Method and system for environmental intelligence monitoring for meat processing

Abstract

This application relates to the field of environmental monitoring technology, and discloses an intelligent environmental monitoring method and system for meat processing. The method includes: collecting environmental parameters through a sensor array within the workshop to construct a real-time environmental parameter matrix; generating a continuous spatial fluctuation field using spatial interpolation, and constructing a fluctuation correlation graph reflecting spatial correlation based on this field; identifying pollution source nodes and propagation paths based on this correlation graph; simulating pollutant diffusion trajectories using a graph convolutional network in conjunction with ventilation data, and determining abnormal diffusion boundaries by comparing with measured data; and finally, dynamically adjusting the sampling density and spatial distribution of the sensor array based on these boundaries to achieve adaptive optimization of the monitoring layout. This invention achieves closed-loop intelligent monitoring from perception and prediction to resource scheduling, significantly improving the monitoring accuracy and early warning capabilities for abnormal diffusion of microorganisms and pollutants in meat processing workshops.

Description

Technical Field

[0001] This application relates to the field of environmental monitoring technology, and in particular to an intelligent environmental monitoring method and system for meat processing. Background Technology

[0002] In the meat processing industry, the cleanliness of the production workshop is a core element in ensuring food safety and preventing microbial contamination and cross-infection. Airborne microorganisms, volatile organic compounds, and other pollutants can easily spread within the workshop with the airflow. If monitoring is not timely, it can lead to the expansion of regional contamination and even cause batch product safety issues.

[0003] Currently, the industry commonly uses a method of deploying sensor networks at fixed locations within the workshop to monitor key parameters such as temperature, humidity, and microbial concentration. However, this traditional static monitoring method has significant shortcomings: First, the fixed layout of the sensor network cannot adaptively adjust according to the actual diffusion path of pollutants and the dynamic changes in the ventilation system, leading to potential monitoring blind spots in high-risk areas and resource redundancy in low-risk areas. Second, existing systems can only provide monitoring data at discrete points, lacking the ability to analyze the continuous spatial distribution of environmental parameters throughout the workshop and the correlation of pollution between areas, making it difficult to accurately locate pollution sources and predict their diffusion trends. Finally, when local parameter anomalies are detected, the system cannot intelligently determine the possible impact range of pollutants, thus failing to guide the accurate and efficient reallocation of monitoring resources.

[0004] Therefore, how to overcome the limitations of static monitoring and construct an intelligent monitoring method that can dynamically perceive the correlation of pollution diffusion, intelligently predict the propagation path, and adaptively optimize the allocation of monitoring resources has become a key technical problem that urgently needs to be solved to improve the proactive defense capability of environmental safety in meat processing workshops. Summary of the Invention

[0005] This application aims to overcome the shortcomings of the prior art and provides an intelligent environmental monitoring method and system for meat processing. Through intelligent technical means, it realizes dynamic perception, intelligent prediction and adaptive optimization of monitoring resources for the spread of pollutants in the workshop, thereby improving the accuracy and efficiency of environmental monitoring in meat processing workshops.

[0006] In a first aspect, this application provides an intelligent environmental monitoring method for meat processing, the method comprising:

[0007] Step 1: Collect environmental parameter data in real time by deploying a sensor array in the meat processing workshop and construct a real-time environmental parameter matrix;

[0008] Step 2: Based on the real-time environmental parameter matrix, apply spatial interpolation algorithms to complete and smooth the data of irregularly distributed points to obtain a continuous spatial fluctuation field;

[0009] Step 3: Divide the region based on the continuous spatial wave field, calculate the correlation coefficient matrix between adjacent regions, and generate a wave correlation diagram accordingly;

[0010] Step 4: In the fluctuation correlation diagram, regions with correlation coefficients exceeding a preset reference threshold are marked as correlation source nodes. The initial propagation path set starting from the correlation source nodes is obtained through a path search algorithm. The correlation source nodes are potential sources of pollutants or other environmental changes.

[0011] Step 5: Combine the initial set of propagation paths with ventilation flow data, and apply graph convolutional networks to simulate the diffusion dynamics of microorganisms and / or pollutants in the workshop, thereby obtaining the predicted propagation trajectory;

[0012] Step 6: Compare the predicted propagation trajectory with the real-time environmental parameter matrix, update the edge weights in the fluctuation correlation graph according to the differences, and determine the abnormal propagation boundary based on the updated correlation graph.

[0013] Step 7: Based on the abnormal diffusion boundary, dynamically adjust the spatial distribution of the sensor array in the workshop and optimize the dot matrix layout configuration.

[0014] Secondly, this application provides an intelligent environmental monitoring system for meat processing, the system comprising:

[0015] The matrix construction module is used to collect environmental parameter data in real time through sensor arrays deployed in meat processing workshops and construct a real-time environmental parameter matrix.

[0016] The preprocessing module is used to complete and smooth the data of irregularly distributed points by applying spatial interpolation algorithms based on the real-time environmental parameter matrix, so as to obtain a continuous spatial fluctuation field.

[0017] The region division module is used to divide regions based on the continuous spatial fluctuation field, calculate the correlation coefficient matrix between adjacent regions, and generate a fluctuation correlation diagram accordingly.

[0018] The path generation module is used to mark regions with correlation coefficients exceeding a preset reference threshold as associated source nodes in the fluctuation correlation diagram, and to obtain an initial set of propagation paths starting from the associated source nodes through a path search algorithm. The associated source nodes are potential sources of pollutants or other environmental changes.

[0019] The trajectory prediction module combines the initial set of propagation paths with ventilation flow data, applies graph convolutional networks to simulate the diffusion dynamics of microorganisms and / or pollutants in the workshop, and then obtains the predicted propagation trajectory.

[0020] The boundary determination module is used to compare the predicted propagation trajectory with the real-time environmental parameter matrix, update the edge weights in the fluctuation correlation graph according to the differences, and determine the abnormal propagation boundary based on the updated correlation graph.

[0021] The configuration generation module is used to dynamically adjust the spatial distribution of the sensor array in the workshop based on the abnormal diffusion boundary, and optimize the dot matrix layout configuration.

[0022] Compared with the prior art, the beneficial effects of the technical solution of this application are at least as follows:

[0023] 1. By using spatial interpolation algorithms, discrete and sparse sensor data are transformed into a continuous spatial distribution of overall environmental parameters (such as microbial concentration) in the workshop, realizing a qualitative change in monitoring cognition from "discrete points" to "continuous fields," and providing a high-quality data foundation for accurate analysis.

[0024] 2. By constructing a fluctuation correlation graph and an application graph convolutional network, dynamic correlation analysis and intelligent prediction of pollutant diffusion were achieved. This can quantify the intensity of pollution correlation between spatial areas and simulate its diffusion path and trend under the influence of ventilation, thereby transforming the monitoring mode from passively receiving alarms to actively predicting and warning, gaining critical time for early intervention;

[0025] 3. This invention is not a static solution; its core lies in a closed-loop feedback mechanism, forming a complete intelligent closed loop of "perception-prediction-decision-optimization." By comparing the prediction results with real-time monitoring data, it can automatically detect cognitive biases, update the model, redefine anomaly boundaries, and ultimately drive adaptive adjustments to the sensor network layout. This allows monitoring resources to dynamically focus on the highest-risk area, significantly improving resource utilization efficiency.

[0026] 4. In summary, this invention enables earlier detection, more accurate source tracing, more reliable prediction, and more efficient monitoring of abnormal spread of microorganisms and pollutants within the workshop. Through dynamic optimization of the layout, it continuously ensures high-precision coverage of key risk areas, fundamentally enhancing the proactive defense capabilities against cross-contamination and ensuring product safety, and significantly improving the proactive environmental safety defense level of meat processing workshops. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram illustrating the overall principle of the intelligent environmental monitoring method of this application;

[0029] Figure 2 This is a flowchart of the intelligent environmental monitoring method for meat processing according to this application;

[0030] Figure 3 This is a schematic diagram of irregularly distributed points in an embodiment of this application;

[0031] Figure 4 This is a schematic diagram of the Kriging space interpolation result in an embodiment of this application;

[0032] Figure 5 This is a schematic diagram of the three-dimensional spatial mesh region division in an embodiment of this application;

[0033] Figure 6 This is a schematic diagram of the significant correlation coefficient matrix in an embodiment of this application;

[0034] Figure 7 This is a schematic diagram of the trajectory intensity distribution and calibration area according to an embodiment of this application;

[0035] Figure 8 This is a schematic diagram of the structure of the intelligent environmental monitoring system for meat processing according to this application. Detailed Implementation

[0036] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0037] like Figure 1The diagram shown illustrates the overall principle of the intelligent environmental monitoring method proposed in this application. This method is designed for the monitoring and closed-loop optimization control of abnormal diffusion of microorganisms and / or pollutants within meat processing workshops. The overall process includes environmental data acquisition, spatial state reconstruction, propagation analysis and prediction, abnormal boundary correction, and adaptive optimization allocation of monitoring resources.

