An intelligent control system for temperature and humidity of an archive room

Abstract

The application relates to the technical field of intelligent regulation and control of an archive storage environment, and discloses an intelligent control system for the temperature and humidity of an archive room. The system comprises an environment monitoring network component, an internal detection engine component, a fusion computing center component and a decision generator component. The environment monitoring network component collects environment data by using a sensor array and generates a high-precision environment state atlas by applying a data assimilation technique. The internal detection engine component reconstructs an internal non-uniform humidity field of archives based on risk identification in the atlas by using a waveform analysis algorithm. The fusion computing center component aligns the time and space attributes of the two, and outputs a comprehensive confidence by using an evidence reasoning method. The decision generator component generates an adaptive control strategy by using a heuristic optimization algorithm according to the confidence. The application realizes all-round accurate perception and intelligent regulation and control from the macro environment of the archive room to the micro environment of the archive entity.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for archival preservation environment, specifically to an intelligent control system for temperature and humidity in archival storage rooms. Background Technology

[0002] Currently, temperature and humidity control in archive storage primarily relies on discretely deployed temperature and humidity sensors within the storage space. Real-time monitoring data is compared with preset thresholds to directly control the operation of air conditioning, humidification, or dehumidification equipment. This approach relies on data from a limited number of points, and its completeness and accuracy are limited by the number and layout of sensors. Errors caused by spatial interpolation make it difficult for the system to accurately reflect the temperature and humidity field distribution throughout the entire storage space, especially failing to effectively identify abnormal conditions in localized areas such as near doors and windows, and in areas where air conditioning systems are ineffective.

[0003] Existing control systems cannot handle complex scenarios such as conflicting monitoring data from multiple areas or dynamic changes in environmental parameters, easily leading to frequent or contradictory control command switching. This not only affects equipment lifespan but also makes it difficult to maintain a stable preservation environment. Current technology equates the control objective with maintaining stable air parameters, completely ignoring the humidity state of the archival materials themselves. The humidity exchange between archival materials and the surrounding air is a dynamic equilibrium process. The internal humidity distribution of archival materials is influenced by multiple factors, including material properties, bulk density, and the airtightness of the storage containers, resulting in significant differences from the air environment.

[0004] Currently, there is a lack of effective means to non-invasively monitor and assess the humidity field inside archival materials. Physical models based on the assumption of uniformity yield calculations that differ significantly from reality, failing to provide early warnings of risks such as mold and embrittlement caused by internal humidity gradients. The core challenge of this invention is to achieve integrated and accurate perception of both the macro-environment of the storage facility and the micro-environment within the archives, and to develop data-driven intelligent control decisions based on this perception. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent temperature and humidity control system for archive storage rooms to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent temperature and humidity control system for an archive storage room, the system comprising:

[0007] An environmental monitoring network component uses a sensor array to collect environmental datasets from an archive, and applies data assimilation techniques to generate an environmental status map based on the environmental datasets.

[0008] The internal detection engine component, based on the risk identification of the environmental state map, reconstructs the humidity field inside the archive through waveform analysis algorithms;

[0009] By integrating the components of the computing center, aligning the spatiotemporal attributes of the environmental state map and the internal humidity field, and using evidence-based reasoning methods, a comprehensive confidence score is output.

[0010] The decision generator component uses heuristic optimization algorithms to generate control strategies based on comprehensive confidence levels.

[0011] Preferably, the method for the environmental monitoring network component to collect environmental parameters includes:

[0012] Multiple types of sensors, including temperature sensors, humidity sensors, and air quality sensors, are deployed in the archive storage room to record spatial point data at fixed time intervals.

[0013] Cluster analysis algorithms are used to identify high- and low-fluctuation areas of environmental parameters and to dynamically prioritize monitoring.

[0014] Adjust the sensor's operating mode according to priority: increase sampling density in high fluctuation areas and decrease sampling density in low fluctuation areas.

[0015] The original data is filtered to remove noise interference and generate an environmental dataset.

[0016] Preferably, the method for generating an environmental status map by the environmental monitoring network component includes:

[0017] Normalize the environmental dataset to unify the data scale;

[0018] A three-dimensional model of the warehouse space is established, and the space is discretized into a regular voxel grid. The Kriging interpolation algorithm is used to calculate the estimated environmental parameters of each voxel to generate a continuously distributed environmental state map. The Gaussian smoothing algorithm is applied to optimize the distribution continuity and output the environmental state map.

[0019] Preferably, the method for risk identification by the environmental monitoring network component includes:

[0020] Define the normal range, warning range, and danger range of environmental parameters; for each voxel value in the environmental state map, compare the parameter ranges and mark the voxel as a green safe zone, a yellow observation zone, or a red danger zone; generate a risk coding map based on the marking results, where different colors represent risk levels.

[0021] Preferably, the method for reconstructing the internal humidity field of the archive using the internal detection engine component includes:

[0022] Based on the risk coding map, infrared thermal imager arrays are deployed in the red danger zone and yellow observation zone; infrared beams are emitted onto the surface of the archive to measure the surface temperature distribution; the internal humidity value is inverted using a heat conduction model, and the humidity gradient is calculated by combining the material's moisture absorption characteristics; a two-dimensional humidity distribution map is generated through image reconstruction technology and mapped onto three-dimensional space to form an internal humidity field.

[0023] Preferably, the method of using a thermal conduction model to invert internal humidity values ​​and combining this with the material's hygroscopic properties to calculate the humidity gradient includes:

[0024] A differential equation for heat conduction of the archival material is established, with surface temperature distribution set as the boundary condition. The gradient descent algorithm is used to iteratively solve the differential equation for heat conduction to obtain an estimate of the internal temperature distribution. Based on the estimate of the internal temperature distribution and the material's moisture absorption isotherm model, the internal relative humidity value is calculated. Based on the discrete point data of the internal relative humidity value in three-dimensional space, the central difference method is used to calculate the humidity gradient vector of each point.

[0025] Preferably, the method for generating a two-dimensional humidity distribution map through image reconstruction technology and mapping it to three-dimensional space to form an internal humidity field includes:

[0026] Multi-angle surface humidity sampling points were acquired by an infrared thermal imager array, and a two-dimensional humidity distribution image was reconstructed using a filtered back projection algorithm. The two-dimensional humidity distribution image was registered with the three-dimensional coordinate system of the archive, and the image pixels were mapped to a three-dimensional voxel grid through perspective transformation. The humidity values ​​in the voxel grid were smoothed using a trilinear interpolation algorithm to generate a continuous three-dimensional internal humidity field.

