocol. When the preset trigger conditions or execution time of the command are met, the corresponding control command data frame is sent to the target device control terminal via the IoT communication link. The system receives command response signals and device status feedback from the device control terminal in real time to confirm the command execution result. The command execution result and the post-execution device status feedback are synchronously injected into the digital twin model of the exhibition hall to update the virtual exhibition hall's operating image and trigger a new round of situational awareness, prediction, and control simulation cycle.

[0053] In practical implementation, parameter intervention simulation is performed in the digital twin model of the exhibition hall to generate a set of preventive control instructions for the physical exhibition hall equipment. This includes the following steps: Select a situation prediction path that requires key attention. In the digital twin model of the exhibition hall, starting from the current virtual state, simulate the operation of the exhibition hall along the situation prediction path. In the example scenario, the selected situation prediction path is N72315->N72316->N72401->N72485->N72560->N80125, where the end node N80125 is the identified potential abnormal node. In the digital twin model of the exhibition hall, starting from the virtual state corresponding to the positioning node N72315, the simulation model is driven to evolve according to the sequence of equipment, environment, and behavioral characteristics represented by each state node on the path. When the simulation reaches a potential abnormal node, the simulation is paused to analyze the key equipment parameters and environmental conditions that caused the operation to enter the potential abnormal node. When the simulation reaches the state corresponding to node N80125, the simulation is paused. By backtracking the logs of the virtual device agent model and the intermediate calculation state of the simulation engine of the exhibition hall environment physical rules, it is found that the key equipment parameter is that the condenser inlet water temperature setpoint of the No. 3 unit of the central air conditioning system deviates from the normal range by 5 degrees Celsius, and the key environmental condition is that the average ambient temperature of the east exhibition area has exceeded 28 degrees Celsius.

[0054] In the digital twin model of the exhibition hall, multiple adjustment schemes were preset for key equipment parameters, and the corresponding parameters of the virtual equipment nodes were modified. For the key parameter of the condenser inlet water temperature setpoint of Unit 3 of the central air conditioning system being too high, three adjustment schemes were preset: Scheme A lowers the setpoint by three degrees Celsius, Scheme B lowers it by five degrees Celsius, and Scheme C lowers it by seven degrees Celsius. In the digital twin model of the exhibition hall, the "condenser inlet water temperature setpoint" attribute of the virtual equipment node CAU-03 was modified, changing it from the current abnormal value to the target value corresponding to Scheme A, Scheme B, and Scheme C. The simulation was restarted from the modified state to observe whether subsequent operation in the digital twin model of the exhibition hall could avoid potential abnormal nodes or transition to a better operating mode. For each adjustment scheme, starting from the modified virtual state, the simulation of the digital twin model of the exhibition hall continued to run for the next six time slices, and the mapping of the subsequent situation evolution path generated by the simulation to the exhibition hall's operational situation evolution map was observed.

[0055] The effectiveness of each adjustment scheme in avoiding anomalies or transitioning to a better mode, as well as the equipment control costs required for implementation, are evaluated. After implementing the adjustment scheme in the digital twin model of the exhibition hall, the operation continues to be simulated for a period of time, and the situation feature vector sequence generated by the simulation is extracted. The situation feature vector sequence is mapped back to the evolution map of the exhibition hall's operation situation to observe whether its evolution path completely avoids potential anomaly nodes, or whether its final state node belongs to a better operation mode community. If anomaly nodes are successfully avoided, the minimum map distance between it and the anomaly node is recorded as a safety margin; if a better community is transitioned to, the performance improvement index of the new community mode compared to the original path's end community mode is recorded. The safety margin and performance improvement index are summarized, and the effectiveness score of the adjustment scheme is calculated using a preset utility function. The range and frequency of equipment parameter changes required by the adjustment scheme are analyzed, and the additional energy consumption estimate and equipment wear estimate caused by implementing the adjustment scheme are calculated by combining the equipment energy consumption model and the equipment wear model. The additional energy consumption estimate and equipment wear estimate are quantified into a control cost score using a preset cost function. The calculation of the effectiveness score and the control cost score is integrated using a multi-attribute decision formula: in: Indicating the adjustment plan The comprehensive evaluation result score indicates that the higher the score, the better the overall performance of the solution. It is a preset coefficient that balances the weights of effectiveness and cost; Indicating the adjustment plan The effectiveness score; and These represent the maximum and minimum effectiveness scores among all candidate adjustment schemes, respectively. Indicates the adjustment plan The cost-of-control rating; and These represent the maximum and minimum control cost scores among all candidate adjustment schemes, respectively. The effectiveness score and control cost score are input into a multi-objective decision model to obtain a comprehensive evaluation result of the adjustment schemes.