[0038] Specifically, temperature and humidity sensors, microbial sensors, and / or pollutant sensors are deployed within the workshop space to collect environmental parameters at each monitoring location, and a real-time environmental parameter matrix is ​​constructed according to spatial location and time series. Based on this real-time environmental parameter matrix, spatial interpolation and continuous field reconstruction are performed on the discrete monitoring point data to obtain a spatial fluctuation field that characterizes the overall environmental state of the workshop. Further, the workshop can be divided into regions based on the spatial fluctuation field, and a fluctuation correlation network between regions can be established by combining the correlation of environmental parameter changes between adjacent regions. After forming the fluctuation correlation network, potential pollution source nodes are identified by combining the degree of regional anomaly, and a set of possible initial propagation paths is extracted; simultaneously, ventilation flow information within the workshop is introduced, coupling the driving effect of ventilation direction and wind speed on inter-regional diffusion with the fluctuation correlation network. Based on this, a graph convolutional network is applied to simulate the dynamic diffusion process of microorganisms and / or pollutants in the workshop regional network, obtaining predicted propagation trajectories with temporal sequence and spatial range information. After obtaining the predicted propagation trajectories, the prediction results are compared with real-time monitoring results to identify regions where there are significant deviations between prediction and measurement, and the abnormal diffusion boundaries are corrected accordingly, forming a more accurate characterization of the current abnormal diffusion range. Furthermore, based on the abnormal diffusion boundary and its internal high-variability sections, high-risk and low-risk areas are identified, and the spatial distribution of the sensor array in the workshop is adaptively optimized and adjusted so that monitoring resources are concentrated in high-risk areas while taking into account the basic monitoring needs of low-risk areas, thereby forming an optimized array layout configuration for the current diffusion situation.

[0039] therefore, Figure 1 The overall diagram reflects the closed-loop intelligent monitoring process of this application, from sensing workshop environmental parameters, constructing a spatial continuous field, analyzing regional propagation relationships, predicting diffusion trends, to correcting abnormal diffusion boundaries and dynamically optimizing the configuration of sensor arrays.

[0040] For ease of understanding, the specific process of the embodiments of this application is described below. Figure 2 The diagram shows a flowchart of the intelligent environmental monitoring method for meat processing provided by the present invention. The flowchart specifically includes the following steps:

[0041] Step 1: Collect environmental parameter data in real time by deploying a sensor array in the meat processing workshop and construct a real-time environmental parameter matrix.

[0042] In one specific embodiment, the process of performing step 1 may specifically include the following steps:

[0043] A sensor array is deployed in the workshop based on a three-dimensional spatial grid, and the sensor nodes collect temperature and humidity data, microbial concentration data, and pollutant concentration data in real time.

[0044] The collected data is organized using three-dimensional spatial coordinates as an index to construct an initial environmental parameter matrix;

[0045] For the initial environmental parameter matrix, a time series smoothing algorithm is applied to remove noise and obtain a smoothed environmental parameter matrix;

[0046] Based on the smoothed environmental parameter matrix, the gradient distribution of environmental parameters in three-dimensional space is calculated, and the local rate of change of environmental parameters is determined.

[0047] Identify high-variability regions where the local rate of change exceeds a preset rate of change threshold and mark them as priority monitoring points;

[0048] The correspondence between priority monitoring points and sensor arrays is analyzed, and the data acquisition frequency of high-variability areas is dynamically adjusted based on the analysis results. New environmental parameter data are collected, and the initial environmental parameter matrix is ​​updated accordingly to obtain the real-time environmental parameter matrix.

[0049] Specifically, the workshop's three-dimensional space is pre-divided into a three-dimensional grid with a preset scale. The grid scale can be set according to the workshop's spatial dimensions, equipment layout density, ventilation conditions, and target monitoring accuracy; for example, a cubic grid with a side length of 1 meter can be used. Sensor nodes are preferably deployed at grid intersections or representative locations corresponding to key functional areas to achieve coverage monitoring of key areas such as production lines, raw material storage areas, and packaging areas. It should be noted that while this deployment method provides basic spatial coverage for workshop environmental parameter collection, it does not mean that all spatial locations will have a completely uniform and missing data distribution at any given time. In actual operation, factors such as partial obstruction, node maintenance, communication interruption, individual sensor failure, or temporary encrypted sampling in local areas may still lead to inconsistent spatiotemporal resolution of local data, resulting in missing data points or irregular distribution points that require subsequent processing.

[0050] Each sensor node integrates one or more sensing units for collecting environmental parameters, including air temperature, relative humidity, pollutant concentration, and parameters characterizing the level of microbial contamination in its local area. Parameters characterizing the level of microbial contamination can include bioaerosol concentration, microbial load in particles, estimated colony-forming units (CFU), ATP fluorescence intensity, or other detectable quantities that reflect the degree of microbial contamination. Each node collects data synchronously or quasi-synchronously according to a preset sampling period and transmits the information, including timestamps and spatial coordinates, to the central processing unit via wired or wireless communication.

[0051] After receiving the data uploaded by each node, the central processing unit organizes it according to its three-dimensional spatial coordinates to construct an initial environmental parameter matrix. In this matrix, each index position corresponds to a spatial grid position within the workshop, and the stored data is a set of values ​​for one or more environmental parameters at that position. Since the raw collected data may be affected by sensor measurement errors, instantaneous disturbances, or transmission noise, the time series data corresponding to each coordinate position can be smoothed, such as by using the moving average method or the exponential moving average method, to obtain more stable smoothed results for temperature, humidity, microbial contamination characterization parameters, and pollutant concentrations in the time dimension, and based on this, a smoothed environmental parameter matrix is ​​formed.

[0052] Based on the smoothed environmental parameter matrix, the spatial gradient distribution of each environmental parameter in three-dimensional space is calculated to characterize the direction and intensity of spatial change of the environmental parameters. Furthermore, the local spatial variation rate of the environmental parameters is determined based on the parameter differences between adjacent spatial grids. The local variation rate preferably characterizes the magnitude of environmental parameter change per unit spatial distance, reflecting the degree of environmental fluctuation within the region. By analyzing the local spatial variation rate of each grid region, areas with significant environmental parameter fluctuations can be identified. These areas typically correspond to potential pollution diffusion initiation areas or areas of abnormally active environment.

[0053] Based on historical monitoring data, health control standards, or empirical thresholds, a local variability rate threshold is pre-set. When the local spatial variability rate of a certain area exceeds this threshold, the area is identified as a high-variability area or a priority monitoring area. In one embodiment, one or more representative grid points can be selected from the high-variability area as priority monitoring points to establish a correspondence between regional risk and specific sensor nodes. High-variability areas indicate that the environmental parameters at the corresponding location change rapidly, typically requiring monitoring with higher temporal resolution.

[0054] After identifying high-variability areas, the spatial correspondence between these areas and existing sensor nodes is analyzed. The sampling frequency is dynamically adjusted only for the sensor nodes corresponding to these areas, without changing their spatial deployment locations. Specifically, the sampling time interval can be adaptively shortened and the data acquisition frequency increased based on the magnitude and persistence of environmental parameter changes in the area, thereby obtaining monitoring data with higher temporal resolution. Thus, while maintaining the overall spatial layout, the sensor array achieves focused and intensive sampling of localized high-risk areas.

[0055] As the sampling frequency increases, higher temporal resolution data collected by the corresponding nodes is continuously fed into the central processing unit and used to update the previously constructed environmental parameter matrix. During the update, newly sampled data can be written to the corresponding spatial coordinates after undergoing the same smoothing process, thus forming a real-time environmental parameter matrix with higher temporal resolution for locally highly variable areas, based on spatial coverage. This matrix retains the spatial distribution information of the overall workshop environmental parameters while enhancing the dynamic detail representation of high-risk areas, providing a data foundation for subsequent irregular distribution point completion, continuous spatial fluctuation field construction, and anomaly diffusion analysis.