[0027] Preferably, the step of the fusion computing center component outputting the comprehensive confidence level using the evidence reasoning method includes:

[0028] The environmental state map and internal humidity field are converted into basic probability assignment functions. The hypothesis space is set to include normal state, abnormal state and critical state. The Dempster combination rule is applied to fuse the probability assignments and calculate the joint confidence degree. The maximum posterior probability is extracted from the joint confidence degree as the comprehensive confidence degree.

[0029] Preferably, the method for setting the hypothesis space in the fusion computing center component includes:

[0030] For the environmental state map, the evidence support is defined based on the color regions of the risk coding map; for the internal humidity field, the evidence support is defined based on the degree to which the humidity value deviates from the baseline; and the support weights are calibrated using training data to optimize the accuracy of probability allocation.

[0031] Preferably, the method by which the decision generator component generates a control strategy includes:

[0032] The comprehensive confidence level is input into the multi-objective optimization framework, and the control objectives are set as risk minimization, stability maximization, and energy efficiency optimization. The simulated annealing algorithm is used to search for the optimal solution of control parameters, including temperature setpoint, humidity setpoint, and equipment switching sequence. The optimal parameter solution sequence is converted into executable control instructions.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] The application of data assimilation technology fuses sensor observation data from sparse locations with the physical model prediction field of the warehouse environment to generate a continuously distributed environmental state map. This method changes the coarse perception mode of traditional technology that directly equates discrete point data with the overall environmental state. The environmental state map can present the detailed spatial distribution of temperature and humidity in a probabilistic form, accurately identifying local high-risk areas that are ignored between sensor locations.

[0035] The waveform analysis algorithm achieves non-invasive reconstruction of the humidity field inside the archive by extracting and inverting features of abnormal fluctuation signals in the environmental state spectrum. This algorithm can analyze signal attenuation and distortion caused by factors such as material hygroscopicity and the sealing of the mounting, thereby quantifying the actual humidity at different depths inside the archive. This directly reveals the true humid and thermal state of the archive, overcoming the limitations of indirect assessment relying solely on air parameters.

[0036] The evidence-based reasoning method addresses the spatiotemporal correlation between environmental maps and internal humidity fields, transforming macroscopic environmental anomalies and microscopic entity risks into fusionable evidence of uncertainty. Through synthesis rules, this method resolves decision-making conflicts that may arise from differences in accuracy and timeliness of multi-source data. This shifts the decision-making basis from a single environmental parameter exceeding limits to a unified risk probability assessment supported by multi-dimensional evidence, enhancing the comprehensiveness and reliability of the judgment.

[0037] Based on comprehensive confidence levels, a heuristic optimization algorithm dynamically searches under multiple constraints to generate an adaptive control strategy. This strategy does not execute pre-set fixed instructions, but intelligently adjusts the priority and execution parameters of the control objective based on real-time assessments of risk level, spatial range, and internal conditions. This enables the system to cope with complex scenarios such as changes in warehouse usage conditions and localized emergencies, achieving a shift from rigid control based on simple rules to flexible and precise control based on risk assessment. This optimizes system operating efficiency while ensuring the safety of the storage environment. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent temperature and humidity control system for archive storage as described in this invention.

[0039] Figure 2 A flowchart of the environmental parameter acquisition method;

[0040] Figure 3 A flowchart of the risk identification method;

[0041] Figure 4Box plot of conflict coefficients for each stage of the intelligent temperature and humidity control system for archive storage;

[0042] Figure 5 This is a chart analyzing the distribution of humidity deviation in archives. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 This invention provides an intelligent temperature and humidity control system for an archive storage facility. The system includes an environmental monitoring network component, an internal detection engine component, a fusion computing center component, and a decision generator component. The environmental monitoring network component uses a sensor array to collect environmental datasets from the archive storage facility and applies data assimilation technology to generate an environmental state map, which represents the temperature and humidity distribution in three-dimensional space. The internal detection engine component reconstructs the internal humidity field of the archives based on risk indicators in the environmental state map using waveform analysis algorithms, thereby detecting humidity changes within the archive materials. The fusion computing center component aligns the spatiotemporal attributes of the environmental state map and the internal humidity field, and outputs a comprehensive confidence score using an evidence-based reasoning method. This confidence score reflects a reliable assessment of the environmental state. The decision generator component uses a heuristic optimization algorithm based on the comprehensive confidence score to generate control strategies, dynamically adjusting the parameters of equipment such as air conditioners and dehumidifiers to ensure a stable environment in the archive storage facility.

[0045] Example 1: See Figure 2 In practice, the deployment and operation of the environmental monitoring network components form the foundation for system data acquisition. A sensor array composed of various sensors, including temperature, humidity, and air quality sensors, is deployed within the archive storage room. Specifically, digital temperature sensors and capacitive humidity sensors are used, while the air quality sensor detects carbon dioxide concentration and volatile organic compound (VOC) content. The sensor array records spatial point data at fixed time intervals, configurable to 5, 10, or 15 minutes. Each spatial point data record includes a timestamp, three-dimensional spatial coordinates, and measured temperature, humidity, and air quality index values. The spatial point data collection covers the three-dimensional space of the archive storage room, with key areas including the interior of the archive equipment, passageways, and areas near air conditioning vents.

[0046] In some embodiments, a clustering analysis algorithm is used to identify high-fluctuation and low-fluctuation areas of environmental parameters. The clustering analysis algorithm employs the K-means algorithm. The algorithm processes spatial point data from a sensor array, treating each sampling point as a data object. The feature vector of each data object consists of temperature readings, humidity readings, and air quality index readings. The K-means algorithm randomly selects K points as initial cluster centers, iteratively assigning each data point to the nearest cluster center and recalculating the cluster centers until they no longer change significantly. Dynamically prioritizing monitoring is based on the clustering results: high-fluctuation areas correspond to clusters with larger data point variances, while low-fluctuation areas correspond to clusters with smaller data point variances. It can be understood that high-fluctuation areas are typically located near warehouse doors and windows or in areas with frequent personnel activity, while low-fluctuation areas are typically located in warehouse corners or enclosed cabinets.

