Module and method for controlling and adjusting humidity of museum

Abstract

The invention discloses a humidity control and adjustment module and method for a museum, and relates to the technical field of intelligent exhibition halls. The method comprises the following steps: constructing a museum BIM model, deploying a high-precision mesh sensor in a museum, extracting a trend term, a periodic term and a residual term by adopting time sequence decomposition, and dynamically adjusting an LSTM network structure; a material water absorption characteristic matrix is established according to cultural relic materials, and showcase control groups are dynamically divided; establishing an equipment efficiency model and realizing multi-equipment task allocation by improving a Hungary algorithm; an exhibition hall digital twinborn body is constructed, and a physical space and a virtual simulation data stream are fused. According to the method, the precise three-dimensional model of the complex museum is generated in a mode of combining three-dimensional scanning and BIM, meanwhile, the material water absorption characteristic matrix is established according to cultural relic materials, showcase control groups are dynamically divided, the equipment efficiency model is established, multi-equipment task allocation is achieved, intelligent adjustment of the temperature and humidity of the museum is improved, and the service life of the museum is prolonged. And museum equipment can be supervised and maintained in time.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent exhibition halls, and particularly relates to a humidity control and adjustment module and method for museums. Background Art

[0002] As a key place for preserving and displaying precious cultural relics, museums have extremely strict and complex requirements for the preservation environment. Temperature and humidity have a crucial impact on the preservation state of museum cultural relics. Controlling the changes in temperature and humidity can not only extend the lifespan of cultural relics and ensure their best preservation and display state; on the contrary, if left to natural changes, it may cause unexpected damage to cultural relics.

[0003] In the past when technology was not yet developed, the humidity control means in museums were relatively simple. At that time, people mainly relied on the following natural wisdoms to adjust humidity: (1) Natural ventilation: Skillfully open doors and windows, and use the natural wind force to make the air flow and adjust the indoor humidity.

[0004] (2) Humidity adjustment materials: Use natural hygroscopic materials such as lime and silica gel to adsorb moisture in the air and reduce humidity.

[0005] (3) Artificial humidification: Place basins, sprayers, etc. indoors to increase the moisture in the air and maintain an appropriate humidity.

[0006] Although these traditional methods can adjust humidity to a certain extent, the effect is limited and difficult to control. Summary of the Invention

[0007] The purpose of the present invention is to provide a humidity control and adjustment module and method for museums. By combining 3D scanning and BIM, an accurate 3D model of a complex museum is generated. At the same time, a material water absorption characteristic matrix is established according to the material of cultural relics, the display cabinet control group is dynamically divided, and an equipment efficiency model is established to realize multi-device task allocation, solving the problems of single humidity control means, low efficiency, and high energy consumption in existing museums.

[0008] To solve the above technical problems, the present invention is realized through the following technical solutions: The present invention is a humidity control and adjustment module and method for museums, including the following steps: Step S1: Integrate the museum BIM model according to the museum building drawings and the data collected by indoor drones; Step S2: Deploy high-precision mesh sensors inside the museum, and use the sliding window algorithm to preprocess the collected data; Step S3: Use time series decomposition to extract the trend term, periodic term, and residual term from the preprocessed data, and dynamically adjust the LSTM network structure; Step S4: Establish a material water absorption characteristic matrix according to the material of cultural relics, and dynamically divide the showcase control groups; Step S5: Establish an equipment efficiency model, and realize multi-device task allocation through an improved Hungarian algorithm; Step S6: Deploy a micro computing unit at the sensor node to achieve local decision-making; Step S7: Construct a digital twin of the exhibition hall, and integrate the physical space and virtual simulation data streams.

[0009] As a preferred technical solution, in the step S1, the construction process of the museum BIM model is as follows: Step S11: Collect site conditions, architectural drawings, and parametric data of special-shaped components, establish a site terrain surface through Revit, and set a unified coordinate system; Step S12: Based on the Revit architectural template, layer by layer establish basic components, and generate special-shaped parts using a parametric family library or Dynamo script; Step S13: Use the MagiCAD plug-in to establish a humidity control system, and customize the pipeline layout rules according to the functions of the museum; Step S14: After the deployment of each exhibition hall of the museum is completed, use a drone to perform 3D scanning to obtain the point cloud data of the internal equipment of the museum; Step S15: Stitch, register, and fuse the point cloud data according to the point cloud data scanned by multiple different acquisition exhibition halls.

