Intelligent exhibition hall control method and system based on digital twinning

Abstract

The invention discloses an intelligent exhibition hall control method and system based on digital twinning, and relates to the field of exhibition hall control, and the method comprises the steps: carrying out the kinematics state calculation of a continuous position flow of a visitor, and precisely mastering the current motion state of the visitor; the key point is that the method is not limited to the current state, but carries out high-precision prediction on the future (moment) movement track of the visitor based on a kinematic model. And along the prediction track, a personalized dynamic environment bubble can be generated in advance in combination with the portrait of the visitor, and the environment atmosphere of the area where the visitor is about to enter is actively and progressively pre-rendered. Finally, the environmental bubbles are reversely analyzed into specific equipment control flows, so that pre-smooth adjustment of lamplight, sound effect and other equipment is realized, and the technical problems of abrupt control and poor experience caused by perception lag and lack of predictive ability of a traditional control scheme are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of exhibition hall control, and more specifically, to a smart exhibition hall control method and system based on digital twinning. BACKGROUND

[0002] With the rapid development of information technology and the Internet of Things, modern smart exhibition halls have gone beyond static display of exhibits and one-way information infusion, and instead pursue a highly personalized, deeply immersive, and seamless interactive visiting experience for visitors. In order to achieve this goal, the environmental control system of the exhibition hall needs to upgrade from a passive event response mode to an intelligent mode that can actively perceive, predict visitor intentions, and prepare the environment in advance. Digital twinning technology provides an ideal technical framework for realizing such proactive intelligent control by constructing a real-time, high-fidelity dynamic model of the physical exhibition hall in a virtual space, making fine-grained modeling of visitor behavior and coordinated control of the environment possible.

[0003] However, the current mainstream smart exhibition hall control scheme, even if it has initially introduced digital management, still mostly remains at the level of instantaneous-passive control based on infrared sensing, video monitoring, and region triggering. This control method has inherent perceptual lag, and the system always reacts after the visitor has entered or triggered a certain preset region, resulting in abrupt and lagging switching of lighting, sound effects, and multimedia content. This after-the-fact interactive experience brings a sense of passivity and fragmentation to the visitors, especially in scenes that pursue artistic atmosphere and immersion, abrupt environmental changes can severely damage the overall visiting experience, making it impossible to achieve the ideal effect of silently and subtly influencing the environment. The fundamental reason for this poor experience is that existing technical solutions lack the ability to understand and predict visitor processes and trends. Traditional control systems perceive isolated, discrete position events, rather than a continuous dynamic process. Since it is impossible to continuously model and prospectively predict the visitor's motion trajectory, speed, and potential intentions, the system can only passively execute the preset script when the visitor hits the next detection point, and cannot pre-render the environment and atmosphere of the area the visitor is about to reach.

[0004] Therefore, this simple logic model based on discrete threshold judgment fundamentally limits the evolution of the exhibition hall control system to a higher, more humanized active service, and cannot eliminate the lag and abruptness of control behavior. SUMMARY

[0005] To solve the above fundamental problem, according to an aspect of the present application, a smart exhibition hall control method based on digital twinning is provided, which includes: obtaining a target visitor original position stream; The kinematic state is calculated and the twin state is updated on the original location flow of the target visitor to obtain the visitor kinematic state at time T; By combining the exhibition hall layout model, trajectory prediction based on the kinematic model and spatial constraints is performed on the visitor's kinematic state at time T to obtain... Predicted trajectory at any given moment; Based on visitor profiles and scenario script libraries, along The predicted trajectory at any given time generates a dynamic environment bubble; Based on the device layout model, the device control weights of the dynamic environment bubble are reverse-analyzed to obtain the device control flow.

[0006] According to another aspect of this application, a smart exhibition hall control system based on digital twins is provided, comprising: The original location stream acquisition module is used to acquire the original location stream of the target visitor; The kinematic state analysis module is used to perform kinematic state calculation and twin state update on the original location flow of the target visitor to obtain the visitor kinematic state at time T. The trajectory prediction module, combined with the exhibition hall layout model, predicts the visitor's trajectory at time T based on the kinematic model and spatial constraints. Predicted trajectory at any given moment; The dynamic environment bubble generation module is used to generate bubbles based on visitor profiles and scene script libraries, following... The predicted trajectory at any given time generates a dynamic environment bubble; The device control flow generation module is used to perform reverse parsing of device control weights on the dynamic environment bubble based on the device layout model to obtain the device control flow.

[0007] Compared with existing technologies, this application provides a smart exhibition hall control method and system based on digital twins, which constructs a new paradigm for smart exhibition hall control from passive response to active prediction. It first accurately grasps the current motion state of visitors by performing kinematic state calculations on the continuous position flow of visitors. Crucially, this application is not limited to the current state, but rather uses a kinematic model to predict the future motion state of visitors. The system accurately predicts the movement trajectory of visitors at any given moment. Following this predicted trajectory, a personalized dynamic environment bubble can be generated in advance, based on visitor profiles, proactively and progressively pre-rendering the atmosphere of the area the visitor is about to enter. Finally, by inversely parsing the environment bubble into specific device control flows, it enables the pre- and smooth adjustment of lighting, sound effects, and other equipment, thus solving the technical problems of abrupt control and poor user experience caused by perception lag and lack of predictive capabilities in traditional control schemes. Attached Figure Description

[0008] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which:

[0009] Figure 1 Flow chart of the method for controlling the smart exhibition hall based on digital twinning according to the embodiment of the present application.

[0010] Figure 2 Data flow diagram of the method for controlling the smart exhibition hall based on digital twinning according to the embodiment of the present application.

[0011] Figure 3 Flow chart of step S2 in the method for controlling the smart exhibition hall based on digital twinning according to the embodiment of the present application.

[0012] Figure 4 Flow chart of step S5 in the method for controlling the smart exhibition hall based on digital twinning according to the embodiment of the present application.

[0013] Figure 5 Block diagram of the system for controlling the smart exhibition hall based on digital twinning according to the embodiment of the present application. DETAILED DESCRIPTION

[0014] Embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. It should be understood that the accompanying drawings and embodiments of the present disclosure are only for exemplary purposes, and are not intended to limit the scope of protection of the present disclosure.

[0015] Based on this, the present application proposes a method for controlling a smart exhibition hall based on digital twinning. Figure 1 Flow chart of the method for controlling the smart exhibition hall based on digital twinning according to the embodiment of the present application. Figure 2 Data flow diagram of the method for controlling the smart exhibition hall based on digital twinning according to the embodiment of the present application. As shown in Figure 1 and Figure 2 The method for controlling the smart exhibition hall based on digital twinning according to the embodiment of the present application includes: S1, obtaining a target visitor original position stream; S2, performing kinematic state calculation and twin state updating on the target visitor original position stream to obtain a T time visitor kinematic state; S3, combining an exhibition hall layout model, performing trajectory prediction based on a kinematic model and spatial constraints on the T time visitor kinematic state to obtain a predicted trajectory at a T+1 time; S4, based on a visitor portrait and a scene script library, generating a dynamic environment bubble along the predicted trajectory at the T+1 time; S5, based on a device layout model, performing reverse analysis of device control weights on the dynamic environment bubble to obtain a device control stream.

[0016] ​​In step S1, the original position stream of the target visitor is acquired. It should be understood that in order to overcome the control lag and abruptness caused by triggering based on discrete and isolated event points in the prior art, and to accurately predict the visitor's behavior process and future trend, the present application first needs to build a data cornerstone that can reflect the dynamics of the physical world of the visitor in real time and continuously. Therefore, performing the step of acquiring the original position stream of the target visitor can capture the complete and uninterrupted movement of the visitor in the exhibition hall space through high-frequency and continuous positioning data acquisition. This provides high-density and high-timeliness original data input for subsequent kinematic state calculation, trajectory prediction and active environment control, thereby changing the control logic from passive response to point to active prediction of flow, which is a fundamental prerequisite for solving the problems described in the background art.

[0017] In an exemplary embodiment of the present application, step S1 is operated as follows: first, a complete indoor high-precision positioning facility is deployed in the exhibition hall. For example, a plurality of ultra-wideband (UWB) positioning base stations are installed on the ceiling or wall of the exhibition hall according to a predetermined geometric layout, ensuring that the signal can cover the entire exhibition hall area without dead angles. At the same time, a portable tag with an embedded UWB chip is provided for each target visitor entering the exhibition hall, which can be a badge, a bracelet or a small pendant, and each tag is assigned a globally unique visitor identifier.