[0056] In practical implementation, preset coefficients The value is set to 0.6, giving it a higher weight in the effectiveness score. For adjustment scheme A, after simulation, the final node of its evolution path belongs to community 2, successfully avoiding the potentially anomalous node community 5. Its minimum graph distance to node N80125 is four hops, indicating a high safety margin and a high effectiveness score. The score is calculated to be 85 points, but it requires a significant adjustment to the condenser temperature, which is estimated to cause an 8% increase in energy consumption and a 0.05 increase in the equipment wear coefficient for the No. 3 central air conditioning unit in the next hour. The control cost score is... The score is 40. For adjustment scheme B, after simulation, the terminal node of its evolution path also belongs to community 2, and the safety margin is the same as that of scheme A. The effectiveness score is... The score is 85 points. The temperature adjustment range is relatively small, the estimated energy consumption increase is 4%, the equipment wear coefficient increases by 0.02, and the control cost score is... The score is 20 points. For adjustment scheme C, after simulation, the final node of its evolution path belongs to community 4. Although it avoids community 5, it enters a non-current time period operation mode, resulting in an effectiveness score. It scores 70 points, with the smallest temperature adjustment range, an estimated 2% increase in energy consumption, and a 0.01 increase in equipment wear coefficient. The control cost score is... It is worth 10 points. Assume... , , , Substitute the values ​​into the formula to calculate and adjust the comprehensive evaluation score of scheme A. Adjust the comprehensive evaluation score of Plan B. Adjust the comprehensive evaluation score of scheme C. Considering both effectiveness and control costs, the optimal adjustment scheme is selected, and the equipment parameters to be adjusted and their target values ​​are transformed into specific, executable equipment control instructions. Based on the calculation results, adjustment scheme B has the highest comprehensive evaluation score of 0.867, therefore, adjustment scheme B is selected as the optimal scheme. The equipment parameter to be adjusted in adjustment scheme B, "the setpoint of the condenser inlet water temperature of the central air conditioning system unit 3," and its target value are transformed into a specific equipment control instruction. The instruction content is to write the setpoint temperature target value to register 40105 at device address 0x03 via the Modbus RTU protocol. All situation prediction paths that require key attention are traversed, and the above steps are repeated for each path. The control instructions generated by all paths are summarized to obtain a set of preventive control instructions.

[0057] The instruction issuance and closed-loop verification module performs executability verification and conflict detection on each instruction in the preventive control instruction set to ensure that the devices involved in the instruction are online and in a controllable state, and that there are no resource or timing conflicts between instructions. In specific implementation, executability verification is completed by querying the real-time status table of the IoT device management platform to confirm that the current status of Unit 3 of the central air conditioning system is "running" and the communication status is "online". Conflict detection is performed by checking the instruction execution time window and resource usage to confirm that there are no other scheduled instructions that need to write the same parameter to the same device within the target time period. The verified instructions are sorted according to their preset execution time windows and encapsulated into a data frame format recognized by the device control protocol. The temperature adjustment instruction for Unit 3 of the central air conditioning system is preset to be executed "immediately" within the execution time window. The instruction is encapsulated into a data frame conforming to the Modbus RTU protocol, including slave address, function code, register address, data, and CRC checksum. When the preset trigger condition or execution time of the instruction is reached, the corresponding control instruction data frame is sent to the target device control terminal through the IoT communication link.