[0010] As a preferred technical solution, in the step S15, combine the point cloud data scanned by the drone in 3D with the museum BIM model, endow the combined BIM model with materials, add textures and lighting for rendering, and at the same time make a detailed list of the constructed equipment information for the indoor equipment scanned by the drone, and associate the operation and maintenance parameters of the equipment to form a digital asset library.

[0011] As a preferred technical solution, in the step S2, the deployment of the high-precision mesh sensor is divided according to spatial density and hierarchical installation; the spatial density division installs high-precision mesh sensors according to the importance of the museum cultural relics; the hierarchical installation division is divided into the ground layer, the showcase layer, and the top layer to install high-precision mesh sensors; the high-precision mesh sensor uses the LoRaWAN protocol to build a self-healing mesh network, which specifically includes a capacitive polymer film sensor, an infrared counter module, a barometric gradient sensor, and an auxiliary monitoring unit; the auxiliary monitoring unit includes a laser particle counter installed in the showcase and a microwave radar deployed at the access control position.

[0012] As a preferred technical solution, in step S2, the preprocessed data also needs to be subjected to outlier detection, and the specific detection method is as follows: Step S21: Determine whether the preprocessed data is approximately normally distributed through a histogram; If it conforms to the approximate normal distribution, then use Criterion for detection; If it is a skewed distribution, then mainly use the box plot method; Step S22: Set The threshold of the criterion, where: upper limit = mean + 3 * standard deviation, lower limit = mean - 3 * standard deviation; The data not between the upper and lower limits is regarded as an outlier; Step S23: Calculate the interquartile range by the box plot method, and the specific formula is as follows: ; Then the data less than and greater than is regarded as an outlier.

[0013] As a preferred technical solution, in step S3, the specific process of dynamically adjusting the LSTM network structure according to the trend term, periodic term, and residual term is as follows: Step S31: Perform inner loop decomposition on the preprocessed data to extract the trend term, periodic term, and residual term; When extracting the trend term, perform LOESS smoothing on the original sequence to extract the initial trend term; When extracting the periodic term, separate the periodic component from the detrended sequence; When extracting the residual term, residual = original sequence - periodic term - seasonal term; Step S32: Dynamically adjust the LOESS window size according to the slope change rate of the trend term; Step S33: Calculate the variance of the periodic component, and if the intensity is lower than the threshold, close the periodic branch; Step S34: Use the processed trend term, periodic term, and residual term as independent input channels; Step S35: Perform branch design on the LSTM network structure and adjust the LSTM hidden layer dimension according to the variance ratio of each component; The design scheme is as follows: Trend branch: Deep LSTM (3 layers) processes slow changes; Periodic branch: Bidirectional LSTM captures periodic bidirectional dependencies; Residual branch: Lightweight LSTM (1 layer) combines with the attention mechanism; Step S36: Fuse the three-way outputs of the LSTM network through a learnable weight matrix, and dynamically update the STL decomposition parameters and LSTM weights based on the prediction error. The specific formula for fusing the three-way outputs of the LSTM network is as follows: ; wherein, .

[0014] As a preferred technical solution, in step S4, the specific process of establishing a material water absorption characteristic matrix according to the cultural relic material and dynamically dividing the showcase control groups is as follows: Step S41: Define a basic material library according to the cultural relic type, and calculate the sensitivity coefficient of each material to humidity change with reference to the historical preservation humidity standards of different materials; Step S42: Construct a matrix with the material type, current humidity deviation, and target humidity range as dimensions ; Step S43: Construct the characteristic vector of each showcase ; Step S44: Calculate the similarity between different showcases, and the calculation formula is as follows: ; In the formula, represents the similarity between showcase i and showcase j, represents the characteristic vector of showcase i, represents the characteristic vector of showcase j, represents the broadband parameter, taking 0.5; Step S45: Take the sum of each row element of the similarity matrix S as the diagonal element, and normalize the Laplacian matrix; Step S45: Cluster the dimension-reduced characteristic vectors to divide the showcase groups.