[0018] When the visitor moves freely in the exhibition hall, the UWB tag worn by the visitor will continuously emit pulsed signals at a predetermined high frequency, for example 10 Hz. The UWB base station network deployed in the exhibition hall will receive these signals in real time and accurately record the arrival time of the signals. These base stations communicate with a central positioning calculation engine through wired or wireless means, which collects signal reception data from at least three different base stations. The positioning calculation engine runs a positioning algorithm such as time difference of arrival (TDOA) inside, which calculates the real-time position coordinates (x, y, z) of the UWB tag in the three-dimensional coordinate system of the exhibition hall by comparing the time difference of the same signal received by different base stations. The calculation process is also executed at a frequency of 10 Hz, ensuring the continuity and real-time nature of the position data.

[0019] The output of this location calculation engine is the raw location stream of the target visitor. This data stream is a continuous sequence of data frames arranged in chronological order, each frame encapsulating the location information of a specific visitor at a specific time. Its data structure can be specifically represented as: (Visitor Unique Identifier, X-axis Coordinate, Y-axis Coordinate, Z-axis Coordinate, Timestamp). For example, within a certain time period, for a target visitor identified as Visitor A, the data stream might appear as three consecutive frames of data as follows: (Visitor A, 10.25, 15.88, 1.65, 1678886400.100), (Visitor A, 10.31, 15.90, 1.65, 1678886400.200), (Visitor A, 10.38, 15.91, 1.65, 1678886400.300). The coordinate values ​​are in meters, accurate to the centimeter level, and the timestamps are standard UNIX timestamps, accurate to the millisecond level.

[0020] In step S2, kinematic state calculation and twin state update are performed on the original location flow of the target visitor to obtain the visitor's kinematic state at time T. Correspondingly, although the original location flow obtained in the previous step has high-frequency and continuous characteristics, it is essentially a discrete observation containing random noise and measurement errors, reflecting only the visitor's point information in space. Directly performing trajectory prediction based on this jagged and unsmooth original data will lead to severe fluctuations and significant deviations in the prediction results, failing to provide a stable and reliable basis for subsequent active environmental control. Therefore, this application performs kinematic state calculation and twin state update on the original location flow of the target visitor to extract the visitor's true, smooth, and complete kinematic state (including position, velocity, and acceleration) from the noisy observation data through filtering and state estimation techniques. This provides a precise and stable dynamic digital twin model foundation for subsequent high-precision trajectory prediction and active environmental rendering.

[0021] In one exemplary embodiment of this application, Figure 3 This is a flowchart of step S2 in the smart exhibition hall control method based on digital twins according to an embodiment of this application. Figure 3 As shown, step S2, which involves performing kinematic state calculation and twin state update on the original location flow of the target visitor to obtain the visitor kinematic state at time T, includes: S21, extracting a single data point from the original location flow of the target visitor at the current time to obtain the original location measurement value at time T; S22, performing kinematic state estimation based on a Kalman filter on the original location measurement value at time T to obtain the estimated state at time T; and S23, performing state vector deconstruction and data object encapsulation on the estimated state at time T to obtain the visitor kinematic state at time T.

[0022] In the above exemplary embodiment, step S2 is operated as follows: firstly, the processing of S21 is performed. Specifically, in one processing cycle, for a specific target visitor, e.g. visitor A, the data frame with the latest timestamp is extracted from its corresponding raw position stream. If the current processing time T corresponds to the timestamp 1678886400.300, the data frame (visitor A, 10.38, 15.91, 1.65, 1678886400.300) is taken from the data stream. Only the three-dimensional coordinate information is extracted from the data frame to form a position measurement vector. This is the raw position measurement value at time T, denoted as Z_T, whose specific value is [10.38, 15.91, 1.65] with the unit of meter.

[0023] The process of S22 is then performed. The Kalman filter is a recursive, optimal state estimation algorithm that can effectively estimate the dynamic, unknown state from a series of incomplete and noisy measurements through a prediction-update cycle. In this embodiment, the Kalman filter is constructed to match the kinematic characteristics of the visitor in three-dimensional space. First, define the state vector x that describes the complete kinematic state of the visitor, which is a nine-dimensional column vector constructed as follows: x = transpose of [p_x, p_y, p_z, v_x, v_y, v_z, a_x, a_y, a_z]. Here, p_x, p_y, p_z represent the visitor's position along the X, Y, Z axes, respectively; v_x, v_y, v_z represent its velocity components; and a_x, a_y, a_z represent its acceleration components. Second, construct the state transition matrix F. This matrix, which is based on the uniform acceleration motion physical model, describes how the state vector x evolves from the previous time (T - At) to the current time (T), where At is the sampling time interval of the positioning data, determined by the data output frequency preset by the underlying positioning facility, for example, 0.1 seconds for a 10 Hz output frequency. The state transition matrix F is a 9x9 matrix whose specific values are determined according to the underlying kinematic formulas. For a single dimension, the current position is a linear combination of the previous position, velocity, and acceleration (p_t = p_{t-1} + v_{t-1} At + 0.5 a_{t-1} At²), the current velocity is a combination of the previous velocity and acceleration (v_t = v_{t-1} + a_{t-1} At), and the acceleration is assumed to remain constant over At (a_t = a_{t-1}). Extending this relationship to three-dimensional space and considering that the motion in each dimension is independent of the others, the resulting 9x9 state transition matrix F is a block diagonal matrix composed of three identical 3x3 submatrices, each of which precisely represents the single-dimensional kinematic relationship described above. Third, define the observation matrix H, which is used to map the nine-dimensional state vector to the three-dimensional measurement space, as it relates the state to the measurements. Since the positioning facility in this embodiment can only directly measure the visitor's position, H is a 3x9 matrix that extracts the position components [p_x, p_y, p_z] from the state vector x. The structure of this matrix is very clear, with the first three columns forming a 3x3 identity matrix, i.e., the elements at positions (1,1), (2,2), and (3,3) are 1, and all other elements are 0. This structure ensures that when H is multiplied by the state vector x, the first three elements of the state vector, i.e., the position components, are accurately and distortion-free selected for comparison with the actual position measurements. Finally, set two key noise covariance matrices: the process noise covariance matrix Q and the measurement noise covariance matrix R.Q is a 9x9 matrix that quantifies the inherent uncertainty of the uniform acceleration motion model, i.e., it acknowledges that the acceleration of the visitor is not constant but subject to random perturbations. The parameter values of the Q matrix can be set by statistical analysis of a large amount of historical trajectory data of visitors to reflect the acceleration variations caused by typical motion patterns of visitors in the exhibition hall (e.g., walking, turning, stopping). R is a 3x3 matrix that represents the measurement error of the UWB positioning facility itself. The diagonal element values of R are determined by performing long-time continuous positioning measurements of the UWB tag under static conditions and calculating the variances of the three-dimensional coordinate outputs. For example, if the measurement standard deviation of the positioning facility on each axis is 0.05 meters, the diagonal elements of the R matrix can be set to 0.0025. The specific process of running the state estimation is divided into two stages at time T. The first stage is prediction. Using the optimal state estimation value x_{T-ΔT} and its error covariance matrix P_{T-ΔT} obtained at time T-ΔT, the prior estimation value of the state at time T is predicted by the state transition equation. At the same time, the prior estimation of the error covariance at time T is also predicted. This step is completely based on the internal motion model and is a pre-estimation of the current state before obtaining new measurement values. The second stage is updating. After receiving the original position measurement value Z_T = [10.38, 15.91, 1.65] at time T, the Kalman gain K is first calculated. The Kalman gain is a key weight matrix, and its size depends on the relative size of the predicted error and the measurement noise. If the measurement noise is small (R value is small), the gain is large, which means that more trust is given to the new measurement value; otherwise, more trust is given to the predicted value of the model. Then, the prior state estimation value obtained in the prediction stage is corrected using the Kalman gain. The correction logic is: the final optimal estimation value is equal to the prior estimation value plus a correction amount, which is proportional to the residual between the actual measurement value and the predicted measurement value. After this step of calculation, the optimal state estimation value at time T is finally obtained, which is the estimated state x_T at time T. For example, after the complete prediction and update cycle, for the input Z_T = [10.38, 15.91, 1.65], the output estimated state at time T, i.e., the state vector x_T, can be a nine-dimensional vector as follows: [10.37, 15.90, 1.65, 0.65, 0.05, 0, -0.5, -0.5, 0] transpose. This vector not only provides a filtered and smoothed position (10.37, 15.90, 1.65), but also provides the instantaneous velocity of the visitor at this moment, i.e., X direction 0.65 m / s, Y direction 0.05 m / s, and acceleration, i.e., X and Y directions are both -0.5 m / s², which indicates that the visitor may be decelerating.