[0058] When the system clock reaches the instruction execution time, the instruction issuance and closed-loop verification module sends the encapsulated Modbus RTU data frame to the PLC controller of Unit 3 of the central air conditioning system via the RS-485 bus. It receives instruction response signals and equipment status feedback from the equipment control terminal in real time to confirm the instruction execution result. After receiving the instruction, the PLC controller returns an acknowledgment frame and, upon completion, sends back a status frame containing the new setpoint through the same communication link. The instruction execution result and the post-execution equipment status feedback are synchronously injected into the digital twin model of the exhibition hall to update the virtual exhibition hall's operational image and trigger a new round of situational awareness, prediction, and control simulation. The instruction issuance and closed-loop verification module injects the received "execution successful" response and new setpoint feedback into the digital twin model of the exhibition hall. The parameters of the virtual equipment node CAU-03 are updated. This update event drives the evolution of the virtual exhibition hall's operational image and serves as a new input to initiate a new round of processing from the operational situation slice.

[0059] It is understandable that the effectiveness evaluation of the adjustment plan relies on the simulation fidelity of the digital twin model of the exhibition hall. In practice, the utility function calculation of the effectiveness score will comprehensively consider multiple dimensions such as safety margin and performance improvement indicators. The safety margin is measured by spectral distance, and the performance improvement indicators include the estimated percentage reduction in energy consumption and the visitor comfort improvement index. In practice, the equipment energy consumption model estimates energy consumption based on equipment power, operating load, and operating time, while the equipment wear model estimates wear based on the number of equipment start-ups and shutdowns, parameter adjustment ranges, and equipment lifespan curves. Optionally, for complex multi-device coordinated control, the control cost score should also consider the additional costs brought about by the coupled operations between devices.

[0060] See Figure 5 This is a diagram illustrating the evolution path of a smart exhibition hall's operation in a two-dimensional feature space. In the scenario of smart exhibition hall situation simulation, this diagram serves as visual evidence of the "state transition logic," helping technical personnel explain to business stakeholders "why the current state evolves into a certain subsequent state," thus improving the credibility of the simulation results. This diagram represents a dimensionality reduction visualization of high-dimensional feature vectors, used to intuitively display the "evolution trajectory of the exhibition hall's operational situation characteristics in two-dimensional space." It helps to quickly identify the feature distribution and evolution direction of state nodes, and is one of the commonly used methods for visualizing situation simulation paths. It can assist in analyzing the trend of state transitions and the feature deviations of abnormal nodes.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0062] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart exhibition hall control system based on the Internet of Things, characterized in that, The system includes: The digital twin synchronization module is used to construct a digital twin model of the exhibition hall and collect the physical space operation flow of the target exhibition hall in real time through the Internet of Things sensor network. The physical space operation flow is synchronously injected into the digital twin model of the exhibition hall to form a dynamically updated virtual exhibition hall operation image. The operational status slicing module is used to perform operational status slicing processing on the virtual exhibition hall's operational image, generate a continuous time slice sequence, and perform status feature distillation extraction on each time slice to obtain a status feature vector that reflects the multi-dimensional operational correlation features within the time slice. The situation evolution map construction module is used to learn the evolution pattern of situation feature vector sequences based on historical operation data and construct the situation evolution map of the exhibition hall. The situation simulation and prediction module is used to map the real-time generated situation feature vector to the exhibition hall operation situation evolution map for node positioning, and perform multi-hop state simulation along the map edge to generate situation prediction paths and potential abnormal nodes for multiple future moments. The preventive control simulation module is used to perform parameter intervention simulation in the digital twin model of the exhibition hall based on the predicted situation path and potential abnormal nodes, and generate a set of preventive control instructions for the physical exhibition hall equipment.