[0015] As a preferred technical solution, in step S5, when constructing the equipment efficiency model, a dynamic database is constructed by collecting equipment energy consumption, processing speed, and fault information in real time through sensors. Based on the historical operation data of the equipment, the future efficiency decay trend of the equipment is predicted through linear regression or LSTM neural network, and the model parameters are dynamically corrected; The specific process of the equipment efficiency model for multi-device task allocation is as follows: Step S51: Convert the equipment efficiency parameters into task allocation costs; Step S52: Change the fixed rule of row / column subtraction in the Hungarian algorithm to be dynamically adjusted according to the real-time efficiency of the equipment; Step S53: Introduce a greedy strategy or Monte Carlo tree search to accelerate the discovery of augmenting paths; Step S54: Input the task set, equipment set, and real-time efficiency parameters into the equipment-task matching scheme model; Step S55: Generate a dynamic cost matrix according to the efficiency model; Step S56: Execute the improved Hungarian algorithm to complete the initial allocation; Step S57: Monitor the execution result and trigger the update of the performance model parameters and the reallocation of the algorithm.

[0016] As a preferred technical solution, in step S6, an embedded microprocessor and a micro flash module are integrated in the sensor node, and a multi-modal sensor is connected through an I²C / SPI interface to form an integrated package structure; when the embedded microprocessor receives data, a quantized regulation model is used for local inference and multi-source data fusion, and the aggregation node completes low-power communication between nodes through the LoRa relative position positioning technology.

[0017] As a preferred technical solution, in step S7, the virtual-real linkage of the constructed museum BIM model is carried out based on the Unity / Unreal engine to construct a digital twin of the exhibition hall; the digital twin of the exhibition hall is used to display the energy consumption, the number of people, and the device health of the museum equipment in real time. At the same time, the administrator can modify the virtual environment parameters through the digital twin of the exhibition hall, preview the renovation effect and then implement physical adjustment.

[0018] The present invention has the following beneficial effects: (1) By combining 3D scanning and BIM, the present invention generates an accurate 3D model of a complex museum. At the same time, a material water absorption characteristic matrix is established according to the material of the cultural relics, the showcase control groups are dynamically divided, and a device performance model is established to realize multi-device task allocation, improve the intelligent regulation of the temperature and humidity in the museum, and timely supervise and maintain the museum equipment.

[0019] (2) By feeding back the LSTM prediction error to the STL parameter adjustment, the present invention forms a dynamic optimization closed loop, and allocates LSTM resources according to the statistical characteristics of trends, cycles, and residuals, and the component drives the network structure to improve the model efficiency.

[0020] (3) The multi-dimensional feature vector composed of the material response coefficient of the showcase, the real-time humidity, and the target humidity range is used in the present invention. Spectral clustering captures the non-linear relationship through the similarity matrix, can identify the group structure of irregular shapes, directly models the relevance between showcases by using the similarity matrix, avoids the complexity of distance calculation in the high-dimensional feature space, improves the flexibility of dynamic adjustment of the showcases, realizes the adaptive optimization of showcase group division with environmental fluctuations, and reduces the control error caused by static division.

[0021] (4) By constructing a device performance model, generating a dynamic cost matrix according to the performance model, and executing the improved Hungarian algorithm to complete the initial allocation; according to the monitoring execution result, triggering the update of the performance model parameters and the reallocation of the algorithm, reducing the energy consumption of the device and improving the task completion time. Of course, it is not necessary for any product implementing the present invention to achieve all the above advantages at the same time. Brief Description of the Drawings

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0023] Figure 1 A humidity control and adjustment module and method for a museum according to the present invention; Figure 2 It is a flowchart for constructing a dynamic LSTM. Detailed Embodiments

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0025] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0026] In order to make the purpose, technical solutions and advantages of the present application clearer, the following Figure 1-2 and embodiments are used to further describe the present application in detail. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

[0027] Please refer to Figure 1 As shown, the present invention is a humidity control and adjustment module and method for a museum, including the following steps: Step S1: Integrate the museum BIM model according to the museum building drawings and the data collected by the indoor drone; Step S2: Deploy high-precision mesh sensors inside the museum, and preprocess the collected data using the sliding window algorithm; Step S3: Use time series decomposition to extract the trend term, periodic term and residual term from the preprocessed data, and dynamically adjust the LSTM network structure; Step S4: Establish a material water absorption characteristic matrix according to the material of the cultural relics, and dynamically divide the showcase control groups; Step S5: Establish an equipment efficiency model, and implement multi-device task allocation through an improved Hungarian algorithm; Step S6: Deploy a micro computing unit at the sensor node to achieve local decision-making; Step S7: Construct a digital twin of the exhibition hall, integrating the physical space and virtual simulation data streams.