[0047] In practice, adjusting the sensor's operating mode according to priority is a dynamic configuration process. In areas of high fluctuation, sampling density is increased, specifically by shortening the fixed time interval, for example, from 10 minutes to 2 minutes. In areas of low fluctuation, sampling density is decreased, specifically by extending the fixed time interval, for example, from 10 minutes to 30 minutes. The adjustment of the sensor's operating mode is accomplished by sending commands from the central controller to the sensor nodes. The sensor nodes have multiple power consumption modes to adapt to different sampling density requirements. The raw data is filtered using digital filters, which can be finite-length impulse response (FTR) filters or infinite-length impulse response (IRF) filters. Filtering removes noise interference, primarily from electromagnetic interference or sensor drift. The resulting environmental dataset is a collection of filtered, time- and space-stamped temperature, humidity, and air quality data.

[0048] In practice, the first step in generating an environmental state map is to normalize the environmental dataset using a min-max normalization method. Unifying the data scale involves mapping temperature, humidity, and air quality data to the [0,1] interval. Temperature data normalization uses the minimum and maximum temperatures allowed by the warehouse design as the range boundaries; humidity data normalization uses the minimum and maximum relative humidity allowed by the warehouse design as the range boundaries; and air quality data normalization uses the safe pollutant concentration threshold as the range boundaries. The normalized environmental dataset facilitates subsequent interpolation calculations.

[0049] In some embodiments, the creation of a 3D model of the warehouse space relies on Building Information Modeling (BIM) data or 3D laser scanning point cloud data, discretizing the space into a regular voxel grid. The size of the regular voxel grid is determined according to the warehouse size and accuracy requirements, with a common voxel size of 10 cm × 10 cm × 10 cm. Each voxel contains a 3D coordinate index for recording environmental parameter values. The Kriging interpolation algorithm is used to calculate the estimated environmental parameters for each voxel. Kriging interpolation is a spatial interpolation method based on a variogram. First, the Kriging interpolation algorithm calculates the semivariogram between sampling points in the environmental dataset, which describes the spatial autocorrelation of the data. Then, the Kriging interpolation algorithm uses the semivariogram to fit a theoretical model, which can be a spherical model, an exponential model, or a Gaussian model. Finally, the Kriging interpolation algorithm calculates weights for each voxel location based on the distance between the voxel and surrounding sampling points and the semivariogram model. The estimated environmental parameters for the voxel are obtained by weighted averaging of the sampling point values. The resulting continuously distributed environmental state map is a 3D data field, in which each voxel stores estimated temperature, humidity, and air quality index values.

[0050] It is understandable that applying the Gaussian smoothing algorithm to optimize the continuity of the distribution is a post-processing step. The Gaussian smoothing algorithm uses a Gaussian kernel function to perform a convolution operation on the 3D data field. The size and standard deviation of the Gaussian kernel function affect the smoothing degree; a larger standard deviation produces a smoother effect. The Gaussian smoothing algorithm iterates through each voxel, calculating the weighted average of its neighborhood voxels, with the weights determined by the Gaussian function. The output environmental state map is a smoothed 3D matrix, where each element corresponds to the environmental parameter value of a voxel. The environmental state map is stored in the system database as a digital file for use by the internal detection engine components. The update frequency of the environmental state map is consistent with the acquisition frequency of the environmental dataset, ensuring that the system can reflect real-time changes in the warehouse environment.

[0051] In practical implementation, the hardware architecture of the environmental monitoring network component includes sensor nodes, aggregation nodes, and a central server. Sensor nodes are responsible for collecting raw data; they are typically battery-powered and have wireless communication capabilities. Aggregation nodes receive data from multiple sensor nodes and are usually located in the center of the storage area to ensure communication quality. The central server runs clustering, filtering, interpolation, and smoothing algorithms and has sufficient computing power to process three-dimensional spatial data. The software component of the environmental monitoring network component includes a data acquisition module, a data processing module, and a data storage module. The data acquisition module controls the sensor operating rhythm, the data processing module executes the aforementioned algorithm processes, and the data storage module saves the environmental dataset and environmental status map to a database.

[0052] Example 2: See Figure 3 In its implementation, the risk identification process is based on the environmental state map. Defining the normal, warning, and hazardous ranges of environmental parameters forms the foundation of risk identification. These ranges are set according to archival protection standards: the normal temperature range is set at 18°C ​​to 22°C, the warning range at 22°C to 24°C, and the hazardous range at temperatures above 24°C or below 18°C; the normal humidity range is set at 40% to 50% relative humidity, the warning range at 50% to 55% relative humidity, and the hazardous range at temperatures above 55% or below 40% relative humidity. For each voxel value in the environmental state map, comparing parameter ranges is an automated process. The system reads the temperature and humidity values ​​stored for each voxel in the three-dimensional matrix of the environmental state map, comparing the temperature value with the normal, warning, and hazardous ranges, and comparing the humidity value with the normal, warning, and hazardous ranges.

[0053] In practice, voxels are marked as green safe zones, yellow observation zones, or red danger zones according to logical judgment rules. If both the temperature and humidity values ​​of a voxel are within the normal range, it is marked as a green safe zone; if either the temperature or humidity value is within the warning range, it is marked as a yellow observation zone; and if either the temperature or humidity value is within the danger range, it is marked as a red danger zone. A risk coding map is generated based on the marking results. The risk coding map is a three-dimensional matrix with the same dimensions as the environmental state map. Each element in the risk coding map matrix stores a color code value, represented by numbers, such as 0 for green, 1 for yellow, and 2 for red. Different colors in the risk coding map represent risk levels: green for safe risk, yellow for observable risk, and red for dangerous risk. After the risk coding map is generated, a spatial index relationship is established between it and the environmental state map.

[0054] In some embodiments, the process of reconstructing the internal humidity field of the archive by the internal detection engine component is initiated after the risk coding map is generated. Based on the risk coding map, an infrared thermal imager array is deployed in the red danger zone and the yellow observation zone. The placement and number of the infrared thermal imager array are determined by the risk coding map. The system identifies all voxel clusters marked in red and yellow in the risk coding map and deploys infrared thermal imagers above the corresponding three-dimensional spatial positions of the clusters. The infrared thermal imager array is deployed in a network, with each infrared thermal imager equipped with a pan-tilt mechanism that can adjust the elevation and horizontal rotation angles of the infrared thermal imager. The infrared thermal imager operates by emitting an infrared beam onto the archive surface. The wavelength of the infrared beam is selected from the mid-wave infrared band, with a wavelength range between 3 and 5 micrometers. The surface temperature distribution is measured by the infrared detector of the infrared thermal imager. The infrared detector converts the received infrared radiation energy into an electrical signal. After amplification and analog-to-digital conversion, the electrical signal forms surface temperature distribution data, which is represented in the form of a two-dimensional thermal image.