[0024] Finally, the process of S23 is performed. First, state vector deconstruction is performed. This process is to parse and split the input nine-dimensional vector according to the pre-defined state vector structure, and map it to independent vectors with clear physical meaning. In this embodiment, the first three elements of the estimated state at time T, i.e., the state vector x_T, are defined as the position coordinates of the visitor in three-dimensional space, the middle three elements are the velocity components, and the last three elements are the acceleration components. According to this rule, the input x_T vector is extracted: the first step extracts the first to third elements of the vector [10.37, 15.90, 1.65] to construct the position vector P_T at time T. This vector indicates that at time T, the estimated position of the visitor is at X axis 10.37 meters, Y axis 15.90 meters, and Z axis 1.65 meters in the exhibition hall coordinate system. The second step extracts the fourth to sixth elements of the vector [0.65, 0.05, 0] to construct the velocity vector V_T at time T. This vector indicates that the instantaneous velocity of the visitor at this moment is 0.65 m / s in the X axis direction, 0.05 m / s in the Y axis direction, and 0 m / s in the Z axis direction. The third step extracts the seventh to ninth elements of the vector [-0.5, -0.5, 0] to construct the acceleration vector A_T at time T. This vector indicates that the instantaneous acceleration of the visitor at this moment is -0.5 m / s² in the X axis direction, -0.5 m / s² in the Y axis direction, and 0 m / s² in the Z axis direction, which may reflect that the visitor is decelerating or preparing to turn. Next, data object encapsulation is performed. This process is to integrate the three physical vectors deconstructed in the previous step with the context information of the state to encapsulate them into a structured data object. This data object not only contains complete kinematic information, but also has unique identity and time. In order to maintain the coherence of the data stream, the visitor identifier and timestamp information used in encapsulation are derived from the original position stream data frame corresponding to time T. Specifically, a data object is created and the following fields are filled in: visitor identifier: obtained from the corresponding data frame, the value is visitor A. timestamp: obtained from the corresponding data frame, the value is 1678886400.300. position vector: the deconstructed position vector P_T, i.e., [10.37, 15.90, 1.65], is assigned to this field. velocity vector: the deconstructed velocity vector V_T, i.e., [0.65, 0.05, 0], is assigned to this field. acceleration vector: the deconstructed acceleration vector A_T, i.e., [-0.5, -0.5, 0], is assigned to this field. Finally, a complete data object is output, which is the kinematic state of the visitor at time T. It clearly and completely records the complete kinematic information of the specific visitor at a specific time.

[0025] In step S3, the kinematic state of the visitor at time T is predicted based on the kinematic model and spatial constraints to obtain The predicted trajectory at time T. It can be understood that, although the previous step has successfully solved the precise and complete kinematic state of the visitor at time T, it is only a precise description of the current snapshot. In order to fundamentally solve the lag and abruptness brought by the instantaneous-passive control mentioned in the background art, the control logic must have foresight, i.e. from perceiving the present to predicting the future. Therefore, the present application further combines the exhibition hall layout model to predict the trajectory of the visitor's kinematic state at time T, evolving the static current state into a dynamic future trajectory, and generating a high-probability moving path map of the visitor in the future. This predicted trajectory is the data basis for realizing environmental pre-rendering and active service, which enables the issuance of control instructions to lead the actual arrival of the visitor, thereby realizing seamless and smooth immersive experience.

[0026] In an exemplary embodiment of the present application, step S3, in combination with the exhibition hall layout model, performs trajectory prediction based on the kinematic model and spatial constraints on the visitor's kinematic state at time T to obtain the predicted trajectory at time T. In an exemplary embodiment of the present application, step S3, in combination with the exhibition hall layout model, performs trajectory prediction based on the kinematic model and spatial constraints on the visitor's kinematic state at time T to obtain the predicted trajectory at time T. In an exemplary embodiment of the present application, step S3, in combination with the exhibition hall layout model, performs trajectory prediction based on the kinematic model and spatial constraints on the visitor's kinematic state at time T to obtain the predicted trajectory at time T.

[0027] In the above exemplary embodiment, step S3 operates as follows: first, perform the processing of S31. This process aims to generate an idealized trajectory that is not constrained by the physical space based on the pure kinematic model. In an exemplary embodiment of the present application, step S31, taking the visitor's kinematic state at time T as the initial state of the kinematic model, performs path deduction on the visitor's kinematic state at time T to obtain the initial predicted trajectory at time T, including: performing path deduction on the visitor's kinematic state at time T with the following formula: ; wherein, is the position vector at time T, is the velocity vector at time T, is the acceleration vector at time T, is a small time step. Here, a total prediction time , for example 2 seconds, is defined, which defines how far into the future needs to be predicted. At the same time, a small time step , for example 0.1 seconds, is defined, which determines the degree of detail of the predicted trajectory. Path deduction is an iterative process, and a total of 20 times, i.e. 20 times. In particular, the total prediction duration The determination aims to balance the time margin required for pre-control and the accuracy decay of long-term prediction, while the setting of the small time step is mainly based on the data sampling frequency of the original position stream to ensure the resolution and computational efficiency of trajectory deduction. Specifically, in the first iteration, the kinematic state at time T is taken as the initial condition, and the path deduction formula is applied to calculate the position at time T+0.1 second. The specific calculation is as follows: new X coordinate = 10.37 + 0.65*0.1 + 0.5*(-0.5)*(0.1)² = 10.4325 meters. New Y coordinate = 15.90 + 0.05*0.1 + 0.5*(-0.5)*(0.1)² = 15.9025 meters. New Z coordinate = 1.65 + 0*0.1 + 0.5*0*(0.1)² = 1.65 meters. At the same time, the velocity vector of the point is updated. In subsequent iterations, the new position and new velocity calculated in the previous step are taken as input, and the process is repeated until all 20 deductions are completed. This process eventually generates a sequence of 21 three-dimensional coordinate points (including the initial point), which is the initial predicted trajectory at time T+2 seconds.

[0028] In particular, when predicting the future behavior of visitors, if a purely kinematic model is used, i.e. , its essence is to regard visitors as a physical particle with no memory and no intention. This model assumes that the future movement of visitors is determined only by the current instantaneous position, velocity, etc. state, completely ignoring the core driving force of human behavior - intention. In the exhibition hall scene, the movement of visitors has a clear purpose, i.e. to visit exhibits, and their path selection and speed change are strongly influenced by the next target exhibit. At the same time, environmental semantics (such as corridors and exhibition areas) and behavior mode switching (such as walking and staying) are also complex behaviors that cannot be expressed by a purely physical model. This simplified prediction method cannot accurately reproduce the inevitable deceleration, stay and wander of visitors in front of key exhibits, nor can it predict the acceleration or turning of visitors in the corridor due to the attraction of distant targets. Therefore, in order to generate a high-fidelity trajectory that truly conforms to the visitor's visit logic and has long-term prediction ability, it is necessary to go beyond simple physical deduction. To this end, the path deduction of the mixed state trajectory prediction based on intention perception in the present application divides the single physical deduction process into two higher-level associated modules: intention prediction and state-dependent trajectory generation, thereby simulating the decision-making process of visitors and generating an initial predicted trajectory that combines purpose, behavior authenticity and long-term prediction ability.