2. The IoT-based intelligent exhibition hall control system according to claim 1, characterized in that, The process of synchronously injecting the physical space operation flow into the digital twin model of the exhibition hall to form a dynamically updated virtual exhibition hall operation mirror includes: The digital twin model of the exhibition hall is used to mirror the physical equipment layout, spatial topology, and environmental sensor network of the target exhibition hall; The physical space operation flow includes heterogeneous device operation signal flow, multi-dimensional environmental perception flow, and visitor activity trajectory flow; The heterogeneous device operation signal stream is parsed and normalized to generate a standardized device status event stream, wherein each device status event includes a device identifier, timestamp, operation type and set of operating parameters. The multidimensional environmental perception stream is spatiotemporally aligned and data is completed to generate a complete gridded environmental state field. The gridded environmental state field discretizes the exhibition space into a three-dimensional grid, and each grid stores the environmental parameter value corresponding to the location of the grid. The visitor activity trajectory flow is cleaned and segmented into behavior segments to generate a behavior sequence composed of continuous behavior units; The standardized device status event stream is mapped to the corresponding virtual device nodes in the exhibition hall digital twin model, and the real-time status attributes of the virtual device nodes are updated. The gridded environmental state field is superimposed onto the spatial grid of the digital twin model of the exhibition hall to update the environmental state attributes of the virtual space; Each behavioral unit in the behavioral sequence is bound to a virtual visitor agent, and the virtual visitor agent is driven to perform the corresponding behavior in the space of the digital twin model of the exhibition hall, and the virtual space impact of the behavior is recorded. Through an event-driven engine, based on device status events, environmental status updates, and virtual visitor agent behavior, the virtual interactions and influences between these events are calculated, enabling the overall state of the digital twin model of the exhibition hall to dynamically evolve over time, forming a virtual exhibition hall operation mirror.

3. The IoT-based intelligent exhibition hall control system according to claim 1, characterized in that, The step of performing operational status slicing processing on the virtual exhibition hall's running image to generate a continuous time slice sequence includes: Each time slice contains a snapshot of all equipment, environment, and visitor status within its corresponding time period; Set a fixed time slice length and sliding step size, and continuously extract a fixed duration of the running state interval on the timeline of the virtual exhibition hall running mirror with the sliding step size as the interval; For each captured running state interval, extract the state change sequence of all virtual device nodes, the environmental parameter change curve of all spatial grids, and the behavioral interaction records of all virtual visitor agents within the running state interval; The sequence of device state changes, the curve of environmental parameter changes, and the records of behavioral interactions within the operating state interval are encapsulated into a structured data packet in chronological order to form a time slice. In the structured data packet, the device state change sequence, environmental parameter change curve and behavior interaction record are indexed by a unified time axis to ensure that data from different dimensions at the same moment can be aligned. All generated time slices are sorted according to their corresponding start timestamps to form the continuous time slice sequence.

4. The IoT-based intelligent exhibition hall control system according to claim 1, characterized in that, The process of performing situation feature distillation extraction on each time slice yields a situation feature vector reflecting the multi-dimensional operational correlation features within the time slice, including: State transition pattern mining is performed on the equipment state change sequence in the time slice to extract the pattern features of equipment operation. The pattern features include periodic patterns, burst patterns and collaborative patterns between related equipment. Spatiotemporal correlation features are extracted from the environmental parameter change curves in the time slice to extract the spatiotemporal features of environmental evolution. The spatiotemporal features include the spatial distribution of parameter gradients, anomaly propagation paths, and the correlation strength with device location. Clustering of behavioral interaction records in time slices is performed to extract group behavior patterns and group characteristics of visitor behavior. These group characteristics include gathering hotspots, movement patterns, and the frequency and type of interaction with devices. Construct a ternary association graph among device pattern features, environmental spatiotemporal features, and behavioral group features. In the graph, nodes represent different types of features, and edge weights represent the co-occurrence strength and causal association strength among features within the time slice. Graph embedding learning is performed on the ternary association graph to encode the association information of all nodes and edges in the graph into a low-dimensional vector of fixed dimension. The low-dimensional vector is the situation feature vector, which contains the comprehensive association state of the device, environment and behavior in the time slice.