[0028] In step S1, the construction process of the museum BIM model is as follows: Step S11: Collect site conditions, architectural drawings, and parametric data of special-shaped components. Use Revit to establish the site terrain surface and set a unified coordinate system; after collecting the architectural drawings, it is necessary to determine the integrity of the drawings. The specific drawings include floor plans, elevation views, sectional views, structural drawings, etc. Ensure that the drawing information is complete, accurate, and clearly marked. Mark the data on the drawings and then import them into the modeling software. Mark key data such as dimensions, elevations, material markings, and door and window positions; after determining that the drawings are complete and correctly marked, use Revit to establish the site terrain surface and set a unified coordinate system to make the coordinate system and spatial position in the software consistent with the actual situation. Step S12: Based on the Revit architectural template, build basic components layer by layer. Generate special-shaped parts using a parametric family library or Dynamo scripts; use drawing tools in the Revit modeling software, such as lines, curves, rectangles, polygons, etc., to draw along the edges and contours of the building on the drawings to outline the basic shape of the building; according to the height information of the building, use the stretching or extrusion function of the Revit modeling software to stretch or extrude the drawn 2D contour along the Z-axis direction to make it a 3D model with a certain height. Refer to the sectional view and elevation view, and gradually add the details of the building, such as doors, windows, balconies, stairs, decorative lines, etc. on the basis of the basic model. Various modeling tools in the modeling software, such as Boolean operations, lofting, chamfering, etc., can be used to create these details. Step S13: Use the MagiCAD plug-in to establish a humidity control system and customize the pipeline layout rules according to the functional areas of the museum; after construction, perform collision detection through multi-disciplinary model merging, generate a collision report, and adjust the pipeline routing, optimizing the net height according to the principle of "small pipes avoid large pipes, non-pressure pipes avoid pressure pipes". Step S14: After the deployment of each exhibition hall of the museum is completed, use a drone to perform 3D scanning to obtain the point cloud data of the internal equipment of the museum. Step S15: Stitch, register, and fuse the point cloud data according to the point cloud data scanned from multiple different acquisition exhibition halls.

[0029] In step S15, the point cloud data obtained by 3D scanning the UAV is combined with the museum BIM model. The combined BIM model is given materials, textures are added, and lighting is set for rendering. According to the actual lighting conditions of the building and the desired effects, appropriate lights are set. The direction and intensity of sunlight can be simulated by setting parallel lights, and the indoor lighting effects can be simulated by adding point lights, spotlights, etc. After completing the model construction, adding material textures, and setting the lights, the rendering operation can be carried out. Rendering parameters such as resolution, rendering quality, and anti-aliasing are set as needed, and then the model is rendered into high-quality pictures or videos to better display the 3D effect of the building. If necessary, lightweight processing can be performed on the as-built model, such as reducing the number of faces and optimizing the texture mapping, etc., to facilitate the later operation and maintenance platform to call. At the same time, an equipment information list is made for the indoor equipment scanned by the UAV, and the operation and maintenance parameters of the equipment (such as equipment model, equipment name, equipment maintenance cycle, etc.) are associated to form a digital asset library. Collision reports, optimization plans, and change records at each stage are sorted out to form a complete BIM result catalog to ensure the consistency between the model and the physical building.

[0030] In step S2, the deployment of the high-precision mesh sensors is divided according to spatial density and hierarchical installation. The spatial density division installs high-precision mesh sensors according to the importance of the museum cultural relics. The spatial density division rules are as follows:

[0031] The core cultural relic area adopts a dual-sensor cross-check mode (the distance between the primary and backup nodes is < 30 cm); dynamic switching of the working node: when the error of the primary node exceeds ±1%RH continuously for 3 times, it will automatically switch.