[0029] Based on this, in one preferred exemplary embodiment of the present application, step S31 takes the kinematic state of the visitor at time T as the initial state of the kinematic model, and performs path deduction on the kinematic state of the visitor at time T to obtain The initial predicted trajectory at time t includes: Obtain a list of candidate exhibits. It's understandable that including all exhibits in the intention prediction calculation within a large exhibition hall would incur enormous and unnecessary computational overhead. To efficiently focus on the visitor's most likely current goal, a pre-screening process is needed to generate a manageable and highly relevant set of candidate exhibits, laying the foundation for subsequent accurate calculations. In practice, this process centers on the visitor's location at time T and invokes a pre-built exhibition hall layout model. First, through spatial proximity analysis, a reasonable decision radius is defined, such as 15 meters, initially considering all exhibits whose physical locations are within this radius. Then, accessibility analysis is performed on these initial candidates, utilizing path planning information from the layout model to eliminate exhibits that, although close in a straight line, are inaccessible due to walls or other obstacles. For example, visitor A is currently located in the main passageway, with exhibits A, B, and C within a 15-meter radius. Exhibit C is located in a closed, independent exhibition room, requiring a longer detour to reach; therefore, it is eliminated through accessibility analysis. Finally, the output list of candidate exhibits is {Exhibit A, Exhibit B}, which will serve as the input for the next step of intent prediction.

[0030] Based on visitor location, the high-level intent prediction module calculates the utility score of each candidate exhibit in the candidate exhibit list using a utility function, and then normalizes it using the Softmax function to obtain a normalized utility score sequence. The candidate exhibit corresponding to the highest normalized utility score in the normalized utility score sequence is selected as the predicted exhibit. .

[0031] in, Indicate candidate exhibits Static attractiveness index (such as historical popularity, whether it is a core exhibit). It is the visitor's current location vector. To candidate exhibits The path distance to the center (not the straight-line distance); It is the visitor's current velocity vector and the direction vector pointing to the exhibit The angle between them represents the inertia of motion, meaning that visitors are more likely to continue moving in the current direction. These are the weighting coefficients for each item. Candidate exhibits The utility score, yes Normalization function, It is a normalized utility score sequence. Is to make Candidate exhibits that achieve the maximum value , is the predicted exhibit. Accordingly, the movement of the visitor is driven by his subjective intention, and a quantitative model is needed to evaluate which exhibit in the candidate list has the greatest appeal to the visitor. To this end, the present application determines a single target exhibit with the highest probability from multiple possibilities through a two-stage process, providing a clear long-term attraction source for subsequent trajectory generation. In specific implementation, first, the high-level intention prediction module calculates the utility score of each candidate exhibit in the candidate exhibit list based on the current state of the visitor and the properties of the candidate exhibits . Among them, the static attraction index is pre-set by the exhibition hall planner based on the importance of the exhibits and historical visitor data, while the weight coefficients are obtained through expert experience tuning or machine learning calibration based on historical trajectory data sets. For example, set the weight , for the candidate exhibit A, its is set to 10, and the calculation is 5 meters, 30 degrees, and for exhibit B, its is set to 4, and the calculation is 12 meters, 90 degrees, the utility score and are calculated by substituting the current state of the visitor and the properties of the exhibits. Then, the Softmax function is used to normalize the obtained utility score sequence, converting it into a probability distribution, ensuring that the sum of probabilities is 1, and the probability of the exhibit with higher utility being selected is greater. Based on the calculation of the above example, the result may be much greater than , and after the Softmax function conversion, the probability distribution like and is obtained. Finally, the candidate exhibit with the highest probability value in the normalized utility score sequence is taken as the predicted exhibit . Therefore, the predicted exhibit is determined to be exhibit A.

[0032] The trajectory generation module calculates the state based on the distance between the predicted exhibit and the current location of the visitor, and in response to the distance between the predicted exhibit and the current location of the visitor being greater than the distance threshold, the visitor state is the moving state, and the position vector of the visitor at the current time is combined by pure kinematics extrapolation and the attraction of the target to obtain the position vector at the next time, that is: .

[0033] wherein, is the distance between and , and is the distance threshold, denotes a simplified kinematic model based on pure physical inertia, denotes vector division by its modulus, a unit vector pointing towards the target exhibit, is a velocity factor used to adjust the strength of the attractive force, its magnitude is comparable to the typical moving speed of visitors. For example, it can be preset as the average moving speed of all visitors in the history data of the exhibition hall, i.e. a scalar, such as 1.2 meters per second, to ensure that the size of the attractive force is reasonable and in line with physical intuition, is the time step and the above is the same, for example 0.1 seconds, is a dynamic weight factor, which depends on the distance to the target, for example when far from the target, is small, the visitor is mainly affected by inertia; when close to the target, becomes large, the target attraction plays a dominant role, guiding the visitor to accurately approach the target and naturally slow down, is a preset distance scale parameter used to control the sensitivity to distance changes. When the value is large, the change is relatively gentle, meaning that the visitor needs to be very close to the target before being significantly affected by the attractive force; When the value is small, the change is relatively sharp, meaning that the visitor will start to be dominated by the attractive force at a relatively long distance, which can be set according to the average size of the exhibition hall or through experimental data, for example, it can be set to half of the average distance between exhibits, is the position vector at the next time.

[0034] In response to the distance between the predicted exhibit and the visitor's current position being less than or equal to the distance threshold, the visitor's state is a stay state, and a small-range random walk simulation is performed on the visitor's current position vector to obtain the next time position vector, i.e. ; wherein, is a multivariate Gaussian distribution with mean 0 and covariance matrix sampling, wherein defines the size and shape of the stay range, is the position vector at the next time. That is, the visitor's behavior pattern is not fixed, and the movement characteristics of the visitor in the process of approaching the target and in front of the target are completely different. Different models need to be used to describe these two states in order to generate trajectories that conform to real behavior. The implementation effect is to generate a smooth trajectory that can naturally transition from travel to stay, accurately reproducing the visitor's deceleration and wandering behavior in front of the exhibit. In specific implementation, the underlying trajectory generation module first determines the distance between the predicted exhibit and the visitor's current position and the preset distance threshold (e.g. 1.5 meters, representing the personal interaction space of the exhibit) are compared. In response to , the visitor state is determined to be the moving state, and the position of the next time instant is calculated by a hybrid weighting formula. The formula is a weighted combination of pure kinematic extrapolation and gravitational motion towards the target, with the weights dynamically calculated by the above formula, achieving the effect of being dominated by inertia when far away and being dominated by gravity when close, and naturally decelerating. In response to , the visitor state is determined to be the staying state, and the position of the next time instant is simulated by performing a random walk around the target point. Among them is a small displacement sampled from a multivariate Gaussian distribution with a mean of 0 and a covariance matrix , simulating the slight movement of the visitor when staying and watching in front of the exhibit. By iteratively performing this step, the complete initial prediction trajectory within the time instant is generated. In this way, the improved path inference model will have semantic perception and long-term prediction ability, and can obtain the destination of the visitor after a few seconds or even tens of seconds through intention prediction, and has behavior authenticity, i.e. accurately reproducing the key behavior patterns of the visitor in the exhibition hall, such as acceleration or turning in the corridor due to the target, and deceleration, staying and wandering in front of the exhibit. Therefore, based on the prediction model, the control method of the smart exhibition hall according to the embodiments of the present application can make more advanced preparatory actions, for example, when it is predicted that the visitor will enter the staying state, the key lighting, multimedia and sound effects of the exhibit are rendered to the best state in advance and smoothly, achieving a smooth and natural control experience.