5. The IoT-based intelligent exhibition hall control system according to claim 4, characterized in that, The construction of the exhibition hall's operational evolution map includes: Collect all time slices generated within the historical operating cycle and their corresponding situation feature vectors; Each situation feature vector is used as a state node in the evolution map of the exhibition hall's operational status. Calculate the situation evolution similarity between any two state nodes. The situation evolution similarity is calculated based on the distance between the two situation feature vectors in the feature space and the proximity of the time slices they represent on the time axis. In the evolution map of the exhibition hall's operational status, a directed edge is established for a pair of state nodes whose status evolution similarity exceeds a preset connection threshold. The direction of the edge points from the earlier node in time to the later node in time, and the weight of the edge is positively correlated with the status evolution similarity. Community discovery is performed on the evolution map of the exhibition hall's operation status, and multiple state nodes with similar evolution patterns are clustered into the same community, with each community representing a typical exhibition hall operation mode. A pattern label is extracted for each community, and transfer conditions are marked for the transfer paths between communities, forming an evolution map of the exhibition hall's operational status with a hierarchical community structure.

6. The IoT-based intelligent exhibition hall control system according to claim 1, characterized in that, The process involves mapping the real-time generated situation feature vectors to the exhibition hall's operational situation evolution map for node localization, and performing multi-hop state deduction along the map edges to generate situation prediction paths and potential abnormal nodes for multiple future time points, including: Calculate the feature similarity between the real-time generated situation feature vector and the features of each state node in the exhibition hall operation situation evolution map; The real-time situation feature vector is mapped to the state node with the highest feature similarity, which serves as the current location node in the map. Starting from the positioning node, traverse all the next-hop nodes connected by its outgoing edges, and calculate the transition probability to each next-hop node based on the weight of the edge. Based on the aforementioned transition probabilities, a random walk algorithm is used to perform multi-step state deduction starting from the location node, generating multiple possible evolution paths starting from the current state. Each path consists of a series of state nodes connected in chronological order. Analyze the terminal nodes of each evolution path, identify nodes that belong to abnormal operation mode communities or are highly similar to historical abnormal nodes, and mark these nodes as potential abnormal nodes. The evolution paths containing the potential abnormal nodes are selected as the situation prediction paths that require special attention.

7. The IoT-based intelligent exhibition hall control system according to claim 6, characterized in that, The step of performing parameter intervention simulation in the digital twin model of the exhibition hall to generate a set of preventative control instructions for the physical exhibition hall equipment includes the following steps: S1: Select a situation prediction path that needs to be focused on, and simulate the operation of the exhibition hall along the situation prediction path in the digital twin model of the exhibition hall, starting from the current virtual state. S2: When the simulation reaches the potential abnormal node, pause the simulation and analyze the key equipment parameters and environmental conditions that caused the operating state to enter the potential abnormal node; S3: In the digital twin model of the exhibition hall, multiple adjustment schemes are preset for the key equipment parameters, and the corresponding parameters of the virtual equipment nodes are modified; S4: Restart the simulation from the state after modifying the parameters, and observe whether the subsequent operation in the digital twin model of the exhibition hall can avoid the potential abnormal nodes or switch to a better operating mode. S5: Evaluate the effectiveness of each adjustment plan in avoiding anomalies or transitioning to a better mode, as well as the equipment adjustment costs required to implement the plan; S6: Considering both effectiveness and control costs, select the optimal adjustment scheme and convert the equipment parameters to be adjusted and their target values ​​in the adjustment scheme into specific, executable equipment control instructions; Traverse all the situation prediction paths that require key attention, repeat steps S1 to S6 for each path, and summarize the control commands generated by all paths to obtain the set of preventive control commands.