[0055] It is understandable that using a heat conduction model to retrieve internal humidity values ​​is a computationally intensive process. This model, based on Fourier's law, considers the thermal conductivity, specific heat capacity, and density parameters of the archival materials. Calculating the humidity gradient by combining the material's hygroscopic properties requires hygroscopic isotherm data of the archival materials, which describe the relationship between the material's equilibrium moisture content and the ambient relative humidity. Generating a two-dimensional humidity distribution map using image reconstruction technology involves multiple steps. This image reconstruction technique employs a filtered back-projection algorithm, which reconstructs a two-dimensional cross-sectional image from surface temperature projection data from multiple angles. Mapping the two-dimensional humidity distribution map to three-dimensional space requires coordinate transformation, the coordinate transformation matrix of which is calculated using the intrinsic and extrinsic parameters of the infrared thermal imager. The final result of forming the internal humidity field is a three-dimensional scalar field. Each voxel in this three-dimensional scalar field stores a relative humidity value, and the internal humidity field has the same spatial range and voxel resolution as the risk coding map.

[0056] In practice, the calibration of the infrared thermal imager array is completed before measurement. The calibration process uses a blackbody radiation source as a standard temperature reference. The blackbody radiation source is located in the archive storage room, and its surface temperature is known and stable. The infrared thermal imager array captures thermal images of the blackbody radiation source, and a temperature correction curve is established based on the difference between the grayscale values ​​of the thermal images and the known temperature. When measuring the surface temperature distribution, the infrared thermal imager array is simultaneously triggered, acquiring multiple thermal images of the same archive surface from different angles. These multiple thermal images are aligned using an image registration algorithm, which extracts feature points from the thermal images and calculates the transformation matrix between the images through feature point matching. The registered multiple thermal images are then fused using a weighted average method to generate a high-precision surface temperature distribution map.

[0057] Optionally, the internal humidity value is derived from the heat conduction model using a numerical solution. The partial differential equation of heat conduction is discretized into a difference equation, which is then iteratively solved on a three-dimensional voxel grid. Boundary conditions are set to measured values ​​from the surface temperature distribution map, and initial conditions are set to the average temperature of the warehouse environment. The iterative solution uses the conjugate gradient method, which solves a large-scale sparse linear system of equations to obtain an estimated temperature value for each voxel within the archive. Based on the estimated internal temperature, the internal relative humidity value is calculated using the material's hygroscopic properties, described by pre-measured adsorption isotherms, which are fitted to the GAB model formula. The humidity gradient is calculated using the central difference method, which calculates the rate of change of humidity for each voxel in the X, Y, and Z directions on a three-dimensional grid. These rates of change constitute the humidity gradient vector. The magnitude of the humidity gradient vector indicates the drastic change in humidity, and the direction of the humidity gradient vector points towards the direction of the fastest increase in humidity.

[0058] In some embodiments, the specific implementation of generating a two-dimensional humidity distribution map through image reconstruction technology relies on a filtered back-projection algorithm, which processes multi-angle surface humidity sampling points acquired by an infrared thermal imager array. The filtered back-projection algorithm first filters the projection data for each angle using a Ram-Lak filter, which suppresses high-frequency noise in the frequency domain. The filtered projection data is then back-projected, tracing each projection value back to the image space along a ray path. All back-projection paths are superimposed to form a two-dimensional humidity distribution map. Mapping the two-dimensional humidity distribution map to three-dimensional space involves a coordinate system: the archive's three-dimensional coordinate system is the world coordinate system, and the infrared thermal imager's image coordinate system is the camera coordinate system. Perspective transformation establishes the correspondence between image pixels and three-dimensional voxels. The perspective transformation matrix is ​​obtained through camera calibration, which uses a calibration plate of known size. A trilinear interpolation algorithm is used to smooth the humidity values ​​in the voxel grid. Trilinear interpolation is based on the humidity values ​​of eight neighboring voxels around a voxel, and the humidity value of the current voxel is calculated through linear weighting. The final generated internal humidity field is stored in system memory as a three-dimensional array. The internal humidity field data format is 32-bit floating-point numbers, with each floating-point number representing the relative humidity percentage value at a voxel location. The internal humidity field and the environmental state map share the same spatial reference frame, enabling the fusion computing center component to directly perform data alignment and fusion operations.

[0059] Example 3: In specific implementation, establishing the thermal conductivity differential equation of the archival material is a fundamental step in retrieving the internal humidity value. The thermal conductivity differential equation of the archival material is constructed based on the theory of unsteady-state thermal conduction. The thermal conductivity differential equation of the archival material considers the anisotropic thermophysical properties of the material, and the variables in the equation include spatial coordinates and time variables. The surface temperature distribution is set as the boundary condition, which is provided by the surface temperature distribution map obtained by the infrared thermal imager array. The surface temperature distribution map is a two-dimensional array that changes with time. The thermal conductivity differential equation of the archival material can be expressed as:

[0060] ;

[0061] Where: symbol Indicates the density of archival materials, symbol The symbol represents the specific heat capacity at constant pressure of archival materials. Represents the temperature field, symbol Indicates time, symbol The thermal conductivity tensor of archival materials, operator Represents the spatial gradient operator, operator This represents the divergence operator.

[0062] In some embodiments, a gradient descent algorithm is used to iteratively solve the heat conduction differential equation. Gradient descent is an optimization algorithm used to minimize an objective function. The objective function is defined as the mean square error between the calculated surface temperature and the measured surface temperature. The gradient descent algorithm starts with an initial guess of the internal temperature distribution, typically set as a uniform temperature field. In each iteration, the gradient descent algorithm calculates the gradient of the objective function with respect to the internal temperature distribution, indicating the direction of the steepest descent of the objective function. Then, the gradient descent algorithm updates the estimated internal temperature distribution along the negative gradient direction, with the update step size controlled by the learning rate parameter. The iterative process continues until the mean square error is less than a preset threshold or the maximum number of iterations is reached, ultimately yielding a stable estimate of the internal temperature distribution.