[0035] Next, the process of S32 is performed. This process ensures the physical reality of the predicted trajectory. The exhibition layout model is a pre-constructed high-precision three-dimensional digital model, which is constructed from the architectural design drawings of the exhibition hall or the on-site point cloud data obtained by laser scanning technology, and is refined by three-dimensional modeling software. This model not only accurately describes the precise spatial coordinates, dimensions and directions of all fixed obstacles (such as walls, showcases, columns) in the exhibition hall in the form of digital geometry, but more importantly, it also performs semantic functional division on the entire space. By assigning attribute labels such as walkable ground, physical obstacles, exhibition item influence area to different geometric regions, the safe path that visitors can freely move and the impassable boundary in the digital space are clearly defined, thereby mapping the exhibition hall as a structured environment for calculation and analysis. Collision detection is performed on each predicted point in the initial predicted trajectory generated in the previous sub-step, i.e., starting from the second point. That is, it is judged whether the three-dimensional coordinates of the point fall into the geometric body labeled as non-walkable region in the exhibition layout model. For example, when deducing to the 5th point, i.e., T+0.5 seconds, the calculated initial predicted point coordinates are [10.80, 15.95, 1.65]. The exhibition layout model shows that there is a virtual wall at X=10.75 meters. Since 10.80>10.75, the predicted point has penetrated the wall, triggering a collision. Once the collision is detected, the path correction mechanism is immediately started. The core of the correction is to adjust the position of the collision point and the velocity vector on which the subsequent deduction is based. An effective correction is to process the velocity component that leads to the collision. In this example, the visitor's X-direction velocity leads to the collision. The correction algorithm will force the X-coordinate of the point to be 10.75 meters, the wall boundary value, and the velocity vector V(T+0.5) of the point will be zeroed in the wall normal direction, i.e., the X-axis direction, or an elastic decay will be applied, while the velocity components in the wall tangent direction, i.e., the Y-axis and Z-axis, will be retained. This corrected point [10.75, 15.95, 1.65] and its corrected velocity vector will serve as the initial state for the next path deduction (calculating the T+0.6 second moment). This collision detection and correction process will accompany the entire path deduction, ensuring that each point in the generated trajectory sequence is located within the walkable region. After completing the processing of all 20 predicted points, an ordered list consisting of 21 three-dimensional coordinate points is finally obtained. This list is the predicted trajectory at the moment, which is a path that conforms to the visitor's current movement trend and is physically realizable in the future 2 seconds.

[0036] In step S4, based on the visitor portrait and the scene script library, the predicted trajectory is generated along the ​The predicted trajectory at the moment generates a dynamic environment bubble. It can be understood that the future predicted trajectory generated in the previous step is only a geometrically valid path in three-dimensional space, which does not contain any environmental interaction intention by itself. In order to convert this pure physical path prediction into a meaningful and personalized visit experience, it is necessary to inject content and soul into it. Therefore, the present application performs a dynamic environment bubble along the predicted trajectory based on the visitor portrait and the scene script library to deeply integrate the path information of where the visitor will go with the identity information of who the visitor is and the content information of what the exhibition hall can provide, thereby actively and prospectively designing a personalized environmental experience scheme wrapped around the future path of the visitor, providing specific and executable blueprints for realizing the final device pre-rendering and active service.

[0037] In an exemplary embodiment of the present application, step S4, based on the visitor portrait and the scene script library, generates a dynamic environment bubble along the predicted trajectory at the moment, including: The predicted trajectory at the moment generates a dynamic environment bubble, including: S41, performing spatial intersection analysis on the predicted trajectory at the moment and the exhibition hall layout model to obtain a potential target list; S42, performing multi-dimensional matching scoring and decision on the potential target list, the visitor portrait and the scene script library to obtain a selected scene script and an activation point; S43, performing instantiation of the selected scene script and the activation point based on the environmental parameter field function to obtain an environmental field function set; S44, performing data integration and packaging on the environmental field function set and the predicted trajectory at the moment to obtain the dynamic environment bubble. The predicted trajectory at the moment generates a dynamic environment bubble, including: S41, performing spatial intersection analysis on the predicted trajectory at the moment and the exhibition hall layout model to obtain a potential target list; S42, performing multi-dimensional matching scoring and decision on the potential target list, the visitor portrait and the scene script library to obtain a selected scene script and an activation point; S43, performing instantiation of the selected scene script and the activation point based on the environmental parameter field function to obtain an environmental field function set; S44, performing data integration and packaging on the environmental field function set and the predicted trajectory at the moment to obtain the dynamic environment bubble. The predicted trajectory at the moment generates a dynamic environment bubble, including: S41, performing spatial intersection analysis on the predicted trajectory at the moment and the exhibition hall layout model to obtain a potential target list; S42, performing multi-dimensional matching scoring and decision on the potential target list, the visitor portrait and the scene script library to obtain a selected scene script and an activation point; S43, performing instantiation of the selected scene script and the activation point based on the environmental parameter field function to obtain an environmental field function set; S44, performing data integration and packaging on the environmental field function set and the predicted trajectory at the moment to obtain the dynamic environment bubble.

[0038] In the above exemplary embodiment, step S4 operates as follows: Firstly, the process of S41 is performed. The pre-constructed exhibition hall layout model contains not only the physical barrier information, but also defines a three-dimensional interaction area for each exhibit or specific area, and associates the corresponding semantic label. The shape of these interaction areas is like a sphere, a cube, and the size is pre-set according to the optimal viewing distance, interactive range and other factors of the exhibit. The process of spatial intersection analysis is an iterative detection loop. The process will traverse each coordinate point on the predicted trajectory and determine the geometric relationship with all the exhibit interaction areas defined in the exhibition hall layout model. The determination includes two cases: one is direct intersection, that is, to determine whether the trajectory point falls within the geometric body of a certain interaction area; the second is proximity detection, that is, to calculate the shortest distance between the trajectory point and the surface of a certain interaction area, and to determine whether the distance is less than a pre-set proximity threshold, for example, 0.5 meters. The setting of the threshold aims to capture the case where the visitor shows a clear intention to approach although he has not directly entered the area. Take a specific embodiment as an example, the 13th point in the predicted trajectory list corresponds to T+1.2 seconds, and its coordinates are [10.95, 16.10, 1.65]. The exhibition hall layout model defines an exhibit named Exhibit A, whose semantic label is [historical relics, Bronze Age], and its interaction area is set as a sphere with a center point at [11.00, 16.15, 1.50] and a radius of 1 meter. By calculation, the distance between the trajectory point and the sphere center is less than the radius, and it is determined as direct intersection. At this time, Exhibit A is identified as a potential target. Continue to traverse, the 19th point in the trajectory list corresponds to T+1.8 seconds, and its coordinates are [11.50, 16.80, 1.65]. Another exhibit named Exhibit B in the model has a semantic label of [modern technology, interactive experience], and its interaction area is a cube. Calculation shows that the trajectory point does not fall into the cube, but the distance between it and the nearest surface of the cube is 0.4 meters, which is less than the pre-set proximity threshold of 0.5 meters. Therefore, Exhibit B is also identified as a potential target. In the entire traversal process, in order to avoid duplication, once a certain exhibit is identified as a potential target, if the subsequent trajectory point again associates with the interaction area of the same exhibit, it will not be added repeatedly. Finally, a structured potential target list is output, each element of the list contains the unique identification of the target exhibit, the semantic label set extracted from the layout model, and the trajectory point that triggered the identification as the activation point information. Based on the above example, the output potential target list is: [{target unique identification: Exhibit A, semantic label: [historical relics, Bronze Age], activation point: {coordinates: [10.95, 16.10, 1.65], time: T+1.2 seconds}}, {target unique identification: Exhibit B, semantic label: [modern technology, interactive experience], activation point: {coordinates: [11.50, 16.80, 1.65], time: T+1.8 seconds}}].