[0056] Step 2: Based on the real-time environmental parameter matrix, apply a spatial interpolation algorithm to complete and smooth the data of irregularly distributed points to obtain a continuous spatial fluctuation field.

[0057] In one specific embodiment, the process of performing step 2 may specifically include the following steps:

[0058] Based on the real-time environmental parameter matrix, identify irregular distribution points caused by missing or uneven data intervals;

[0059] The Kriging spatial interpolation algorithm is applied to estimate the data of irregularly distributed points. The Kriging spatial interpolation algorithm estimates the environmental parameter values ​​of irregularly distributed points by weighted average based on the environmental parameter values ​​of adjacent known points. The weights are determined by spatial distance and variogram.

[0060] The estimated values ​​are used to fill the real-time environmental parameter matrix to obtain the complete parameter matrix, and the initial continuous spatial fluctuation field is constructed based on this matrix using the surface fitting method.

[0061] Calculate the gradient of the initial continuous spatial wave field, determine the wave direction and intensity, and obtain the wave vector field;

[0062] Based on the analysis of the wave vector field, spatial connectivity is identified to identify potential wave propagation channels;

[0063] The data on the wave propagation channel is compared with the original measured data at the corresponding position in the real-time environmental parameter matrix, and the deviation value is calculated. If the deviation value exceeds the preset deviation threshold, the parameters of the Kriging space interpolation algorithm are adjusted to correct the initial continuous space wave field and obtain the final continuous space wave field.

[0064] Specifically, during meat processing, the spatial distribution of environmental parameters is not always uniform and complete. Due to factors such as local node failures, communication interruptions, maintenance shutdowns, localized encrypted sampling, and inconsistent sampling frequencies in different areas, the real-time environmental parameter matrix obtained in step 1 may still contain some missing spatial location data, sparse local sampling, or inconsistent spatiotemporal resolution. Such issues can affect the continuous understanding of the overall environmental status of the workshop. Therefore, it is necessary to complete the relevant data points and construct a continuous spatial fluctuation field based on the completed data to provide a foundation for subsequent regional correlation analysis and anomaly diffusion identification.

[0065] Irregularly distributed points mainly include: data gaps where no valid observations were obtained due to node failure, communication anomalies, or maintenance, and sparsely sampled points where observations exist but the surrounding sampling points are significantly sparse, resulting in discontinuous local spatial distribution. It should be noted that a monitoring value of zero does not automatically constitute an irregularly distributed point. Only when, after considering node status, data validity markers, temporal continuity, and the completeness of neighborhood sampling, is it confirmed that the data is missing, distorted, or locally insufficiently sampled, is the corresponding location identified as a target point requiring completion.

[0066] After identifying irregularly distributed points, spatial interpolation algorithms are used to complete the data, with Kriging interpolation being the preferred method. This algorithm is based on the assumption that spatially adjacent locations are correlated. Under the condition that local spatial statistical characteristics are approximately stable, it estimates the target point based on the observed values ​​of nearby valid sample points and their spatial correlation structure. For different environmental parameters such as temperature, humidity, microbial contamination characterization parameters, and pollutant concentrations, corresponding interpolation models can be established and the estimation process executed independently. This processing method is a univariate interpolation method, which can meet the need for continuous estimation of key environmental parameters in workshop environmental monitoring scenarios while maintaining simplicity.

[0067] Specifically, spatial differences between neighboring sample points can be statistically analyzed for each environmental parameter, and corresponding variogram models can be fitted. The variogram model can be selected based on the fitting effect of the semivariogram scatter distribution, such as a spherical model, exponential model, or Gaussian model; preferably, a model with smaller fitting residuals and better reflection of local spatial correlation is chosen. Based on the selected variogram model, the interpolation weights of the target point relative to known neighboring points are calculated, thereby obtaining the parameter estimates for the location to be filled in. Through the above processing, missing points or insufficiently sampled points in the real-time environmental parameter matrix are filled in, forming a parameter matrix with a more complete spatial distribution. It should be noted that the main role of spatial interpolation is data completion and continuous estimation; it is not equivalent to the overall smooth reconstruction in the subsequent continuous field representation.

[0068] After obtaining the complete parameter matrix, a three-dimensional spatial field reconstruction is performed based on this matrix to obtain an initial continuous spatial fluctuation field. This step can employ volume field fitting, mesh reconstruction, or other continuous spatial representation methods to convert the environmental parameter values ​​on the discrete mesh into a three-dimensional continuous distribution form, thereby characterizing the overall variation trend of environmental parameters in the workshop space. Compared to the original discrete matrix, the continuous spatial fluctuation field can more intuitively describe the direction, intensity, and local coherence of environmental parameters in space. Therefore, the "smoothing" in step 2 mainly refers to performing continuous spatial representation and local fluctuation trend reconstruction on the discrete data after completing the spatial interpolation, rather than simply equating "smoothing" with Kriging interpolation itself.

[0069] In one implementation, the spatial variations of workshop environmental parameters can be considered to satisfy approximately stationary or locally stationary characteristics within a local time window, meaning that the correlation of data from nearby locations is mainly determined by spatial distance and local structure. For temperature, humidity, contaminant concentration, and microbial contamination characterization parameters in adjacent areas of a meat processing workshop, there is usually usable spatial correlation within a short time and local range; therefore, using the Kriging method for local completion is reasonable. If the number of sample points for a certain parameter in a local area is insufficient or the semivariogram fitting result is unstable, the neighborhood range can be narrowed or the minimum neighborhood sample size can be increased before performing interpolation to improve the robustness of the estimation results.

[0070] After the initial continuous spatial fluctuation field is constructed, its gradient distribution in three-dimensional space can be further calculated to obtain the gradient direction and gradient intensity at each spatial location. The gradient direction reflects the direction of the fastest increase in environmental parameters, and the gradient intensity reflects the local spatial variation amplitude. Based on the gradient information of all spatial locations, a fluctuation vector field can be formed to characterize the spatial variation trend of environmental parameters. It should be noted that this fluctuation vector field reflects the spatial variation trend of environmental parameters, rather than the actual transport direction of pollutants in the presence of complex ventilation disturbances. Therefore, the connectivity paths identified based on gradient consistency should be understood as potential fluctuation propagation channels or candidate diffusion trend channels, rather than the final determination of the actual pollution propagation path.

[0071] When identifying potential pathways, the space can be discretized into several adjacent nodes, and these nodes can be connected based on the consistency of their gradient directions. For example, when the angle between the gradient directions of adjacent nodes is less than a preset threshold, they can be considered to have a relatively consistent local change trend, thus establishing a connectivity relationship. This threshold can be set based on historical monitoring data, workshop ventilation conditions, or validation results. Based on this, potential connectivity paths extending from high-variance areas to surrounding areas can be identified, characterizing potential spatial fluctuation corridors of environmental parameters. These potential pathways are mainly used for regional correlation analysis and propagation prediction in subsequent steps and should not be directly equated to actual pollution diffusion paths driven by ventilation.

[0072] To improve the reliability of the continuous spatial fluctuation field, a verification and correction mechanism is further introduced. Verification points are selected in the aforementioned potential channels or other areas with significant changes. The estimated values ​​of the initial continuous spatial fluctuation field at these locations are compared with the corresponding original valid observations in the real-time environmental parameter matrix, and the deviation is calculated. If the average or maximum deviation of multiple verification points exceeds a preset deviation threshold, it indicates that the current interpolation model or spatial field reconstruction result differs significantly from the measured situation in a local area, requiring parameter correction. The deviation threshold can be set by considering sensor measurement accuracy, historical data fluctuation range, and allowable monitoring error.

[0073] During calibration, parameters directly related to the spatially relevant structure can be adjusted first, such as the variogram model type, nugget value, range, sill value, neighborhood search radius, minimum number of neighborhood samples, or at least one of the anisotropy parameters. Spatial interpolation and 3D spatial field reconstruction for the corresponding parameters are then re-executed. After one calibration, the verification point deviation is recalculated. Iteration can be stopped when both the average and maximum deviations of the verification points are below a preset deviation threshold, or when the deviation improvement over two consecutive iterations is below a preset convergence threshold. The current result is then determined as the final continuous spatial fluctuation field. Through this feedback calibration process, the degree to which the continuous spatial fluctuation field conforms to the actual environmental fluctuations in the workshop can be improved.