[0032] The hierarchical installation division is divided into the ground layer, the display cabinet layer, and the top layer for installing high-precision mesh sensors. The hierarchical installation division rules are as follows:

[0033] The high-precision mesh sensors use the LoRaWAN protocol to build a self-healing mesh network (transmission distance up to 200 m), which specifically includes a capacitive polymer film sensor, an infrared counter module, a barometric gradient sensor, and an auxiliary monitoring unit. The capacitive polymer film sensor is used to measure with an accuracy of ±0.5%RH (relative humidity) and ±0.1°C (temperature); the infrared counter module is used to synchronously monitor the crowd density around the display cabinet; the barometric gradient sensor is used for the linkage control of the ventilation system; the auxiliary monitoring unit includes a laser particle counter installed in the display cabinet and a microwave radar deployed at the access control position. The laser particle counter is used to monitor the dust concentration in the display cabinet (accuracy 0.1 μm); the microwave radar is used to count the residence time of personnel.

[0034] When preprocessing the collected data using the sliding window algorithm, it specifically includes: 1. Window construction and data maintenance Define a data window of a fixed size (such as 10 sampling points), store historical data through a circular queue or a ring buffer, ensure that new data overwrites the oldest data, and each time new sampled data is added to the window, the oldest data in the window is synchronously removed to keep the data in the window always the latest sequence.

[0035] When setting the window, the larger the window, the stronger the smoothing effect, but the real-time performance decreases; it is recommended to dynamically adjust according to the signal frequency and interference intensity (such as choosing a small window for high-frequency signals and a large window for low-frequency noise); 2. Pulse interference detection and suppression Perform a threshold judgment on the difference between the new data and the previous valid value (such as setting the maximum allowable deviation Δ). If it exceeds Δ, it is determined as pulse interference, discard the current value and replace it with the window mean or median; after sorting the data in the window and taking the median, it can effectively suppress single-pulse interference (such as when the window size is 5, take the 3rd value after sorting); the threshold Δ is set based on the normal fluctuation range of the signal, usually taking 2 to 3 times the standard deviation of the historical data, or calibrated through experiments; 3. Moving average filtering Calculate the arithmetic mean of the data in the window as the final output, weaken the influence of random noise, and dynamically maintain the window sum through an accumulator to avoid full-sum each time.

[0036] In step S2, outlier detection is also required for the preprocessed data. The specific detection method is as follows: Step S21: Judge whether the preprocessed data is approximately normally distributed through a histogram; If it conforms to the approximate normal distribution, then use the criterion for detection; If it is a skewed distribution, mainly use the box plot method; Step S22: Set the threshold of the criterion, where: upper limit = mean + 3 * standard deviation, lower limit = mean - 3 * standard deviation; regard the data not between the upper and lower limits as outliers; Step S23: Calculate the interquartile range using the box plot method. The specific formula is as follows: ; Then regard the data less than and greater than as outliers.

[0037] When the proportion of detected outliers is less than 5%, directly remove the outlier data; if the proportion of detected outliers is greater than 5%, fill the outliers with the median, mean or adjacent data.

[0038] Please refer to Figure 2 As shown, in step S3, the specific process of dynamically adjusting the LSTM network structure according to the trend term, periodic term, and residual term is as follows: Step S31: Perform inner-loop decomposition on the preprocessed data to extract the trend term, periodic term, and residual term; When extracting the trend term, perform LOESS smoothing on the original sequence to extract the initial trend term; when extracting the periodic term, separate the periodic components from the detrended sequence; when extracting the residual term, residual = original sequence - periodic term - seasonal term; Step S32: Dynamically adjust the LOESS window size according to the slope change rate of the trend term. When the slope increases, shrink the window to improve sensitivity; Step S33: Calculate the variance of the periodic component. If the intensity is lower than the threshold, close the periodic branch; Step S34: Use the processed trend term, periodic term, and residual term as independent input channels; Step S35: Design branches for the LSTM network structure and adjust the LSTM hidden layer dimension according to the variance ratio of each component; The design scheme is as follows: Trend branch: Deep LSTM (3 layers) processes slow changes; Periodic branch: Bidirectional LSTM captures periodic bidirectional dependencies; Residual branch: Lightweight LSTM (1 layer) combined with an attention mechanism; Step S36: Fuse the three-way outputs of the LSTM network through a learnable weight matrix, and dynamically update the STL decomposition parameters and LSTM weights based on the prediction error. The specific formula for fusing the three-way outputs of the LSTM network is as follows: ; where .