[0063] It is understandable that calculating the internal relative humidity value based on the estimated internal temperature distribution combined with the material's hygroscopic isotherm model is a physicochemical process. The material's hygroscopic isotherm model describes the relationship between the equilibrium moisture content of the archival material and the relative humidity of the environment at a certain temperature. During the calculation, the estimated internal temperature distribution is used as input, combined with the material's equilibrium moisture content data, and the relative humidity value at each voxel location is obtained through table lookup or interpolation. Based on the discrete point data of the internal relative humidity value in three-dimensional space, the discrete point data corresponds to the nodes of the voxel grid. The central difference method is used to calculate the humidity gradient vector at each point. The central difference method is a numerical differentiation method. For points inside the grid, the central difference method uses the humidity values ​​of six adjacent points to calculate the partial derivatives in the X, Y, and Z directions. For boundary points, forward difference or backward difference methods are used. The three calculated partial derivatives constitute the humidity gradient vector, which reflects the rate and direction of humidity change in space.

[0064] In practical implementation, generating a two-dimensional humidity distribution map through image reconstruction technology requires processing multi-angle surface humidity sampling points acquired by an infrared thermal imager array. These sampling points originate from measurement data from infrared thermal imagers at different orientations, and each sampling point contains two-dimensional image coordinates and a humidity value. A filtered back-projection algorithm, belonging to the tomographic reconstruction algorithm family, is used to reconstruct the two-dimensional humidity distribution map. The algorithm first filters the humidity data for each projection angle using a ramp filter convolution kernel. The filtered projection data is then used to reconstruct the image through a back-projection operation, which uniformly distributes the values ​​on each projection ray to the corresponding paths in the image space. The back-projection results from all angles are superimposed to form the two-dimensional humidity distribution map, which is in grayscale image format, with pixel values ​​representing relative humidity.

[0065] Optionally, registering the two-dimensional humidity distribution image with the three-dimensional coordinate system of the archive involves spatial transformation. The registration process is achieved through feature point matching, setting markers with known three-dimensional coordinates on the archive surface. The two-dimensional coordinates of these markers are identified in the infrared thermal image, establishing the correspondence between the two-dimensional image coordinates and the three-dimensional world coordinates. Perspective transformation maps image pixels to a three-dimensional voxel grid; the perspective transformation matrix consists of the camera's intrinsic and extrinsic parameter matrices. Perspective transformation maps each pixel in the image to its corresponding position in the three-dimensional voxel grid, taking into account the camera's perspective distortion. A trilinear interpolation algorithm is used to smooth the humidity values ​​in the voxel grid. This algorithm calculates the values ​​of internal points based on the values ​​of the eight vertices of the cube. For each voxel, its corresponding cube cell is found, and a weighted average is calculated based on the humidity values ​​of its eight adjacent vertices. The weights are determined by the distance from the voxel to each vertex; closer vertices have larger weights. A continuous three-dimensional internal humidity field is generated; this field is a three-dimensional array of floating-point numbers.

[0066] In some embodiments, the material parameters for the heat conduction model inversion process need to be accurately determined. The density of the archival material is measured by the gravimetric method. The isobaric specific heat capacity of the archival material is obtained using differential scanning calorimetry, and the thermal conductivity of the archival material is measured using the heat flow meter method. The parameters of the material hygroscopic isotherm model are determined by the static weighing method, where archival material samples are placed in constant temperature environments with different humidity levels to measure the equilibrium moisture content. The implementation of the gradient descent algorithm needs to consider computational efficiency, and a stochastic gradient descent algorithm is used to accelerate convergence. In each iteration of the stochastic gradient descent algorithm, a subset of surface temperature points are randomly selected to calculate the objective function, reducing the computational load. The learning rate parameter adopts an adaptive adjustment strategy, with a larger initial learning rate to accelerate convergence and a smaller learning rate in later stages to improve accuracy. The grid resolution of the internal temperature distribution estimate is matched to the spatial resolution of the infrared thermal imager, typically setting the voxel size to be smaller than the minimum resolvable size of the infrared thermal imager. The calculation results of the humidity gradient vector are stored in a sparse matrix format, storing only non-zero gradient values ​​to reduce storage space. The visualization of the three-dimensional internal humidity field uses isosurface plotting technology, where isosurface plotting displays the spatial distribution surface of a specific humidity value. The entire internal humidity field reconstruction process is performed in parallel on the graphics processing unit (GPU), leveraging its parallel architecture to accelerate large-scale data processing. The reconstructed 3D internal humidity field data is then transmitted via network to the fusion computing center components for further processing.

[0067] Example 4: In specific implementation, the process of the integrated computing center component using the evidence reasoning method to output the comprehensive confidence level begins in the data preparation stage. The core step is to convert the environmental state map and internal humidity field into a basic probability assignment function. The environmental state map is a three-dimensional data field containing temperature, humidity, and air quality parameters, while the internal humidity field is a three-dimensional data field describing the relative humidity distribution within the archival materials. Converting the basic probability assignment function requires defining an identification framework, which includes a set of mutually exclusive and complete propositions that the system may be in. The hypothesis space is defined to include three basic propositions: normal state, abnormal state, and critical state. A normal state indicates that the environmental parameters fully meet the archival preservation requirements; an abnormal state indicates that the environmental parameters exceed the safe range and pose a threat to the archives; and a critical state indicates that the environmental parameters are on the edge of safety and require close monitoring. A basic probability assignment function is generated for each voxel data point in the environmental state map. This basic probability assignment function assigns a probability mass to each proposition, with the probability mass ranging from 0 to 1, and the sum of the probability masses of all propositions equals 1. A basic probability assignment function is also generated for each voxel of data in the internal humidity field. The basic probability assignment function of the internal humidity field calculates the probability mass based on the degree to which the humidity value deviates from the standard value.

[0068] In some embodiments, applying the Dempster combination rule to fuse probability assignment involves the computational synthesis of multiple evidence sources. The Dempster combination rule is a fundamental method in evidence theory for fusing multiple independent pieces of evidence. An environmental state map provides the first set of evidence sources, and an internal humidity field provides the second set. These two sets of evidence sources are fused after spatiotemporal alignment. The Dempster combination rule calculates the joint confidence level through orthogonal summation. Orthogonal summation first calculates the conflict coefficient, reflecting the degree of evidence conflict. The conflict coefficient represents the degree to which different evidence sources completely contradict each other regarding the support for the proposition set. Then, the Dempster combination rule proportionally redistributes the conflict coefficients to each proposition, resulting in a normalized joint basic probability assignment function. The maximum posterior probability is extracted from the joint confidence level as the comprehensive confidence level. The maximum posterior probability is the probability value corresponding to the proposition with the highest probability quality in the joint basic probability assignment function. The comprehensive confidence level is a value between 0 and 1; a higher value indicates a higher credibility of the proposition being true. The comprehensive confidence level has the same spatial distribution as the environmental state map and the internal humidity field, forming a three-dimensional confidence field that is output to the decision generator component.