[0074] Therefore, step 2 completes the spatial representation and reconstructs the continuous field of the discrete environmental data obtained in step 1, transforming the environmental parameters from discrete monitoring point representations to continuous spatial representations, and forming basic spatial field data that can be used for subsequent regional division, correlation mapping, and anomaly diffusion analysis. This step mainly solves the problem of incomplete spatial understanding caused by local missing measurements, sparse sampling, and inconsistent spatiotemporal resolution.

[0075] Figure 3 It demonstrates the identification of irregularly distributed points, using scatter plots and red dashed boxes to mark areas with missing or irregularly distributed data, and using different symbols to mark valid and missing data points. Figure 4 The results of Kriging space interpolation are presented, and the distribution of the interpolated data is shown through a contour map. Red dots in the map represent interpolation supplement points, while black dots represent original valid data points. This map intuitively demonstrates how the interpolation algorithm fills in irregular data points and shows the effect of applying Kriging space interpolation to complete environmental data and construct spatial fluctuation fields.

[0076] Step 3: Divide the region based on the continuous spatial wave field, calculate the correlation coefficient matrix between adjacent regions, and generate a wave correlation diagram accordingly.

[0077] In one specific embodiment, the process of performing step 3 may specifically include the following steps:

[0078] The continuous spatial wave field is divided into regions based on a preset three-dimensional spatial grid;

[0079] For each pair of adjacent regions, calculate the Pearson correlation coefficient between their environmental parameter sequences;

[0080] Based on all Pearson correlation coefficients, a correlation coefficient matrix indexed by regional units is constructed.

[0081] The correlation coefficient matrix is ​​filtered according to a preset significance threshold, and significant correlation pairs with coefficient values ​​exceeding the threshold are retained to obtain a simplified matrix.

[0082] An initial fluctuation correlation graph is constructed based on a simplified matrix, with regions as nodes and significantly correlated pairs as edges.

[0083] Calculate the degree of each node in the initial fluctuation correlation graph. The degree represents the number of other nodes associated with that node. The node with the highest degree is determined as the central region.

[0084] The environmental parameter sequence of the central region is compared with the environmental parameter sequences of adjacent points in the continuous spatial fluctuation field. Based on the comparison results, the edge weights of the initial fluctuation correlation graph are updated to obtain the fluctuation correlation graph, which reflects the correlation strength of environmental parameter fluctuations.

[0085] Specifically, based on the continuous spatial fluctuation field, the physical space of the workshop is divided into multiple non-overlapping regional units according to a preset spatial resolution. The spatial resolution can be consistent with the sensor deployment grid, or it can be set to a finer three-dimensional grid according to the analysis accuracy requirements. Each regional unit is uniquely identified by its spatial location and corresponds to a set of environmental parameter data of the continuous spatial fluctuation field within that region. To facilitate subsequent correlation analysis, the continuous field data within each regional unit can be extracted in chronological order as a regional-level environmental parameter time series; for example, this time series can be formed by the average value, weighted average value, or representative point value of each sampling point within the region. Thus, the regional units correspond one-to-one with the nodes in the figure, and the data object corresponding to the node is a regional-level time series, rather than the instantaneous value of a single discrete point.

[0086] After obtaining the time series of each regional unit, correlation analysis is performed on spatially adjacent regional pairs. Preferably, the Pearson correlation coefficients of adjacent regions on environmental parameters such as temperature, humidity, microbial contamination characterization parameters, and pollutant concentrations are calculated separately to characterize the degree of coordinated fluctuation within the monitoring time window. Alternatively, correlation coefficients of multiple environmental parameters can be calculated separately and then weighted or averaged to obtain a comprehensive correlation index between regional pairs. It should be noted that the Pearson correlation coefficient mainly reflects the degree of synchronous linear correlation of changes in environmental parameters in adjacent regions. Therefore, the correlation constructed in this step is mainly used to characterize the strength of spatial fluctuation correlation and provide a candidate structural basis for subsequent propagation analysis, rather than being directly equivalent to a physical propagation path with a definite temporal causal relationship.

[0087] After calculating the correlation for all adjacent region pairs, a correlation coefficient matrix is ​​constructed using the region units as row and column indices. To reduce interference from weak correlations and random noise, a significance threshold can be set, such as based on historical data distribution, statistical significance test results, or empirical thresholds. Correlation coefficients with absolute values ​​below this threshold are set to zero, while region pairs with absolute values ​​reaching or exceeding the threshold are retained, thus obtaining a sparse, simplified matrix. An initial fluctuation correlation graph is constructed based on this simplified matrix, where each node corresponds to a region unit, and each edge represents a significant fluctuation correlation between two adjacent regions. The initial weight of the edge can be determined by the comprehensive correlation index of the corresponding region pair. In this way, the continuous spatial fluctuation field is transformed into a regional-level spatial correlation network, facilitating subsequent graph structure analysis of key regions and highly correlated paths.

[0088] To identify key regions with strong connectivity in the network, the degree of each node (the number of edges directly connected to that node, i.e., the number of its neighboring nodes) can be calculated, and the region corresponding to the node with the highest degree or the node with a preset leading degree can be identified as a candidate central region. It should be noted that this central region only represents a region in the current fluctuating network that has many significant connections with other regions, reflecting its importance in the statistical correlation structure. It can serve as a reference region for subsequent local weight correction, but is not directly equivalent to the actual pollution source or physical diffusion center.

[0089] In one embodiment, to improve the accuracy of the initial fluctuation correlation diagram in reflecting local actual fluctuation characteristics, a boundary weight correction process based on the central region is introduced. The time series of regional environmental parameters corresponding to the central region is extracted, and the time series of regional environmental parameters corresponding to each spatially directly adjacent region are also extracted. The two are compared time-by-time, and the average absolute difference or other difference indicators between the central region and each adjacent region are calculated. It should be noted that the objects involved in the comparison are all regional time series; if the original data in the continuous spatial fluctuation field originates from multiple points, the data within the adjacent regions should first be averaged, sampled from representative points, or equivalently aggregated before being used for inter-regional difference calculation to ensure the consistency of the comparison object's hierarchical level.

[0090] Based on the aforementioned differences, local weight correction is performed on the edges directly connected to the central region in the initial fluctuation correlation graph, rather than updating all edges simultaneously. During correction, the difference index is first normalized to convert it into a dimensionless adjustment quantity, and then the original edge weights are corrected according to a preset mapping relationship. For example, if the normalized difference between the central region and a neighboring region is large, the weight of that neighboring edge can be appropriately reduced; if the normalized difference is small, the original edge weight is retained or moderately increased. Thus, the corrected local edge weights can better reflect the strength of the actual fluctuation consistency between the central region and neighboring regions. After completing the above local correction, the final fluctuation correlation graph is obtained.

[0091] Through the above processing, the resulting fluctuation correlation diagram can characterize the statistical correlation strength of environmental parameter fluctuations between different areas of the workshop, and highlight the regional structures with strong connectivity and significant correlation. The main function of this diagram is to transform the continuous spatial fluctuation field into a graph structure representation that can be used for subsequent path search and propagation trend analysis, providing a regional network foundation for identifying potentially highly correlated regions, candidate source nodes, and their associated paths.

[0092] Figure 5This diagram illustrates the division of a 3D spatial grid into regions, showing the arrangement of the sensor array. Each black dot represents the center of a region, while red dots indicate the specific locations of the sensors. This diagram demonstrates how a 3D spatial grid is used to divide the region and determine the sensor placement within each region. Figure 6 This diagram illustrates a significant correlation coefficient matrix, showing the correlations between different regions. The value of each matrix element represents the Pearson correlation coefficient between the regions, with color intensity reflecting the strength of the correlation. Correlation coefficients higher than 0.7 are considered significantly correlated.

[0093] Step 4: In the fluctuation correlation diagram, regions with correlation coefficients exceeding a preset reference threshold are marked as correlation source nodes. The initial propagation path set starting from the correlation source nodes is obtained through a path search algorithm. The correlation source nodes are potential sources of pollutants or other environmental changes.

[0094] In one specific embodiment, the process of performing step 4 may specifically include the following steps:

[0095] Based on historical environmental data, a preset reference threshold is set to identify nodes corresponding to areas where the correlation coefficient exceeds the preset reference threshold, and these nodes are marked as associated source nodes.