[0039] In step S4, the specific process of establishing a material water absorption characteristic matrix based on the cultural relic material and dynamically dividing the display case control groups is as follows: Step S41: Define a basic material library according to the cultural relic type (metal, paper, textile, etc.), refer to the historical preservation humidity standards of different materials (such as 0 - 40%RH for metal, 40 - 50%RH for calligraphy and painting), and calculate the sensitivity coefficient of each material to humidity changes (such as 0.8 for metal, 1.2 for calligraphy and painting, and 1.5 for textile), reflecting the sensitivity of the material to humidity; Step S42: Construct a matrix with the material type, current humidity deviation, and target humidity range as dimensions, and dynamically update the coefficient to adapt to environmental fluctuations; Step S43: Construct the feature vector of each display case , the material response coefficient is used to reflect the sensitivity of the material to humidity; the real-time humidity is used to dynamically adjust the clustering weight; the span of the target humidity range reflects the strictness of the control requirement. Step S44: Calculate the similarity between different display cabinets. The calculation formula is as follows: ; In the formula, represents the similarity between display cabinet i and display cabinet j, represents the feature vector of display cabinet i, represents the feature vector of display cabinet j, represents the broadband parameter, which can be set to 0.5 through cross-validation or empirical values, or can be dynamically adjusted according to the humidity fluctuation range value, increasing when the fluctuation is large to enhance fault tolerance; Step S45: Use the diagonal elements as the sum of each row element of the similarity matrix S, ‌, and normalize the Laplacian matrix. The specific formula is as follows: ; Step S45: Cluster the feature vectors after dimensionality reduction to divide the display cabinet groups. When the humidity control error within the group exceeds the threshold (such as MAE > 5%RH), recalculate the feature vectors when adding / removing display cabinets.

[0040] In step S5, when constructing the equipment efficiency model, the equipment energy consumption (kW·h / task), processing speed (tasks / hour), and failure information (failure rate) are collected in real time through sensors to build a dynamic database. Based on the equipment historical operation data (such as the energy consumption curve of the constant humidity system, the temperature control record of the display cabinet), the future efficiency decay trend of the equipment is predicted through linear regression or LSTM neural network, and the model parameters are dynamically corrected; The specific process of the equipment efficiency model for multi-device task allocation is as follows: Step S51: Convert the equipment efficiency parameters into task allocation costs; Step S52: Change the fixed rule of row / column subtraction in the Hungarian algorithm to be dynamically adjusted according to the real-time efficiency of the equipment; Step S53: Introduce a greedy strategy or Monte Carlo tree search to accelerate the discovery of augmenting paths; Step S54: Input the task set (such as the humidity control requirements of the display cabinet), the equipment set (such as constant humidity machines, local humidifiers), and the real-time efficiency parameters ‌ into the equipment-task matching scheme model; Step S55: Generate a dynamic cost matrix according to the efficiency model; Step S56: Execute the improved Hungarian algorithm to complete the initial allocation; Step S57: Monitor the execution results, trigger the update of the performance model parameters and algorithm reallocation, achieve efficient allocation of device resources, and at the same time ensure system stability and energy consumption economy.

[0041] In step S6, an embedded microprocessor (ARM Cortex-M series or RISC-V architecture) and a micro flash module are integrated in the sensor node, and multi-modal sensors (temperature and humidity, acceleration, optical sensors) are connected through I²C / SPI interfaces to form an integrated package structure (size ≤ 3 cm × 3 cm), supporting hot-pluggable deployment. A quantized lightweight AI model (such as MobileNet compressed to 50 - 200 KB) is used for local inference, and Kalman filtering and D-S evidence theory are combined to achieve multi-source data fusion, with a decision response time ≤ 10 ms‌; through dynamic voltage and frequency scaling (DVFS) and task scheduling algorithms (improved Hungarian algorithm or ant lion optimization algorithm), calculation tasks are dynamically allocated according to the remaining energy of the node, and the typical power consumption is reduced by 60% - 80%‌; a hierarchical topology structure is constructed, the edge nodes perform local threshold judgment and simple event response, and the aggregation nodes complete complex pattern recognition, and low-power communication between nodes is achieved through LoRa relative position positioning technology.