[0069] In practical implementation, converting the environmental state map into a basic probability assignment function requires determining the membership function. The membership function defines the degree of support each environmental parameter value provides for the three propositions. The membership function for temperature is a trapezoidal function, for humidity it is a trigonometric function, and for air quality it is a Gaussian function. Refer to Table 1, which uses humidity as an example to illustrate the conversion rules for basic probability assignment.

[0070] Table 1: Basic Probability Allocation and Conversion Rules for Humidity Parameters

[0071]

[0072] It is understandable that the conversion of the internal humidity field into a basic probability allocation function employs a similar principle but with different parameter thresholds. The basic probability allocation function for the internal humidity field considers differences in material properties. The safe humidity range for paper archival materials differs from that for film archival materials, requiring adjustments to the probability allocation parameters based on the archival type. When applying Dempster's combination rule to fuse probability allocations, potential conflicts between the environmental state map and the internal humidity field need to be addressed. A conflict resolution mechanism is activated when the conflict coefficient exceeds a preset threshold. This conflict resolution mechanism is based on evidence reliability weighting, with reliability weights obtained through training with historical data. The evidence reliability weight for the environmental state map is typically higher than that for the internal humidity field because the measurement data for the environmental state map is more direct and accurate.

[0073] In its implementation, the Dempster combination rule is calculated using matrix operations, representing the basic probability allocation function of each source of evidence as a probability vector. This probability vector contains four elements: the first three elements correspond to the probability allocation values ​​for the normal, abnormal, and critical states, respectively, and the fourth element corresponds to the probability allocation value for uncertainty. The Dempster combination rule constructs a joint probability matrix, which is the result of the outer product of two probability vectors. Then, the conflict coefficient is calculated, which is the sum of the probability values ​​of all completely conflicting proposition pairs in the joint probability matrix. The conflict coefficient is calculated as the sum of the probabilities of all evidence combinations that do not support the same proposition. An excessively large conflict coefficient indicates significant contradictions among the sources of evidence requiring special handling. The conflict coefficient is processed using a normalization method, proportionally redistributing the conflict coefficient to non-conflicting proposition combinations. The normalization factor is 1 minus the conflict coefficient. The final joint basic probability allocation function contains the updated probability values ​​of the three propositions and an uncertainty value.

[0074] Optionally, a comparison-ranking algorithm is used to extract the maximum posterior probability from the joint confidence score. This algorithm compares the probability values ​​of the three propositions—normal state, abnormal state, and critical state—in the joint basic probability assignment function. The proposition with the highest probability value is selected as the best estimate of the current state, and its probability value is output as the comprehensive confidence score. When the probability values ​​of two propositions are very close, a conservative decision-making strategy is activated. This strategy always selects the proposition with the higher risk as the output. For example, when the probability value of the abnormal state is 0.48 and the probability value of the normal state is 0.47, even though the probability value of the normal state is slightly lower, the system will choose the abnormal state as the final proposition because the abnormal state represents a higher risk and needs to be prioritized. The spatial distribution of the comprehensive confidence score is consistent with the original environmental state map. Each voxel location has a corresponding comprehensive confidence score value, forming a three-dimensional confidence score distribution field. The comprehensive confidence score data is stored in a floating-point array format, with the array dimensions being exactly the same as the environmental state map, facilitating direct access by subsequent decision-making components.

[0075] Optionally, the real-time performance of the evidence reasoning method is ensured through parallel computing technology, with the fusion computing center component equipped with a graphics processing unit (GPU) for accelerated computation. The GPU simultaneously processes the basic probability allocation calculations for multiple voxel locations, dividing the 3D data field into multiple computational blocks for parallel processing. Matrix operations in the Dempster combination rules utilize the GPU's parallel thread architecture for high-speed computation, with evidence fusion calculations for a single voxel completed within a single thread block. The update frequency of the overall confidence score is synchronized with the sampling frequency of the environmental monitoring network component, typically set to update every 5-10 minutes to ensure the system can respond promptly to environmental changes. The fusion computing center component also includes an evidence quality assessment module, which monitors the integrity and consistency of the input data. When data quality does not meet requirements, the evidence weight is reduced or re-collection of data is requested. The final output overall confidence score includes not only numerical results but also an additional credibility index. This credibility index reflects the degree of conflict and data quality during the evidence fusion process, providing additional reference information for the decision generator component.

[0076] See Figure 4 The box plot illustrates the distribution characteristics of conflict coefficients during the evidence reasoning process in the six stages of the intelligent temperature and humidity control system for the archive storage: the initial stage, monitoring stage 1, monitoring stage 2, fusion stage 1, fusion stage 2, and decision-making stage. The boxes in the box plot reflect the quartile range of the conflict coefficients at each stage: the lower edge represents the first quartile (25th quartile), the middle line of the box represents the second quartile (median, 50th quartile), and the upper edge represents the third quartile (75th quartile). Whiskers show the range of values ​​excluding outliers, with scatter points representing outliers. As observed in the figure, the upper edges of the boxes in fusion stages 1 and 2 are relatively high, indicating that the upper quartiles of the conflict coefficients in these two stages are at a high level, and the overall degree of conflict between the evidence is relatively high. The boxes in the decision stage are generally lower, indicating that the median and quartile ranges of the conflict coefficients in this stage are smaller, and the degree of conflict during evidence fusion is relatively lower. The distribution of outliers in each stage reflects the dispersion of the conflict coefficients. For example, fusion stage 1 has low outliers, while fusion stage 2 has high outliers, reflecting the individual differences in evidence conflict within different stages. These distribution characteristics can provide data support for the system's evidence reliability assessment and conflict resolution mechanism optimization at different stages, helping to improve the accuracy of comprehensive confidence calculation and thus optimize the generation of temperature and humidity control strategies.

[0077] Example 5: In specific implementation, the method for setting the hypothesis space in the fusion computing center component begins with the structured definition of input evidence. The evidence support for defining the environmental state map depends on the color region information of the risk coding map. The risk coding map is a three-dimensional color-coded map generated by the environmental monitoring network component, where green safe zones indicate environmental parameters are within the normal range, yellow observation zones indicate environmental parameters are within the warning range, and red danger zones indicate environmental parameters are within the danger range. The quantification of evidence support is based on the spatial distribution characteristics of the color regions, and the system assigns an initial support weight to each color region. For example, in a specific archive storage monitoring example, the risk coding map shows that the northeast corner area is a red danger zone, accounting for approximately 15% of the total storage area; the central area is a yellow observation zone, accounting for approximately 25%; and the remaining areas are green safe zones, accounting for approximately 60%. For this specific example, the system assigns a higher abnormal state support weight to the red danger zone, a higher critical state support weight to the yellow observation zone, and a higher normal state support weight to the green safe zone.