[0096] For any associated source node, traverse all its connected edges in the oscillating association graph and construct an initial path starting from any associated source node;

[0097] Based on path length and edge weight, calculate the path score for each initial path and filter out high-scoring paths with scores higher than a preset score threshold.

[0098] Perform similarity analysis on all high-scoring paths, merge similar paths, and obtain an initial path set;

[0099] Based on the initial path set, a depth-first search algorithm is applied to expand the potential branch paths to obtain an expanded path set.

[0100] The extended path set is compared with the fluctuation correlation graph, cyclic paths are removed, and the effective path set is determined.

[0101] Calculate the path score for each path in the set of valid paths, sort the set of valid paths according to the path scores, and obtain the initial propagation path set.

[0102] Specifically, a reference threshold is set based on historical monitoring data or the current edge weight distribution in the graph. This threshold can be determined based on the high quantile of the edge weight statistical distribution; for example, the 80th quantile of the edge weight distribution can be selected as the threshold in this embodiment. When the edge weight between a node in a certain region and its adjacent nodes reaches or exceeds this reference threshold, it indicates a strong statistical correlation between the node and its surrounding regions. This node can then be marked as a candidate correlation source node, forming a set of candidate starting nodes. It should be noted that candidate correlation source nodes represent initial analysis objects with strong regional correlation in the current fluctuation correlation network, used for subsequent candidate path searching, and are not directly equivalent to confirmed pollution source areas or actual transmission sources.

[0103] After determining the set of candidate starting nodes, candidate paths are constructed in the fluctuation correlation graph, starting from each candidate source node. Since the fluctuation correlation graph obtained in step 3 mainly reflects the statistical correlation between regions, and its edges represent the strength of the correlation, the paths obtained in this step should be understood as candidate paths extracted based on the correlation structure, used to represent possible diffusion trends or propagation candidate relationships, rather than deterministic propagation trajectories whose physical direction has already been determined in this step. Specifically, starting from any candidate source node, a first-order path can be formed by traversing its directly connected adjacent edges, and then extended outwards to higher-order adjacent nodes, generating multiple candidate paths within a preset search depth. The search method can adopt a breadth-first search approach, and the maximum search depth can be set according to the workshop space scale, regional division granularity, and the density of the correlation network, such as 3 to 5 layers.

[0104] To select more representative paths from the candidate paths, a path score is introduced as a path evaluation metric. For example, the path score can be determined based on the combined weights of each edge on the path and the path length, such as using the average edge weights, weighted average, or other indicators that can characterize the overall interconnectedness of the paths. The path score measures the overall interconnectedness and extension cost of the candidate path in the fluctuation interconnection graph; a higher score indicates that the path maintains a strong regional interconnection over a shorter length. It should be noted that this score is an evaluation metric used for path selection and ranking and does not directly represent the actual probability of pollutant propagation. Based on a preset score threshold, high-scoring paths can be selected from the initial candidate paths as input for subsequent redundancy removal processing.

[0105] Since different candidate starting nodes may correspond to partially overlapping node sequences, or multiple paths searched from the same starting node may have local overlap, it is necessary to perform similarity analysis and redundancy removal on high-scoring paths. For example, the degree of node overlap can be used as the criterion for judging path similarity. When the node sets contained in two paths have a high degree of overlap, they are considered to have high similarity in regional association structure. For similar paths, the one with the higher path score can be retained as the representative path, or they can be merged according to preset rules to form an initial set of redundant paths. This similarity analysis is mainly used to eliminate duplicate or highly similar candidate association paths; its evaluation basis is still the similarity of the regional node sets, rather than the propagation direction itself.

[0106] After obtaining the initial path set, path expansion can be performed based on the terminal nodes of each path in the initial path set. In one embodiment, a depth-first search method can be used to recursively expand the terminal nodes along unvisited adjacent edges until a preset expansion depth is reached or no new expandable nodes are found, thus obtaining an expanded path set. Subsequently, the expanded path set is checked cyclically to remove cyclic paths with duplicate nodes, retaining the set of valid paths without cycles.

[0107] After obtaining the set of valid paths, a path score is calculated for each path to prioritize them. This score is determined based on factors such as the combined strength of the edge weights on the path, the path length, and the continuity of the path structure. It reflects the relative priority and overall significance of the path in the current fluctuation correlation graph, rather than a strict physical propagation probability. Sorting the set of valid paths according to their scores yields the initial propagation path set. It should be noted that this initial propagation path set refers to the candidate set of paths extracted from the regional fluctuation correlation network based on statistical correlation strength and path structure priority. Its purpose is to provide structured input for subsequent processes combining ventilation flow data, graph convolutional network prediction, and online calibration, rather than directly determining the true diffusion direction, propagation order, and final propagation boundary in this step.

[0108] Through the above processing, step 4 extracts candidate correlation source nodes from the fluctuation correlation graph, generates candidate correlation paths, completes path screening and redundancy removal, and forms an initial path set that can be used for subsequent dynamic propagation prediction. This step further organizes the implicit high correlation structure between regions into a priority path candidate set, providing an input basis for introducing ventilation direction constraints and dynamic diffusion simulation in subsequent steps, thereby completing the transition from static regional correlation analysis to dynamic propagation trend analysis.

[0109] Step 5: Combine the initial set of propagation paths with ventilation flow data, and apply graph convolutional networks to simulate the diffusion dynamics of microorganisms and / or pollutants in the workshop, thereby obtaining predicted propagation trajectories.

[0110] In one specific embodiment, the process of performing step 5 may specifically include the following steps:

[0111] Acquire ventilation flow data, which should include at least the speed and direction of airflow within the workshop;

[0112] The initial propagation path set and ventilation flow data are input into a pre-constructed graph convolutional network. The graph convolutional network has a wave correlation graph as its graph structure and environmental parameters as the initial features of the nodes.

[0113] Neighborhood information is aggregated through the convolutional layers of a graph convolutional network, and the feature representation of each node is iteratively updated to simulate the diffusion dynamics of microorganisms and / or pollutants along the graph structure under ventilation-driven conditions.

[0114] The initial propagation trajectory is extracted based on the simulation results. The initial propagation trajectory represents the spatiotemporal sequence of microorganisms and / or pollutants spreading outward from the source node.

[0115] Integrate the initial propagation trajectory over time steps to calculate its cumulative spatial impact and obtain the trajectory intensity distribution;

[0116] The trajectory intensity distribution is compared with the initial propagation path set. The accuracy of the simulation is verified based on the comparison results, and the calibration parameters are determined.

[0117] Adjust the parameters of the graph convolutional network according to the calibration parameters, and re-simulate to obtain the predicted propagation trajectory including the corrected diffusion boundary.

[0118] Specifically, after obtaining the initial set of propagation paths, in order to predict the dynamic diffusion process of microorganisms and / or pollutants in the workshop, the initial set of propagation paths is combined with ventilation flow data and input into a pre-constructed graph convolutional network for diffusion simulation, thereby obtaining the predicted propagation trajectory.

[0119] The pre-constructed graph convolutional network uses the fluctuation correlation graph obtained in step 3 as its base graph structure. In this graph, nodes correspond to regional units within the workshop, and edges represent the fluctuation correlations between these regional units. An adjacency matrix is ​​constructed based on the regional correlations obtained in step 3, with elements representing the initial correlation strength between adjacent regions. The initial features of each node consist of environmental parameters of the corresponding regional unit, including at least one or more of microbial concentration, pollutant concentration, temperature, and humidity, thus forming a node feature matrix. The initial propagation path set does not replace the graph structure itself but participates in network computation as prior path information. To facilitate input into the graph convolutional network, the initial propagation path set can be transformed into node-level prior features and / or edge-level prior weights. Node-level prior features characterize whether the corresponding node is within the coverage area of ​​the initial propagation path set; edge-level prior weights characterize whether the corresponding edge is located on a candidate propagation path covered by the initial propagation path set. For example, nodes located in the initial propagation path set can be labeled with a first preset value, and nodes not located in the initial propagation path set can be labeled with a second preset value to form path prior node features; and edges located in the initial propagation path set can be assigned initial edge weights higher than non-path edges to form path prior edge features. Thus, the initial propagation path set is transformed into a structured prior input that can participate in model computation in this step, rather than existing only as a functional description.