[0042] In step S7, based on the Unity / Unreal engine, the virtual-real interaction of the constructed museum BIM model is carried out, and a digital twin of the exhibition hall is constructed to support two-way data driving between the physical space and the virtual model; the digital twin of the exhibition hall can also be embedded with a predictive maintenance module, and the operation data of the equipment is analyzed through an LSTM neural network to predict the fault risk 72 hours in advance. The digital twin of the exhibition hall is used to display more than 20 core indicators such as the energy consumption, number of visitors, and equipment health of the museum equipment in real time‌. At the same time, the administrator can modify the virtual environment parameters through the digital twin of the exhibition hall and preview the renovation effect before implementing physical adjustments. It should be noted that in the above system embodiments, the various units included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0043] In addition, those of ordinary skill in the art can understand that all or part of the steps in the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0044] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A humidity control and adjustment module and method for a museum, characterized in that, It includes the following steps: Step S1: Integrate the museum BIM model based on the museum building drawings and the data collected by indoor drones; Step S2: Deploy high-precision mesh sensors inside the museum, and use the sliding window algorithm to preprocess the collected data; Step S3: Use time series decomposition to extract the trend term, periodic term, and residual term from the preprocessed data, and dynamically adjust the LSTM network structure; Step S4: Establish a material water absorption characteristic matrix according to the material of the cultural relics, and dynamically divide the showcase control groups; Step S5: Establish an equipment efficiency model, and implement multi-device task allocation through an improved Hungarian algorithm; Step S6: Deploy a micro computing unit at the sensor node to achieve local decision-making; Step S7: Build a digital twin of the exhibition hall, and integrate the physical space and virtual simulation data streams.

2. The humidity control and adjustment module and adjustment method for a museum according to claim 1, characterized in that, In the said Step S1, the construction process of the museum BIM model is as follows: Step S11: Collect site conditions, building drawings, and parametric data of special-shaped structures, establish the site terrain surface through Revit, and set a unified coordinate system; Step S12: Based on the Revit architectural template, layer by layer establish basic components, and generate the special-shaped parts using a parametric family library or Dynamo scripts; Step S13: Use the MagiCAD plug-in to establish a humidity control system, and customize the pipeline layout rules according to the museum's functional areas; Step S14: After the deployment of each exhibition hall of the museum is completed, use drones to perform 3D scanning to obtain the point cloud data of the internal equipment of the museum; Step S15: Stitch, register, and fuse the point cloud data according to the point cloud data scanned by multiple different acquisition exhibition halls.

3. The humidity control and adjustment module and adjustment method for a museum according to claim 2, characterized in that In the said Step S15, combine the point cloud data obtained by the 3D scanning of the drone with the museum BIM model, endow the combined BIM model with materials, add textures and lighting for rendering, and at the same time make a construction equipment information list for the indoor equipment scanned by the drone, and associate the operation and maintenance parameters of the equipment to form a digital asset library.

4. A humidity control and adjustment module and adjustment method for a museum according to claim 1, characterized in that, In the said Step S2, the deployment of the high-precision mesh sensors is divided according to spatial density and hierarchical installation; the spatial density division installs high-precision mesh sensors according to the importance of the museum cultural relics; the hierarchical installation division is divided into the ground layer, showcase layer, and top layer to install high-precision mesh sensors; the high-precision mesh sensors use the LoRaWAN protocol to build a self-healing mesh network, which specifically includes a capacitive polymer film sensor, an infrared counter module, a barometric gradient sensor, and an auxiliary monitoring unit; the auxiliary monitoring unit includes a laser particle counter installed in the showcase and a microwave radar deployed at the access control position.

5. A humidity control and adjustment module and adjustment method for a museum according to claim 1, characterized in that, In the said Step S2, the preprocessed data also needs to be subjected to outlier detection. The specific detection method is as follows: Step S21: Judge whether the data is approximately normally distributed through a histogram for the preprocessed data; If it conforms to the quasi-normal distribution, use the criterion for detection; If it is skewed, mainly use the box plot method; Step S22: Set the threshold of the criterion, where: upper limit = mean + 3 * standard deviation, lower limit = mean - 3 * standard deviation; data not between the upper and lower limits are regarded as outliers; Step S23: Calculate the interquartile range using the box plot method. The specific formula is as follows: ； will be less than and greater than data as outliers.