[0039] The process of S42 is then performed. This step also requires the invocation of two core preset databases: the visitor profile database and the scene script database. The visitor profile database is a structured database that stores the characteristics of each visitor. These characteristics can be obtained from ticket or registration information when the visitor enters, such as the visitor's chosen identity, such as a professional scholar, a general audience, a student, or dynamically generated by analyzing their stay time in front of different types of exhibits during the visit. In this embodiment, the profile data of the current visitor A is retrieved from the database, which contains: {Visitor Type: Professional Scholar, Interest Preference: [Historical Artifact, Ancient History]}. The scene script database is a collection of a series of interaction rules pre-arranged by the exhibition hall designer. Each script defines a complete environmental control scheme and binds specific trigger conditions. For example, the database contains the following two scripts: Script J1: its trigger condition T1 is {Visitor Type Requirement: [Professional Scholar, Researcher], Semantic Label Requirement: Historical Artifact}; its effect is to start the exhibit spotlight and play a deep academic explanation audio track. Script J2: its trigger condition Tj2 is {Visitor Type Requirement: [General Audience, Family Parent-Child], Semantic Label Requirement: Interactive Experience}; its effect is to activate the floor guide light effect and play an interesting science popularization animation. The matching score process will combine each target (index k) in the potential target list with each script (index j) in the scene script database in pairs, and apply the scoring formula for calculation. In an exemplary embodiment of the present application, step S42, multi-dimensional matching scoring and decision making are performed on the potential target list, visitor profile, and scene script database to obtain the selected scene script and activation point, including: multi-dimensional matching scoring of the potential target list, visitor profile, and scene script database is performed using the following formula: ; wherein, is the type of visitor profile, is the semantic label of the th potential target in the potential target list, is the trigger condition of the th script in the scene script database, is the visitor type matching function, if meets the visitor type requirement of , it is 1, otherwise it is 0; is the semantic label similarity function, and are preset weight coefficients, which are pre-set by the exhibition hall policy maker according to the operation target, used to adjust the relative importance of the two dimensions of visitor identity and exhibit content in personalized recommendation, such as is 0.6, is 0.4, is and The matching degree. For example, for the first target item A in the potential target list, i.e., k=1, its semantic label For [historical artifacts, ...], begin iterating through the script library: match script J1, i.e., j=1: type of visitor A. For professional scholars, this meets the visitor type requirements of script J1. Therefore, the visitor type matching function The value is 1. Semantic tag for item A. Semantic requirements of script J1 The historical artifacts are highly similar, and the semantic tag similarity function is accurate. A high score, denoted as 1.0, is obtained through calculation, for example, based on the cosine similarity of word vectors. The final score is: Matches script J2, i.e., j=2: Visitor A's type The visitor type requirement of Script J2 is not met. Therefore The value is 0. The semantic tag also does not match. The score is very low, set to 0.1. The final score is: Next, for the second target item B in the potential target list (i.e., k=2), whose semantic label Sk2 is [modern technology, ...], the script library is traversed again: it matches script J1 (i.e., j=1): the visitor type is satisfied. The value is 1. However, the semantic tags do not match. The score is very low, set at 0.2. The final score is: Matches script J2, i.e., j=2: Visitor type not satisfied. The final score is 0. The score will be very low. After all combinations have been scored, the decision-making mechanism will compare all scores and select the combination with the highest score. In this example, The highest score globally is achieved. Therefore, the decision is as follows: select script J1 as the scene script to be executed, and select the activation point corresponding to the potential target item A associated with this highest score, i.e., {coordinates: [10.95, 16.10, 1.65], time: T+1.2 seconds}, as the trigger position for this script.

[0040] The process of instantiation is to parse each environmental effect defined in the script J1 and convert it into a concrete environmental parameter field function. The script J1 contains two core effects: starting the spotlight on the exhibit and playing the in-depth academic explanation audio track. First, the effect of starting the spotlight on the exhibit is instantiated. This effect is converted into a light intensity field function that describes the distribution of light intensity in space. In order to simulate the natural effect of the spotlight being brightest at the center and smoothly decaying towards the periphery, a Gaussian function is chosen as its mathematical model. The light intensity field function is defined as: Intensity_Light(p) = I_base + I_max x exp(-||p-p_activation||2 / (2 x sigma_light2)). Where p is an independent variable representing any three-dimensional coordinate point (x, y, z) in the exhibition space, I_base is the basic ambient light intensity of the exhibition hall, which is a preset value, for example, 20 lux, to ensure that the exhibition hall maintains basic lighting even without triggering effects, I_max is the maximum additional light intensity that the spotlight effect can achieve, which is predefined by the parameters of script J1, for example, 300 lux, to ensure that the exhibit can be significantly illuminated. p_activation is the center point of the function, which directly uses the coordinates of the input activation point [10.95, 16.10, 1.65], so that the center of the light effect is accurately aligned with the exhibit location that the visitor will focus on. sigma_light is the standard deviation of the Gaussian function, which controls the size of the light spot and the softness of the edge. This value is also preset by script J1, for example, sigma_light = 1.5 meters, representing that the light intensity decays to about 60% at a distance of 1.5 meters from the center. Next, the effect of playing the in-depth academic explanation audio track is instantiated. This effect is converted into an acoustic field volume field function that describes the distribution of sound volume in space. Similarly, in order to create a private auditory space centered on the exhibit and gradually weakening towards the periphery, a Gaussian function is also used for modeling. The acoustic field volume field function is defined as: Volume_Audio(p) = V_max x exp(-||p-p_activation||2 / (2 x sigma_audio2)). Where p is also an arbitrary three-dimensional coordinate point in space, V_max is the maximum volume of the explanation audio track, which is preset by script J1, for example, 65 decibels (dB), to ensure a clear auditory experience in the central area. p_activation still uses the input activation point coordinates [10.95, 16.10, 1.65], so that the sound seems to come from the exhibit. sigma_audio is the spatial influence radius of the sound field, which is preset by script J1, for example, sigma_audio = 2.5 meters. This value can be different from the sigma value of the light to define a sound bubble that is wider than the light spot range, ensuring that visitors can hear the explanation within a larger range near the exhibit.Thus, the originally abstract script is instantiated into two concrete, continuous mathematical functions. The final output is a set containing these two functions, i.e., the aforementioned environmental field function set: {light_function: Intensity_Light(p), audio_function: Volume_Audio(p)}. This function set constitutes a complete, computable environmental state description, which defines the expected light intensity and volume magnitude at every spatial coordinate point around the activation point.

[0041] Finally, the process of S44 is performed. First, a data object, i.e., the dynamic environment bubble, is instantiated, which is logically a container containing the complete experience plan of a specific visitor in the future period of time. Then, field filling is performed on this data object to complete data integration and encapsulation: the first step is to store the context information of the current process into the object. This includes the unique identifier of the visitor A inherited from the original data stream, and the reference timestamp T of generating the dynamic environment bubble, i.e., 1678886400.300, indicating that the plan is made based on the state of the visitor at this time. The second step is to store the complete predicted trajectory at T+2 seconds, i.e., an ordered list containing 21 three-dimensional coordinate points, as a field in the dynamic environment bubble object, which defines the range and path of the environment bubble in space and time. Finally, the environmental field function set generated in the previous step, i.e., a set containing two specific mathematical functions of Intensity_Light(p) and Volume_Audio(p), is stored as a field in the dynamic environment bubble object, which defines how the environmental parameters should change with the spatial position inside the environment bubble. Finally, the output is this completely filled data object dynamic environment bubble, such as {visitor unique identifier: "visitor A", generation timestamp: 1678886400.300, predicted trajectory: [[x0, y0, z0], [x1, y1, z1],..., [x 20 ,y 20 ,z 20The environmental field function set is defined as follows: {light_function:Intensity_Light(p)=20+300×exp(-||p-[10.95,16.10,1.65]||² / (2×1.5²)),audio_function:Volume_Audio(p)=65×exp(-||p-[10.95,16.10,1.65]||² / (2×2.5²))}}. It is a logically unified, self-contained data unit. Specifically, this dynamic environmental bubble is a comprehensive information body that explicitly declares: for visitor A, the expected movement path within the next 2 seconds is defined by the encapsulated trajectory data, while the ideal light intensity and sound environment along this path and its surrounding space are precisely described by the encapsulated environmental field function.

[0042] In step S5, based on the device layout model, the device control weights of the dynamic environment bubble are reverse-analyzed to obtain the device control flow. That is, the dynamic environment bubble generated in the previous step is a comprehensive data blueprint containing the visitor's future path and an idealized environmental parameter field. It precisely depicts the ideal lighting, sound field, and other expected states in future spacetime in the form of continuous mathematical functions. However, the physical exhibition hall environment is composed of a limited number of discrete devices (such as lighting fixtures and speakers) with fixed positions. Each of these devices has independent control parameters and influence ranges. Therefore, this idealized, continuous environmental blueprint cannot be directly executed by physical devices. To accurately map this forward-looking design scheme from the digital twin space to the physical reality, a reverse solution process is required. Therefore, in this application, based on the device layout model, the device control weights of the dynamic environment bubble are reverse-analyzed to transform this continuous, abstract environmental field function model into a set of discrete, immediately executable, and precise control command flows for each specific physical device.

[0043] In one exemplary embodiment of this application, Figure 4 This is a flowchart of step S5 in the smart exhibition hall control method based on digital twins according to an embodiment of this application. Figure 4 As shown, step S5, based on the device layout model, performs inverse analysis of the device control weights in the dynamic environment bubble to obtain the device control flow, including: S51, analyzing the device control weights in the dynamic environment bubble... S52. The predicted trajectory at time T is used to locate the trajectory point to obtain the reference point at time T; S53. The influence domain equipment is screened for the reference point at time T, the equipment layout model and the environmental field function set in the dynamic environment bubble to obtain the relevant equipment set; S54. The field function is sampled and the state is calculated for the relevant equipment set, the equipment layout model and the environmental field function set in the dynamic environment bubble to obtain the equipment control flow.