[0120] Ventilation flow data includes at least the velocity and direction of airflow within the workshop, which can be obtained from wind speed and direction sensors deployed at air supply outlets, return air outlets, and key locations within the workshop. To facilitate coupling with the area map structure, the ventilation flow data is matched according to area units to obtain the ventilation characteristic information corresponding to each area. Furthermore, for any adjacent area A and area B in the map, based on the relationship between the direction of the line connecting area A to area B and the dominant local airflow direction, the airflow velocity component in the direction from area A to area B is determined, and this airflow velocity component is used to characterize the ventilation driving intensity propagating from area A to area B. Thus, the original undirected area association relationship is transformed into directional edge features.

[0121] In one embodiment, the initial association strength obtained in step 3, the path prior edge weights formed by the initial propagation path set, and the directional driving strength formed by ventilation flow data can be dimensionlessly processed, and propagation edge weights can be generated according to a preset fusion rule. The propagation edge weights are used to characterize the information propagation strength between adjacent regions under the combined effects of statistical association, path prior, and ventilation driving. For example, the propagation edge weights can be determined by weighting the following: initial association strength, path prior edge weights, and directional driving strength, where each weight can be preset or updated during training. Based on the propagation edge weights, a propagation matrix for message passing in the graph convolutional network is constructed, thereby enabling the graph convolutional network to move beyond the limitations of the standard symmetric normalized adjacency matrix and support directional inter-region propagation modeling.

[0122] During network computation, the aforementioned node feature matrix is ​​used as node input, and the propagation matrix is ​​used as edge propagation constraints to perform multiple rounds of graph convolution or message passing iterations. In the t-th iteration, each node aggregates the feature information of its neighboring nodes after the propagation edge weight modulation, and calculates it together with the node's features from the previous round to obtain the node's updated features in the t-th round. As the number of iterations increases, the node features gradually absorb the influence of a larger neighborhood, thereby characterizing the discrete-time diffusion process of microorganisms and / or pollutants in the regional graph structure. Therefore, the graph convolutional network in this step is essentially a graph message passing model that introduces path priors and ventilation direction constraints based on a fixed regional graph structure.

[0123] Graph convolutional networks can be implemented using a combination of offline training and online prediction. During offline training, training samples can be constructed using historical monitoring data, historical abnormal diffusion records, historical ventilation flow data, and regional environmental parameter sequences within the corresponding time periods. The input consists of node features, path prior features, and ventilation-driven features from several consecutive time points, while the output labels are the pollutant concentration, concentration increment, and / or affected state of each regional unit at the next time point. When the output is a continuous numerical value, mean squared error loss and / or mean absolute error loss can be used; when the output represents an affected state, cross-entropy loss can be used; when both continuous numerical values ​​and state categories are output simultaneously, a combined loss function consisting of regression and classification loss terms can be used. After training, the network parameters are saved for online execution.

[0124] During the online operation phase, the current regional environmental parameters, the prior information of the initial propagation path set, and the current ventilation flow data are input into the trained graph convolutional network to perform forward propagation, obtaining the prediction results for each regional unit within one or more discrete time steps. The prediction results include at least one of the following for each regional unit at the corresponding time step: the predicted concentration value, the concentration increment, or the affected state. Based on the prediction results of each discrete time step, regional nodes that first reach a preset influence threshold can be identified, and these regional nodes are recorded in chronological order, thus forming an initial propagation trajectory expanding outward from the source node. Then, the node prediction values ​​within each time step are accumulated or integrated to obtain the cumulative influence value of each regional node within the prediction time window, forming the trajectory intensity distribution.

[0125] To verify the consistency between the simulation results and the path prior and real-time monitoring results, nodes with a cumulative influence value greater than a preset influence threshold in the trajectory intensity distribution are identified as the affected node set. The overlap between this affected node set and the path node set covered by the initial propagation path set is compared. Simultaneously, the error between the predicted results and the latest measured environmental parameters is calculated. When the node set overlap is lower than a preset requirement, and / or the error between the predicted and measured values ​​exceeds a preset error threshold, calibration is performed on the graph convolutional network. The calibration parameters specifically include at least one of the following: fusion coefficients for fusing the initial association strength, path prior edge weights, and directional driving strength; edge weight update parameters in the propagation matrix; and message passing weight parameters in the graph convolutional layer. If necessary, the step size parameter used during online updates may also be included. During calibration, a calibration objective function containing a prediction error term and a node set inconsistency penalty term can be constructed, and the above calibration parameters are iteratively updated with small steps by minimizing this calibration objective function.

[0126] After parameter updates, diffusion simulation is re-executed based on the same region map structure, and the propagation trajectory and affected node set are extracted again. If the updated prediction error and the degree of inconsistency in the node set both meet the preset requirements, the corresponding result is determined as the calibrated predicted propagation trajectory. If the requirements are still not met, parameter updates and diffusion simulations continue until the preset requirements are met, or the improvement magnitude of multiple consecutive iterations is lower than the preset convergence threshold. The final result is the predicted propagation trajectory containing the corrected diffusion boundary.

[0127] Through the above processing, this step unifies the fluctuation correlation graph obtained in step 3, the initial propagation path set obtained in step 4, and the current ventilation flow data into node inputs, edge propagation constraints, and path prior constraints that can be processed by graph convolutional networks. It realizes iterative simulation of the diffusion process of microorganisms and / or pollutants on the regional graph structure, and outputs predicted propagation trajectories with temporal sequence and spatial range information, providing a basis for subsequent abnormal diffusion boundary correction and optimal allocation of monitoring resources.

[0128] Figure 7 The visualization results of the trajectory intensity distribution and the calibration area are shown. Contour lines are used to represent the distribution of trajectory intensity. The red dashed box in the figure represents the calibration area, indicating that the predicted results in this area deviate significantly from the actual data, and parameter adjustments are needed to improve the accuracy of the simulation. The color blocks in the figure reflect different values ​​of trajectory intensity (dark areas represent higher intensities).

[0129] Step 6: Compare the predicted propagation trajectory with the real-time environmental parameter matrix, update the edge weights in the fluctuation correlation graph based on the differences, and determine the abnormal propagation boundary based on the updated correlation graph.

[0130] In one specific embodiment, the process of performing step 6 may specifically include the following steps:

[0131] The predicted propagation trajectory is compared with the real-time environmental parameter matrix, and the difference between the predicted value at the trajectory point and the measured value at the corresponding position in the matrix is ​​calculated.

[0132] The difference value is compared with a preset difference threshold to identify outliers where the difference value exceeds the threshold.

[0133] For outliers, the weights of adjacent edges in the fluctuation correlation graph are updated by merging the difference values ​​into the original edge weights in a weighted average manner.

[0134] Based on the fluctuation correlation graph with updated weights, the correlation coefficients between regions are recalculated to obtain the corrected correlation coefficient matrix.

[0135] Based on the modified correlation coefficient matrix, the connected component analysis method is used to extract the preliminary anomalous diffusion boundary; morphological operations are performed on the preliminary anomalous diffusion boundary to smooth the boundary line and obtain a smooth boundary.

[0136] The smooth boundary is compared with the predicted propagation trajectory to calculate the boundary deviation and determine the boundary confidence.

[0137] High-confidence boundaries with a confidence level higher than a preset confidence threshold are selected and used as anomalous diffusion boundaries.

[0138] Specifically, the predicted propagation trajectory is a time-sequential sequence of regional propagation points. Each trajectory point corresponds to a regional cell in the workshop spatial grid and is further mapped to a corresponding node in the fluctuation correlation diagram. The real-time environmental parameter matrix represents the measured environmental parameter values ​​of each regional cell at the current moment or within the current sliding time window. Based on a unified regional cell index, each trajectory point in the predicted propagation trajectory can be compared with the measured values ​​of the corresponding regions in the real-time environmental parameter matrix to obtain the difference between the predicted and measured values, thus characterizing the degree of deviation between the model prediction and on-site monitoring at that region.

[0139] The above-mentioned difference values ​​are compared with a preset difference threshold to identify anomalies where the difference exceeds the threshold. The difference threshold can be determined based on the sensor measurement error range, historical fluctuation statistical characteristics, or a preset warning tolerance. Anomalies reflect areas where there is a significant deviation between the current predicted trajectory and the actual monitoring results, indicating that there may be localized enhanced diffusion, changes in ventilation conditions, or dynamic disturbances that cannot be fully represented by the original associated structures.