8. The IoT-based intelligent exhibition hall control system according to claim 7, characterized in that, The evaluation of the effectiveness of each adjustment scheme in avoiding anomalies or transitioning to a better mode, and the equipment adjustment costs required for implementing the scheme, includes: After implementing the adjustment plan in the digital twin model of the exhibition hall, the simulation continues for a period of time in the future, and the sequence of situation feature vectors generated by the simulation is extracted. Map the situation feature vector sequence back to the exhibition hall operation situation evolution map, and observe whether its evolution path completely avoids the potential abnormal nodes, or whether its final state node belongs to a better operation mode community. If the abnormal node is successfully avoided, record the minimum graph distance between it and the abnormal node as a safety margin; if it is switched to a better community, record the performance improvement index of the new community mode compared with the original path end community mode. The safety margin and performance improvement indicators are summarized, and the effectiveness score of the adjustment scheme is calculated through a preset utility function; The range and frequency of equipment parameter changes required by the adjustment plan are analyzed. Combined with the equipment energy consumption model and equipment wear model, the additional energy consumption and equipment wear caused by implementing the adjustment plan are estimated. The estimated additional energy consumption and equipment wear are quantified into a control cost score using a preset cost function; The effectiveness score and the regulation cost score are input into a multi-objective decision model to obtain a comprehensive evaluation result of the adjustment scheme.

9. The IoT-based intelligent exhibition hall control system according to claim 1, characterized in that, The construction of the digital twin model of the exhibition hall includes: Collect architectural information model data, equipment asset list and physical topology connection relationship of the target physical exhibition hall, and construct a static three-dimensional geometric model of equipment-space association; Based on the equipment asset list, a virtual device agent model is constructed for each type of physical equipment, which includes multi-level attributes and behavioral rules. The multi-level attributes include at least equipment identification, model specifications, performance parameters and communication protocols. The behavioral rules define the dynamic response relationship of equipment state as control commands, operating parameters and external environment change. In the static three-dimensional geometric model, based on the physical topology connection relationship and the actual spatial coordinates of the devices, the corresponding virtual device proxy model is instantiated and deployed to form a virtual device node network; An integrated simulation engine for physical rules of the exhibition hall environment is used to simulate the propagation and interaction of light, temperature, humidity and air flow in a multi-dimensional spatial grid. A virtual sensor network and a virtual actuator network are embedded in the static three-dimensional geometric model. The virtual sensor network is used to perceive the state of the mesh environment calculated by the environmental physical rules simulation engine during the simulation. The virtual actuator network is used to receive control commands and drive the corresponding virtual device agent model to perform actions. Establish data interaction and causal logic connections between the virtual device node network, virtual sensor network, virtual actuator network and environmental physical rule simulation engine to form a high-fidelity simulation system that can dynamically respond to and evolve the state of equipment control commands, environmental disturbances and visitor interactions, namely the digital twin model of the exhibition hall.

10. The IoT-based intelligent exhibition hall control system according to claim 1, characterized in that, The system also includes: a command issuance and closed-loop verification module, comprising: Each instruction in the preventive control instruction set is subject to executability verification and conflict detection to ensure that the equipment involved in the instruction is online and its status is controllable, and that there are no resource or timing conflicts between the instructions. The verified instructions are sorted according to their preset execution time windows and encapsulated into a data frame format recognized by the device control protocol. When the preset triggering conditions or execution time of the instruction are met, the corresponding control instruction data frame is sent to the target device control terminal through the Internet of Things communication link; Receive command response signals and equipment status feedback returned by the equipment control terminal in real time to confirm the command execution results; The execution results of the instructions and the feedback of the device status after execution are synchronously injected into the digital twin model of the exhibition hall to update the virtual exhibition hall operation image and trigger a new round of situational awareness, prediction and control simulation cycle.

Images