6. A humidity control and adjustment module and adjustment method for a museum according to claim 1, characterized in that, In the said Step S3, the specific process of dynamically adjusting the LSTM network structure according to the trend term, periodic term, and residual term is as follows: Step S31: Perform inner-loop decomposition on the preprocessed data to extract the trend term, periodic term, and residual term; Step S32: Dynamically adjust the LOESS window size according to the slope change rate of the trend term; Step S33: Calculate the variance of the periodic component. If the intensity is lower than the threshold, close the periodic branch; Step S34: Use the processed trend term, periodic term, and residual term as independent input channels; Step S35: Design branches for the LSTM network structure and adjust the dimension of the LSTM hidden layer according to the variance ratio of each component; Step S36: Fuse the three-way outputs of the LSTM network through a learnable weight matrix, and dynamically update the STL decomposition parameters and LSTM weights based on the prediction error.

7. A humidity control and adjustment module and adjustment method for a museum according to claim 1, characterized in that, In step S4, the specific process of establishing a material water absorption characteristic matrix according to the cultural relic material and dynamically dividing the showcase control group is as follows: Step S41: Define a basic material library according to the cultural relic type, and calculate the sensitivity coefficient of each material to humidity change with reference to the historical preservation humidity standards of different materials; Step S42: Construct a matrix with material type, current humidity deviation, and target humidity range as dimensions ; Step S43: Construct the feature vector of each display cabinet ; Step S44: Calculate the similarity between different showcases. The calculation formula is as follows: ； In the formula, represents the similarity between showcase i and showcase j, represents the feature vector of showcase i, represents the feature vector of showcase j, represents the broadband parameter, taking 0.5; Step S45: Take the sum of each row element of the similarity matrix S with diagonal elements as the sum, and normalize the Laplacian matrix; Step S45: Cluster the dimension-reduced feature vectors to divide the showcase groups.

8. A humidity control and adjustment module and adjustment method for a museum according to claim 1, characterized in that, In step S5, when constructing the equipment efficiency model, a dynamic database is constructed by collecting equipment energy consumption, processing speed, and fault information in real time through sensors. Based on the historical operation data of the equipment, the future efficiency decay trend of the equipment is predicted through linear regression or LSTM neural network, and the model parameters are dynamically corrected; The specific process of the equipment efficiency model for multi-device task allocation is as follows: Step S51: Convert the equipment efficiency parameters into task allocation costs; Step S52: Change the fixed rule of row / column subtraction in the Hungarian algorithm to be dynamically adjusted according to the real-time efficiency of the equipment; Step S53: Introduce a greedy strategy or Monte Carlo tree search to accelerate the discovery of augmenting paths; Step S54: Input the task set, equipment set, and real-time efficiency parameters into the equipment-task matching scheme model; Step S55: Generate a dynamic cost matrix according to the efficiency model; Step S56: Execute the improved Hungarian algorithm to complete the initial allocation; Step S57: Monitor the execution result and trigger the update of the efficiency model parameters and algorithm reallocation.

9. A humidity control and adjustment module and adjustment method for a museum according to claim 1, characterized in that In step S6, an embedded microprocessor and a micro flash module are integrated in the sensor node, and a multi-modal sensor is connected through the I²C / SPI interface to form an integrated package structure; when the embedded microprocessor receives data, a quantized regulation model is used for local inference and multi-source data fusion, and the aggregation node completes low-power communication between nodes through LoRa relative position positioning technology.

10. A humidity control and adjustment module and adjustment method for a museum according to claim 1, characterized in that In step S7, based on the Unity / Unreal engine, the virtual-real linkage of the constructed museum BIM model is carried out to construct a digital twin of the exhibition hall; the digital twin of the exhibition hall is used to display the energy consumption, number of visitors, and equipment health of the museum in real time. At the same time, the administrator can modify the virtual environment parameters through the digital twin of the exhibition hall, preview the renovation effect and then implement physical adjustments.

Images