[0044] In the above exemplary embodiment, step S5 is operated as follows: Firstly, the process of S51 is performed. The trajectory point positioning is a target-oriented extraction process. Its target is to locate the activation point which is identified and finally selected in the whole step S4 to trigger the activation of the personalized scenario script. This activation point is the logical and spatial core of the experience blueprint carried by the whole dynamic environment bubble, and all the environmental field functions are instantiated around this point. Therefore, this activation point becomes the most ideal reference benchmark when performing the device control reverse analysis at the current T time. The implementation process is as follows: Firstly, the input dynamic environment bubble object is received. Then, the data structure of the object is parsed to extract the information of the activation point. In the previous step S42, through multi-dimensional matching score and decision, the combination of script J1 and exhibition A is finally selected, and the activation point associated with the exhibition A is determined, which is the closest point to the exhibition A on the predicted trajectory, and its specific information is {coordinate: [10.95, 16.10, 1.65], time: T+1.2 seconds}. All the instantiated environmental field functions (such as Intensity_Light(p) and Volume_Audio(p)) are constructed with the coordinate [10.95, 16.10, 1.65] of the activation point as the spatial center p_activation. Therefore, directly from the information contained in the dynamic environment bubble, the activation point coordinate which is predetermined and used as the instantiation benchmark of the field function is read out. Finally, the extracted three-dimensional coordinate vector [10.95, 16.10, 1.65] is taken as its only output, and this coordinate point is named as the reference point at T time, which means the spatial reference point used when performing the device control calculation at the current T time, not indicating that the point is the actual position of the visitor at T time.

[0045] Then the process of S52 is performed. It should be understood that the device layout model is a high-precision three-dimensional structured database, which is the static skeleton of the exhibition hall digital twin. It is determined in the exhibition hall construction or digital transformation stage by integrating building information model (BIM), computer-aided design (CAD) drawings, or using a three-dimensional laser scanner to scan the physical space to generate point cloud data, and then through artificial modeling and information annotation. The model accurately records the all-around information of each controllable physical device in the exhibition hall. For each device in the model, at least the following fields are included: a unique device identifier such as spotlight-07, a device type such as spotlight, precise three-dimensional spatial coordinates, and a key maximum effective influence radius. This radius is a value set in advance according to the physical characteristics of the device, such as the power and beam angle of the lamp, the rated power, directivity, and installation height of the speaker, etc. It defines the maximum spatial range of the device that can produce meaningful physical effects such as visible light and audible sound pressure. For example, a high-power spotlight installed on a 5-meter-high ceiling can have an effective influence radius of 8 meters; while a directional speaker for near-field interaction may only have a radius of 3 meters. The implementation process of the influence domain device screening is an iterative filtering process based on spatial proximity. The process is centered on the reference point p_ref at time T, and traverses each device in the device layout model. In each iteration, the following calculations and judgments are performed: extract the three-dimensional coordinates of the current traversed device, denoted as p_device. Calculate the Euclidean distance between the device and the reference point p_ref: Distance = ||p_device-p_ref||. Query the maximum effective influence radius of the device from the device layout model, denoted as R_max. Compare the calculated distance with the radius. If Distance < R_max, the device is determined to be a relevant device, and its unique device identifier is added to a temporary set. For example, the coordinates of the reference point p_ref in this application are [10.95, 16.10, 1.65]. The device layout model contains the following device information: Device One: {Identifier: spotlight-07, Type: spotlight, Coordinates: [11.00, 16.00, 4.50], Maximum Effective Influence Radius: 8.0 meters}. Device Two: {Identifier: spotlight-08, Type: spotlight, Coordinates: [15.00, 18.00, 4.50], Maximum Effective Influence Radius: 8.0 meters}. Device Three: {Identifier: speaker-03, Type: speaker, Coordinates: [11.50, 15.50, 3.00], Maximum Effective Influence Radius: 5.0 meters}. Device Four: {Identifier: spotlight-15, Type: spotlight, Coordinates: [25.00, 30.00, 4.50], Maximum Effective Influence Radius: 8.0 meters}. The screening process is as follows: for spotlight-07, the distance from the reference point is about 2.85 meters.Since 2.85 < 8.0, Spot-07 is determined as a relevant device. For Spot-08, the distance between it and the reference point is calculated to be about 4.86 meters. Since 4.86 < 8.0, Spot-08 is also determined as a relevant device. For Speaker-03, the distance between it and the reference point is calculated to be about 1.58 meters. Since 1.58 < 5.0, Speaker-03 is also determined as a relevant device. For Spot-15, its coordinates are far away from the reference point, and the calculated distance is much larger than its 8.0 meters of influence radius, so this device is ignored. After traversing all the devices in the device layout model that are related to light and sound, the identifiers of all the devices determined as relevant form a set, which is the set of relevant devices. In this example, the output set of relevant devices is: {Spot-07, Spot-08, Speaker-03}.

[0046] Finally, the process of S53 is performed. In addition to providing the coordinates of the devices, the device layout model also provides a device influence domain model for each device, which is a positive physical model function describing the precise mathematical relationship between the control parameter of the device, such as the brightness level 0-255, and the physical effect it produces in space, such as the light intensity at a specific distance and angle. The implementation of this step mainly consists of two stages: field function sampling and state calculation.

[0047] First, field function sampling is performed. This stage aims to determine what the ideal environment state should be at the reference point p_ref. The process takes the coordinates of the reference point p_ref [10.95, 16.10, 1.65] as the argument and substitutes it into each of the environment field functions encapsulated in the dynamic environment bubble to perform evaluation. For the light environment, substitute p_ref into the light intensity field function Intensity_Light(p). Since p_ref is the center point p_activation of the field function, the exponential term is exp(0) = 1, so the calculated expected light intensity I_target = 20 + 300 x 1 = 320 lux. For the acoustic environment, substitute p_ref into the sound field volume field function Volume_Audio(p). Similarly, the calculated expected volume V_target = 65 x 1 = 65 decibels. These two calculated values, 320 lux and 65 decibels, are the final targets for the subsequent state calculation.

[0048] Next, state computation is performed. The goal of this stage is to compute the specific control parameter values for each device in the relevant device set, so that the sampling goal described above is collectively achieved. The process starts with the lighting devices. There are two fixtures in the relevant device set, spotlight-07 and spotlight-08. Since multiple devices are working together, they need to be assigned reasonable contribution weights. The weights are assigned based on the potential influence capability of the device on the reference point, inversely proportional to the square of the distance. From the device layout model, the distance of spotlight-07 to the reference point d_07 is found to be about 2.85 meters, and the distance of spotlight-08 to the reference point d_08 is about 4.86 meters. Based on this, their normalized weights are calculated: the weight of spotlight-07 w_07 = 1 / (2.85)2~ 0.123. The weight of spotlight-08 w_08 = 1 / (4.86)2~ 0.042. After normalization, w_07_norm ~ 0.75, w_08_norm ~ 0.25. Then, the total desired light intensity I_target is distributed among the two spotlights according to the weights: the target light intensity that spotlight-07 needs to contribute I_target_07 = 320 x 0.75 = 240 lux. The target light intensity that spotlight-08 needs to contribute I_target_08 = 320 x 0.25 = 80 lux. Finally, the control parameter for each device is solved using the inverse function of the device impact domain model stored in the device layout model. Specifically, the device impact domain model itself is a forward physical model function that calculates the physical effect (e.g. light intensity) at a specific spatial point given an input device control value (e.g. brightness level), while its inverse function performs the opposite calculation. The inverse function takes as input the desired physical effect value to be achieved at a specific location, e.g. 240 lux light intensity at 2.85 meters, and combines known spatial relationships such as distance, angle, to inversely solve for the device control parameter value that must be set to achieve this effect. This inverse function model is pre-established based on mathematical modeling of the device's physical characteristics such as the light fixture's distribution curve file, or by actual calibration testing of the device to fit the functional relationship between physical effect and control parameter. For spotlight-07, its inverse function of the impact domain model is called, taking as input the request to produce 240 lux light intensity at 2.85 meters, and the model inversely calculates that its brightness level should be set to 220, i.e. within a control range of 0-255. For spotlight-08, its inverse function is similarly called, taking as input the request to produce 80 lux light intensity at 4.86 meters, and the model inversely calculates that its brightness level should be set to 145. For the sound device, speaker-03, since it is the only sound source in the relevant device set, it will independently take on the entire sound field creation task. Its distance to the reference point d_audio is about 1.58 meters. Its inverse function of the impact domain model is called, taking as input the request to produce 65 decibels sound volume at 1.58 meters, and the model inversely calculates that its volume level should be set to 85, i.e. within a control range of 0-100.