[0140] Based on the identified outliers, a local edge weight update is performed on the fluctuation correlation graph. Specifically, taking the node corresponding to the outlier as the center, the adjacent edges directly connected to that node are selected as the edges to be updated. During the update, the difference values ​​are first converted into dimensionless adjustment quantities, such as through normalization or interval mapping, to uniformly convert differences of different dimensions or magnitudes into standardized indicators that can be used for edge weight correction, and then weighted and fused with the original edge weights. In this way, the edge weights connected to the outlier areas are locally corrected according to the current prediction deviation. The purpose of this update step is to inject real-time monitoring feedback into the original correlation graph, forming a set of corrected edge weights that reflect the latest local anomaly information, providing structural priors and key areas of focus for subsequent regional correlation reassessment, rather than generating new correlation coefficients independently of the original data.

[0141] After completing the aforementioned local edge weight correction, the correlation of the time series of relevant regions in the correlation graph is recalculated based on the latest real-time environmental parameter matrix to obtain the corrected correlation coefficient matrix. Preferably, a sliding time window incorporating the latest monitoring data is used during the recalculation to ensure that the correlation results retain historical continuity while reflecting current abnormal trends. This recalculation is still based on the latest regional-level measured data series, and the aforementioned edge weight update is used to indicate the local connectivity relationships that should be prioritized for correction, serving as an intermediate feedback process for the dynamic adjustment of the correlation graph. Thus, the corrected correlation coefficient matrix can simultaneously reflect the spatial correlation state after real-time data changes and graph structure feedback updates.

[0142] Based on the corrected correlation coefficient matrix, candidate spatial ranges for anomalous diffusion are further extracted. Specifically, connections in regions reaching a preset strong correlation threshold are preserved, and connected component analysis is performed to divide the interconnected regional nodes into several connected regions. Among them, connected components containing a large number of anomalous points, having a high degree of overlap with the predicted propagation trajectory, or having a large regional scale can be identified as candidate regions for anomalous diffusion. Extracting the spatial contours of the outer regional units of this candidate region yields the preliminary anomalous diffusion boundary. It should be noted that this boundary represents the outer edge of the highly correlated anomalous region obtained under the current "prediction-measurement-feedback correction" framework, and belongs to the candidate boundary for anomalous diffusion obtained through data-driven identification.

[0143] Because the initial boundary is affected by spatial discretization and local noise, edge burrs or local gaps may appear. Therefore, morphological smoothing can be performed on it to obtain a more continuous and stable boundary profile. The smoothed boundary is then compared with the predicted propagation trajectory in this round to assess the degree of matching between the boundary and the trajectory. For example, the spatial deviation between the boundary point set and the trajectory point set can be used as the evaluation criterion; for instance, the boundary deviation can be calculated based on the Hausdorff distance or other spatial distance indicators, and this deviation can be mapped to the boundary confidence score. It should be noted that this confidence score is used to reflect the degree of consistency between the smoothed boundary and the predicted trajectory in spatial coverage, and is one of the evaluation indicators of boundary reliability, but not the sole criterion for judging the true physical boundary.

[0144] Finally, based on a preset confidence threshold, the boundary with higher confidence is selected as the abnormal diffusion boundary at the current moment. The abnormal diffusion boundary obtained in this step is determined by continuously comparing the predicted propagation trajectory with real-time monitoring data, and through local edge weight correction, correlation reassessment, connected region extraction, boundary smoothing, and consistency assessment. It can be used to characterize the main spatial impact range of the current abnormal diffusion and provide a basis for subsequent early warning, key area review, and monitoring resource adjustment.

[0145] Through the above processing, step 6 realizes a closed-loop feedback update from "predicted trajectory" to "real-time corrected abnormal diffusion boundary": first, abnormal areas are identified by using the deviation between prediction and actual measurement, then the local correlation structure is corrected with the abnormal area as the center, and the spatial correlation is re-evaluated in combination with the latest monitoring data, and finally the abnormal diffusion boundary that is more consistent with the current monitoring status is extracted, thereby improving the dynamic adaptability and result stability of diffusion range identification.

[0146] Step 7: Based on the abnormal diffusion boundary, dynamically adjust the spatial distribution of the sensor array in the workshop and optimize the dot matrix layout configuration.

[0147] In one specific embodiment, the process of performing step 7 may specifically include the following steps:

[0148] Based on the abnormal diffusion boundary, high-variability sections where the rate of change of environmental parameters exceeds the preset change threshold are classified as high-risk areas;

[0149] For any high-risk area, the number of sensor nodes deployed is increased based on its boundary area and the preset sensor coverage density.

[0150] Based on the adjusted deployment numbers, sensor nodes are redistributed in high-risk areas to generate an initial dot matrix layout.

[0151] The average boundary distance is then calculated based on the shortest Euclidean distance from each sensor node to the abnormal diffusion boundary in the initial dot matrix layout.

[0152] Determine if the average boundary distance is greater than a preset distance threshold. If so, iteratively adjust the sensor node positions until the coverage requirements are met, and obtain the iterative dot matrix layout.

[0153] Based on the iterative dot matrix layout, sparse sampling in low-risk areas is integrated to achieve a global balanced allocation of monitoring resources;

[0154] Simulate the data acquisition process under balanced allocation and obtain data acquisition efficiency indicators;

[0155] Determine if the efficiency index is better than the preset standard. If so, confirm that the current layout is an optimized dot matrix layout configuration. If not, iteratively adjust the sensor node positions until a configuration scheme that meets the efficiency requirements is obtained.

[0156] Specifically, after obtaining the boundary of abnormal diffusion, to improve the targeting of monitoring in the abnormal diffusion area, the spatial distribution of the sensor array within the workshop is dynamically adjusted to form a dot matrix layout configuration more suitable for the current risk distribution. Dynamically adjusting the spatial distribution of the sensor array within the workshop does not simply refer to the on-site disassembly and relocation of fixed-installation sensors, but rather to performing at least one of the following on-site adjustments: position adjustment, node start / stop switching, and / or temporary deployment of monitoring nodes with reconfigurability, based on the optimized layout results. Specifically, dynamic adjustment can be achieved in the following ways: First, using movable monitoring nodes installed on mobile carriers, sliding rail mechanisms, lifting mechanisms, or hoisting mechanisms, and adjusting the actual positions of the nodes according to the optimization results; Second, for monitoring nodes pre-deployed in multiple candidate locations, switching the active and dormant states of the corresponding nodes according to the optimization results to change the actual spatial distribution of data collection; Third, temporarily deploying or setting up portable monitoring nodes in high-risk areas to form the adjusted monitoring array together with fixed nodes.

[0157] Based on the real-time environmental parameter matrix, the rate of change of environmental parameters in each regional unit within the abnormal diffusion boundary is calculated, and regional units with a rate of change exceeding a preset threshold are classified as high-variability zones; adjacent or connected high-variability zones constitute high-risk areas. The rate of change can be determined based on the amount of change in environmental parameters between adjacent time steps, and the threshold can be set based on historical fluctuation baselines, sensor measurement error ranges, or preset early warning sensitivities.

[0158] For any high-risk area, the required number of target nodes is determined based on its boundary area and preset sensor coverage density. This target number is then compared with the currently active number of nodes to determine the number of nodes that need to be newly deployed, relocated, or switched over. Subsequently, sensor nodes are redistributed within the high-risk area to generate an initial dot matrix layout. This initial dot matrix layout can be tailored to the shape of the high-risk area and the location of the anomalous diffusion boundary, employing a boundary-adjacent arrangement, a balanced arrangement within the area, or a hybrid combination of both, to ensure monitoring capabilities at both the edge of anomalous diffusion and within highly active areas.

[0159] After establishing the initial dot matrix layout, its boundary coverage effect is evaluated. Specifically, for each sensor node in the initial dot matrix layout, the shortest Euclidean distance to the anomalous diffusion boundary is calculated, and the shortest Euclidean distances are statistically analyzed to calculate the average boundary distance. The average boundary distance characterizes the overall fit of the sensor nodes to the anomalous diffusion boundary. The smaller the average boundary distance, the closer the sensor nodes are to the anomalous diffusion boundary, and the better the coverage fit of the layout to the anomalous diffusion boundary.