[0078] In some embodiments, defining the evidence support level for an internal humidity field requires establishing a deviation calculation model between the humidity value and a baseline value, where the baseline value is determined based on the optimal storage humidity for the archival materials. For example, the baseline humidity for paper archives is set to 45%RH, and for film archives, it is set to 30%RH. The degree of deviation is quantified by calculating the relative deviation, which is equal to the absolute difference between the measured humidity value and the baseline humidity value divided by the baseline humidity value. In a specific application, when it is detected that a certain area contains paper archives, the system calls the baseline humidity of 45%RH for paper archives. If the internal humidity field shows that the humidity value of that area is 55%RH, then the relative deviation is (55-45) / 45=0.22, or 22%. Based on this degree of deviation, the system calculates the evidence support level according to a preset mapping relationship; the greater the degree of deviation, the higher the support weight for the abnormal state. It is understood that different archival materials have different humidity sensitivities, and the definition of evidence support level needs to consider the differences in material characteristics.

[0079] In practice, calibrating the support weights using training data is a crucial step in optimizing the accuracy of probability allocation. The training data comes from historical monitoring records in the archives. These records include environmental condition maps, internal humidity field data, and records of actual archive damage events. The calibration process employs the maximum likelihood estimation method, which adjusts the support weight parameters to maximize the consistency between the model's predictions and actual observations. For example, the system collected daily environmental condition maps, internal humidity field data, and weekly archive condition inspection reports from the past six months, obtaining 180 sets of valid training samples. These training samples included records of three actual discovered archive mold events. The environmental condition maps for these events all showed a red danger zone ratio exceeding 20%, and the internal humidity field showed an average humidity deviation exceeding 25%. Based on this training data, the system automatically adjusted the support weight for red danger zones to abnormal states from an initial value of 0.7 to 0.85, and simultaneously adjusted the support weight for high humidity deviation to abnormal states from 0.6 to 0.78.

[0080] Optionally, the calibration process for the support weights is optimized using the gradient descent algorithm. The gradient descent algorithm iteratively adjusts the weight parameters to minimize the loss function. The loss function is defined as the cross-entropy between the predicted and actual states. In each iteration, the gradient of the loss function with respect to the weight parameters is calculated, and the parameters are updated along the negative gradient direction. The learning rate is set to 0.01, and the weight parameters converge to a stable value after 1000 iterations. The accuracy of the optimized probability assignment is verified using a test dataset containing 30 sets of historical data that were not used in training. The degree of agreement between the predicted and actual states is calculated as the evaluation metric. A degree of agreement of 90% or higher is considered to meet the accuracy requirements for probability assignment; otherwise, the support weights need to be recalibrated.

[0081] Optionally, the definition of evidence support also considers the time factor, assigning higher weight to recent data. The system introduces a time decay factor, which decreases exponentially. The weighting coefficient for data from the most recent seven days is 1.0, for data from the eighth to the fourteenth day it is 0.7, for data from the fifteenth to the thirtieth day it is 0.5, and data older than thirty days is not included in the support calculation. This time-weighted mechanism ensures that the evidence support reflects the latest environmental conditions, improving the timeliness of the status assessment. In a specific example, the western area of ​​the warehouse experienced abnormal humidity last week. Although it has returned to normal this week, due to the time decay factor, the recent abnormality in this area will still have a significant impact on the current support calculation.

[0082] In practice, the definition of evidence support for different color regions employs fuzzy logic, which addresses the uncertainty of color boundary regions. The boundaries of color regions in the risk coding diagram are not absolutely clear, exhibiting transition zones. The system defines a membership function for each color region: a trapezoidal function for the green safe zone, a triangular function for the yellow observation zone, and a trapezoidal function for the red danger zone. For example, a voxel at a color boundary may simultaneously belong to both the green safe zone and the yellow observation zone, possessing a green membership of 0.4 and a yellow membership of 0.6. This membership relationship is considered when calculating evidence support, with a weighted average of the support weights for each color. This approach improves the accuracy of the evidence support definition and avoids abrupt changes in boundary regions.

[0083] Optionally, training data quality control is achieved through a data cleaning process, which removes invalid data and outliers. Invalid data includes sensor fault records and data gaps during communication interruptions, while outliers refer to data points that clearly deviate from physical laws, such as humidity values ​​exceeding 100% RH. After data cleaning, high-quality training samples are retained to ensure the reliability of support weight calibration. Training data is updated regularly, with new data added to the training set monthly while older data older than three months is discarded to maintain the model's adaptability and accuracy. The support weight calibration cycle is set to once a week, with the system automatically executing the calibration procedure during off-peak business periods, without affecting normal monitoring operations.

[0084] In practice, the accuracy of probability assignment is evaluated using a combination of metrics, including precision, recall, and F1 score. Precision assesses the proportion of actually anomalous states among those predicted as anomalous; recall assesses the proportion of correctly predicted states among those actually anomalous; and the F1 score is the harmonic mean of precision and recall. The system sets the accuracy standards for probability assignment at a precision of at least 85%, a recall of at least 80%, and an F1 score of at least 0.82. A weekly probability assignment accuracy evaluation report is generated, and an alert is triggered when any metric falls below the standard value, indicating the need to check the quality of the training data or adjust the support weight definition rules. Through this continuous evaluation and optimization mechanism, the accuracy of evidence support definition and probability assignment is ensured to remain at a high level.