[0049] After calculating precise control values ​​for all devices in the relevant device set, these results are integrated and encapsulated into a structured instruction sequence, namely the device control flow. Its format is a list, where each element is a specific control instruction. In this embodiment, the output device control flow is: [{Device ID: Spotlight-07, Parameter: Brightness, Value: 220, Timestamp: T}, {Device ID: Spotlight-08, Parameter: Brightness, Value: 145, Timestamp: T}, {Device ID: Speaker-03, Parameter: Volume, Value: 85, Timestamp: T}]. This device control flow will be sent to the exhibition hall's underlying device control interface to complete the final driving of the physical environment.

[0050] In summary, the digital twin-based smart exhibition hall control method based on the embodiments of this application is explained, which constructs a new paradigm for smart exhibition hall control from passive response to active prediction. It first accurately grasps the current motion state of visitors by performing kinematic state calculations on the continuous position flow of visitors. Crucially, this application is not limited to the current state, but rather uses a kinematic model to predict the future motion state of visitors. The system accurately predicts the movement trajectory of visitors at any given moment. Following this predicted trajectory, a personalized dynamic environment bubble can be generated in advance, based on visitor profiles, proactively and progressively pre-rendering the atmosphere of the area the visitor is about to enter. Finally, by inversely parsing the environment bubble into specific device control flows, it enables the pre- and smooth adjustment of lighting, sound effects, and other equipment, thus solving the technical problems of abrupt control and poor user experience caused by perception lag and lack of predictive capabilities in traditional control schemes.

[0051] Figure 5 This is a block diagram of a digital twin-based smart exhibition hall control system according to an embodiment of this application. Figure 5 As shown, the digital twin-based smart exhibition hall control system 100 according to an embodiment of this application includes: an original location flow acquisition module 110, used to acquire the original location flow of the target visitor; a kinematic state analysis module 120, used to perform kinematic state calculation and twin state update on the original location flow of the target visitor to obtain the visitor's kinematic state at time T; and a predicted trajectory generation module 130, used to combine the exhibition hall layout model to perform trajectory prediction based on the kinematic model and spatial constraints of the visitor's kinematic state at time T to obtain the kinematic state of the visitor. Predicted trajectory at any moment; Dynamic environment bubble generation module 140, used to generate trajectories based on visitor profiles and scene script library, along... The predicted trajectory at any time generates a dynamic environment bubble; the device control flow generation module 150 is used to perform inverse analysis of the device control weights of the dynamic environment bubble based on the device layout model to obtain the device control flow.

[0052] Here, those skilled in the art can understand that the specific operations of each of the steps in the above-described digital-twin-based intelligent exhibition hall control system have been described in detail above with reference to the description of the digital-twin-based intelligent exhibition hall control method of Figures 1 to 4 , and thus repetitive descriptions thereof will be omitted.

Claims

1. A digital-twin-based intelligent exhibition hall control method, characterized in that, The method comprises the following steps: acquiring a target visitor original position stream; performing kinematic state calculation and twin state updating on the target visitor original position stream to obtain a T-time visitor kinematic state; In combination with the exhibition hall layout model, the kinematic state of the visitor at T time is predicted based on the kinematic model and the space constraint to obtain the predicted trajectory at T time. time. Based on the visitor profile and the scene script library, a dynamic environment bubble is generated along the predicted trajectory of the moment the predicted trajectory of the moment performing reverse analysis on a dynamic environment bubble based on a device layout model to obtain a device control stream. 2.The digital-twin-based intelligent exhibition hall control method according to claim 1, characterized in that, The method of performing kinematic state calculation and twin state updating on the target visitor original position stream to obtain a T-time visitor kinematic state comprises the following steps: extracting a single data point at a current time from the target visitor original position stream to obtain a T-time original position measurement value; performing Kalman filter-based kinematic state estimation on the T-time original position measurement value to obtain an estimated state at the T time; performing state vector deconstruction and data object encapsulation on the estimated state at the T time to obtain the T-time visitor kinematic state. 3.The digital-twin-based intelligent exhibition hall control method according to claim 2, characterized in that, By combining the exhibition hall layout model, trajectory prediction based on the kinematic model and spatial constraints is performed on the visitor's kinematic state at time T to obtain... The predicted trajectory at any given time includes: Taking the kinematic state of the visitor at time T as the initial state of the kinematic model, the kinematic state of the visitor at time T is deduced to obtain the initial predicted trajectory at time T T. Will The initial predicted trajectory at each moment is subjected to collision detection with the exhibition hall layout model, and based on the collision detection results, [the following is applied]. The initial predicted trajectory at time [time] is corrected to obtain the [property / method]. Predicted trajectory at any given moment. 4.The digital-twin-based intelligent exhibition hall control method according to claim 3, characterized in that, Taking the kinematic state of the visitor at time T as an initial state of the kinematic model, performing path deduction on the kinematic state of the visitor at time T to obtain an initial predicted trajectory at time T , comprising: performing path deduction on the kinematic state of the visitor at time T by using the following formula: ; wherein, is a position vector at time T, is a velocity vector at time T, is an acceleration vector at time T, is a small time step. 5.The digital-twin-based intelligent exhibition hall control method according to claim 1, wherein, Based on the visitor profile and the scene script library, a dynamic environment bubble is generated along The predicted trajectory of the moment generates a dynamic environment bubble, including: on performing a spatial intersection analysis of the predicted trajectory and the layout model of the showroom at the time instant to obtain a list of potential targets; performing multi-dimensional matching scoring and decision-making on a potential target list, a visitor portrait, and a scene script library to obtain a selected scene script and an activation point; performing instantiation based on an environmental parameter field function on the selected scene script and the activation point to obtain a set of environmental field functions; to an environmental field function set and data integration and encapsulation of the predicted trajectory at the time instant to obtain the dynamic environmental bubble. 6.The digital-twin-based intelligent exhibition hall control method according to claim 5, characterized in that, The process involves multi-dimensional matching, scoring, and decision-making based on a potential target list, visitor profile, and scenario script library to obtain selected scenario scripts and activation points. This includes performing multi-dimensional matching and scoring on the potential target list, visitor profile, and scenario script library using the following formula: ;in, Types of visitor profiles The first in the potential target list Semantic labels for potential targets For the first scene script library The triggering conditions for each script For visitor type matching functions, if satisfy If the visitor type requirement is met, the value is 1; otherwise, it is 0. For semantic label similarity function, and For preset weighting coefficients, yes and The degree of matching. 7.The digital-twin-based intelligent exhibition hall control method according to claim 1, wherein, The method of performing reverse analysis on a dynamic environment bubble based on a device layout model to obtain a device control stream comprises the following steps: For dynamic environment bubbles The predicted trajectory at time T is used to locate trajectory points to obtain a reference point at time T; performing influence domain device screening on a reference point at the T time, a device layout model, and a set of environmental field functions in the dynamic environment bubble to obtain a set of related devices; performing field function sampling and state calculation on the set of related devices, the device layout model, and the set of environmental field functions in the dynamic environment bubble to obtain the device control stream.

8. An intelligent exhibition hall control system based on digital twinning, characterized in that, The method comprises the following steps: an original position stream acquisition module for acquiring a target visitor original position stream; a kinematic state analysis module for performing kinematic state calculation and twin state updating on the target visitor original position stream to obtain a T-time visitor kinematic state; a predicted trajectory generation module, configured to combine the exhibition hall layout model, and perform trajectory prediction based on a kinematics model and a space constraint on a kinematics state of the visitor at the T time to obtain a predicted trajectory at the T time; a predicted trajectory at the T time. a dynamic environment bubble generation module for generating a dynamic environment bubble along a predicted trajectory of the visitor based on the visitor profile and a scene script library; at the moment in time; a device control stream generation module for performing reverse analysis on a dynamic environment bubble based on a device layout model to obtain a device control stream.

Images