[0085] See Figure 5This study reveals the relative humidity deviation distribution and overall characteristics of different types of archives in the archive storage facility, providing crucial data support for intelligent temperature and humidity control of archives. Subplot a shows box plots of the relative humidity deviation for paper archives and film archives: the box plots visually present the location and dispersion of data through boxes, whiskers, and outliers. Specifically, the relative humidity deviation boxes for paper archives are concentrated in the lower range, with a median of approximately 10%, indicating that their overall deviation from the baseline value (45%RH) is relatively small, with low data dispersion and only a few cases of higher deviation. The relative humidity deviation boxes for film archives shift significantly upwards, with a median of approximately 43%, and outliers (approximately 90%), indicating that film archives (baseline humidity 30%RH) have a significantly greater humidity deviation and more drastic data fluctuations. This is closely related to the differences in humidity sensitivity between film and paper archives and the hygroscopic characteristics of their materials, reflecting the different humidity control requirements of the two types of archives in the storage environment. Subplot b is a histogram showing the overall distribution of relative humidity deviation: with relative deviation on the x-axis and frequency on the y-axis, combined with the red dashed line representing the average (24.0%), the distribution characteristics of overall humidity deviation can be seen. Most frequencies are concentrated in the low deviation range (0-20%), but there is a certain proportion of high deviation cases. The average of 24.0% reflects the overall deviation level. This provides a data foundation for the integrated computing center component to assess the comprehensive confidence level of environmental conditions and internal humidity fields, and also provides a basis for the decision generator component to consider the overall deviation degree and distribution pattern when formulating temperature and humidity control strategies. In summary, through the combination of box plots and histograms, the statistical characteristics of relative humidity deviation in archives are clearly presented from two dimensions: classification comparison and overall distribution. This provides intuitive and professional data analysis support for risk identification, evidence reasoning, and control strategy generation in the intelligent temperature and humidity control system for archive storage, helping to achieve precise control of the archive storage environment based on data-driven principles.

[0086] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent temperature and humidity control system for an archive room, characterized in that, The system comprises: An environmental monitoring network component that collects an environmental dataset of the archive using a sensor array and applies a data assimilation technique to generate an environmental state map based on the environmental dataset; An internal detection engine component that reconstructs an internal humidity field of the archive based on risk identification of the environmental state map through a waveform analysis algorithm; A fusion computing center component that aligns the spatiotemporal attributes of the environmental state map and the internal humidity field and outputs a comprehensive confidence using an evidential reasoning method, including: Converting the environmental state map and the internal humidity field into basic probability assignment functions, setting a hypothesis space including a normal state, an abnormal state, and a critical state; applying a Dempster combination rule to fuse the probability assignments and calculate a joint trust degree; and extracting a maximum a posteriori probability from the joint trust degree as the comprehensive confidence; A decision generator component that generates a control strategy using a heuristic optimization algorithm based on the comprehensive confidence.

2. The intelligent temperature and humidity control system for archives warehouse according to claim 1, characterized in that, The method for collecting environmental parameters by the environmental monitoring network component comprises: Deploying multiple types of sensors, including temperature sensors, humidity sensors, and air quality sensors, inside the archive room to record spatial point data at fixed time intervals; Using a clustering analysis algorithm to identify high-volatility and low-volatility regions of environmental parameters and dynamically dividing monitoring priorities; Adjusting sensor operating modes according to priorities, increasing sampling density in high-volatility regions and reducing sampling density in low-volatility regions; Filtering raw data to remove noise interference and generating an environmental dataset.

3. The intelligent temperature and humidity control system for archives warehouse according to claim 2, characterized in that, The method for generating an environmental state map by the environmental monitoring network component comprises: Normalizing the environmental dataset to unify data scales; Establishing a three-dimensional model of the room space, discretizing the space into a regular voxel grid; using a Kriging interpolation algorithm to calculate environmental parameter estimates for each voxel and generating a continuously distributed environmental state map; and applying a Gaussian smoothing algorithm to optimize distribution continuity and output the environmental state map.

4. The intelligent temperature and humidity control system for archives warehouse according to claim 3, characterized in that, The method for risk identification by the environmental monitoring network component comprises: Defining normal, warning, and danger ranges of environmental parameters; comparing parameter ranges for each voxel value in the environmental state map to mark the voxel as a green safe zone, a yellow observation zone, or a red danger zone; and generating a risk coding map based on the marking results, where different colors represent risk levels.

5. The intelligent temperature and humidity control system for archives warehouse according to claim 4, characterized in that, The method for reconstructing an internal humidity field of the archive by the internal detection engine component comprises: According to the risk coding map, arranging an infrared thermal imager array in the red danger zone and the yellow observation zone; emitting an infrared light beam to the archive surface to measure surface temperature distribution; using a heat conduction model to invert internal humidity values, combining material moisture absorption characteristics to calculate humidity gradients; generating a two-dimensional humidity distribution map through image reconstruction technology and mapping it to a three-dimensional space to form an internal humidity field.

6. The intelligent temperature and humidity control system for archives warehouse according to claim 5, characterized in that, The method for using a heat conduction model to invert internal humidity values, combining material moisture absorption characteristics to calculate humidity gradients comprises: A heat conduction differential equation of the archive material is established, and a surface temperature distribution is set as a boundary condition; a gradient descent algorithm is used to iteratively solve the heat conduction differential equation to obtain an internal temperature distribution estimate; according to the internal temperature distribution estimate, in combination with a material moisture absorption isotherm model, an internal relative humidity value is calculated; and based on discrete point data of the internal relative humidity value in a three-dimensional space, a central difference method is used to calculate a humidity gradient vector of each point.

7. The intelligent temperature and humidity control system for archives warehouse according to claim 6, characterized in that, The method for generating a two-dimensional humidity distribution map through an image reconstruction technique and mapping to a three-dimensional space to form an internal humidity field comprises: Multi-angle surface humidity sampling points acquired by an infrared thermal imager array are collected, a filtered back-projection algorithm is used to reconstruct a two-dimensional humidity distribution image; the two-dimensional humidity distribution image is registered with a three-dimensional coordinate system of the archive room, and image pixels are mapped to a three-dimensional voxel grid through perspective transformation; a trilinear interpolation algorithm is used to smooth the humidity values in the voxel grid to generate a continuous three-dimensional internal humidity field.

8. The intelligent temperature and humidity control system for archives warehouse according to claim 7, characterized in that, The method for setting a hypothetical space in the fusion computing center component comprises: For the environmental state atlas, an evidence support degree is defined according to a color region of a risk coding map; for the internal humidity field, an evidence support degree is defined according to a degree of deviation of a humidity value from a benchmark; support degree weights are calibrated through training data to optimize the accuracy of probability allocation.

9. The intelligent temperature and humidity control system for archives warehouse according to claim 8, characterized in that, The method for the decision generator component to generate a control strategy comprises: The comprehensive confidence is input into a multi-objective optimization framework, control objectives are set as risk minimization, stability maximization and energy efficiency optimization; a simulated annealing algorithm is used to search for optimal solutions of control parameters, including temperature set values, humidity set values and device switching time sequences; and the optimal parameter solution sequence is converted into executable control instructions.

Images