[0160] The average boundary distance is compared with a preset distance threshold to determine whether the current preliminary dot matrix layout meets the boundary coverage requirements. If the average boundary distance is greater than the preset distance threshold, it indicates that the overall fit of the sensor nodes relative to the abnormal diffusion boundary in the current layout is insufficient, requiring further adjustments to node positions, node activation status, and / or the number of local nodes. If the average boundary distance is not greater than the preset distance threshold, the current layout can be considered to meet the boundary coverage requirements. The preset distance threshold can be set based on the abnormal diffusion boundary identification accuracy requirements, sensor spatial resolution requirements, or actual deployment experience. It should be noted that the average boundary distance is mainly used to evaluate the closeness of the sensor layout to the abnormal diffusion boundary; it is a boundary coverage evaluation indicator, primarily reflecting the layout's ability to monitor and fit the edge area of ​​abnormal diffusion, and is not the sole evaluation criterion for the overall spatial coverage capability within high-risk areas.

[0161] If the initial dot matrix layout does not meet the aforementioned boundary coverage requirements, an iterative adjustment process will begin. For mobile monitoring nodes, optimization can be achieved by adjusting their target locations within high-risk areas; for pre-deployed candidate nodes, optimization can be achieved by switching the start / stop status of nodes at different candidate locations; for portable monitoring nodes, optimization can be achieved by changing their installation locations. After each adjustment, the average boundary distance is recalculated, and the "layout adjustment - boundary coverage assessment" process is repeated until the preset distance threshold requirement is met, resulting in an iterative dot matrix layout that meets the boundary coverage requirements.

[0162] After obtaining the iterative dot matrix layout, sparse sampling configuration is implemented in low-risk areas to achieve a globally balanced allocation of monitoring resources. The node density in low-risk areas is lower than that in high-risk areas, but the minimum number of monitoring nodes required to meet basic situational awareness, anomaly detection, and communication redundancy is retained. Preferably, the number of nodes retained in low-risk areas should at least meet the requirements for periodic collection of key environmental parameters, anomaly detection, and communication redundancy, thereby avoiding monitoring blind spots or alarm delays due to excessive reduction. By integrating the dense layout in high-risk areas with the sparse layout in low-risk areas, a globally balanced sensor distribution scheme is obtained.

[0163] To further evaluate the operational effectiveness of this distribution scheme, the data acquisition process under this scheme is simulated to obtain data acquisition efficiency indicators. Efficiency indicators can comprehensively consider at least one or more of the following: effective data acquisition volume, data refresh latency in key areas, communication load, and system energy consumption. Preferably, before calculating the comprehensive efficiency indicator, each indicator is normalized and then fused according to preset weights to ensure comparability between indicators of different dimensions. The resulting efficiency indicator reflects the overall performance of the current deployment in terms of monitoring effectiveness and operational costs.

[0164] The efficiency index is compared with the preset efficiency standard (based on historical best values ​​or design target settings). If the efficiency index is better than the preset standard, the current layout can be confirmed as an optimized dot matrix layout configuration; if it does not meet the preset standard, the node positions, node start / stop status, or the number of local nodes are iteratively adjusted until a configuration scheme that meets the efficiency requirements is obtained, provided that the coverage requirements of the high-risk area boundary are met.

[0165] Through the above processing, step 7 achieves adaptive reconfiguration of the sensor array based on the abnormal diffusion boundary: first, high-risk areas are identified and local monitoring density is increased; then, node layout is constrained by boundary coverage requirements; and finally, global balance optimization is performed by combining the basic monitoring needs of low-risk areas and system operating efficiency. This transforms monitoring resources from a static, uniform distribution to a dynamic configuration oriented towards abnormal diffusion areas, improving monitoring accuracy in high-risk areas while considering system energy consumption, timeliness, and overall deployment efficiency.

[0166] The above describes the intelligent environmental monitoring method for meat processing in the embodiments of this application. The following describes the intelligent environmental monitoring system for meat processing in the embodiments of this application. Please refer to [link / reference]. Figure 8 The present application provides a schematic diagram of the structure of an intelligent environmental monitoring system for meat processing, which includes:

[0167] The matrix construction module 10 is used to collect environmental parameter data in real time through a sensor array deployed in the meat processing workshop and construct a real-time environmental parameter matrix.

[0168] The preprocessing module 20 is used to complete and smooth the data of irregularly distributed points by applying a spatial interpolation algorithm based on the real-time environmental parameter matrix, so as to obtain a continuous spatial fluctuation field.

[0169] The region division module 30 is used to divide regions based on the continuous spatial fluctuation field, calculate the correlation coefficient matrix between adjacent regions, and generate a fluctuation correlation diagram accordingly.

[0170] The path generation module 40 is used to mark regions with correlation coefficients exceeding a preset reference threshold as associated source nodes in the fluctuation correlation graph, and to obtain an initial set of propagation paths starting from the associated source nodes through a path search algorithm, wherein the associated source nodes are potential sources of pollutants or other environmental changes.

[0171] The trajectory prediction module 50 is used to combine the initial set of propagation paths with ventilation flow data, apply graph convolutional networks to simulate the diffusion dynamics of microorganisms and / or pollutants in the workshop, and then obtain the predicted propagation trajectory.

[0172] The boundary determination module 60 is used to compare the predicted propagation trajectory with the real-time environmental parameter matrix, update the edge weights in the fluctuation correlation graph according to the differences, and determine the abnormal diffusion boundary based on the updated correlation graph.

[0173] The configuration generation module 70 is used to dynamically adjust the spatial distribution of the sensor array in the workshop and optimize the dot matrix layout configuration based on the abnormal diffusion boundary.

[0174] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An intelligent environmental monitoring method for meat processing, characterized in that, The methods include: Step 1: Collect environmental parameter data in real time by deploying a sensor array in the meat processing workshop and construct a real-time environmental parameter matrix; Step 2: Based on the real-time environmental parameter matrix, apply spatial interpolation algorithms to complete and smooth the data of irregularly distributed points to obtain a continuous spatial fluctuation field; Step 3: Divide the region based on the continuous spatial wave field, calculate the correlation coefficient matrix between adjacent regions, and generate a wave correlation diagram accordingly; Step 4: In the fluctuation correlation diagram, regions with correlation coefficients exceeding a preset reference threshold are marked as correlation source nodes. The initial propagation path set starting from the correlation source nodes is obtained through a path search algorithm. The correlation source nodes are potential sources of pollutants or other environmental changes. Step 5: Combine the initial set of propagation paths with ventilation flow data, and apply graph convolutional networks to simulate the diffusion dynamics of microorganisms and / or pollutants in the workshop, thereby obtaining the predicted propagation trajectory; Step 6: Compare the predicted propagation trajectory with the real-time environmental parameter matrix, update the edge weights in the fluctuation correlation graph according to the differences, and determine the abnormal propagation boundary based on the updated correlation graph. Step 7: Based on the abnormal diffusion boundary, dynamically adjust the spatial distribution of the sensor array within the workshop and optimize the dot matrix layout configuration; Step 5 includes: acquiring ventilation flow data, which includes at least the velocity and direction of airflow within the workshop; inputting the initial propagation path set and ventilation flow data into a pre-constructed graph convolutional network (GCNN), which uses a fluctuating correlation graph as its graph structure and environmental parameters as initial node features; aggregating neighborhood information through the convolutional layers of the GCNN, iteratively updating the feature representation of each node, and simulating the diffusion dynamics of microorganisms and / or pollutants along the graph structure driven by ventilation; extracting the initial propagation trajectory based on the simulation results, which represents the spatiotemporal sequence of microorganisms and / or pollutants spreading outward from the source node; integrating the initial propagation trajectory over a time step to calculate its cumulative spatial impact and obtain the trajectory intensity distribution; comparing the trajectory intensity distribution with the initial propagation path set, verifying the simulation accuracy based on the comparison results, and determining calibration parameters; adjusting the parameters of the GCNN according to the calibration parameters and re-simulating to obtain the predicted propagation trajectory containing the corrected diffusion boundary; Step 7 includes: classifying high-variability areas where the rate of change of environmental parameters exceeds a preset threshold as high-risk areas based on the abnormal diffusion boundary; increasing the number of sensor nodes deployed in any high-risk area based on its boundary area and a preset sensor coverage density; redistributing sensor nodes in the high-risk area based on the adjusted number of deployments to generate a preliminary dot matrix layout; calculating the average boundary distance based on the shortest Euclidean distance from each sensor node in the preliminary dot matrix layout to the abnormal diffusion boundary; determining whether the average boundary distance is greater than a preset distance threshold, and if so, iteratively adjusting the sensor node positions until the coverage requirements are met to obtain an iterative dot matrix layout; integrating sparse sampling in low-risk areas based on the iterative dot matrix layout to achieve a global balanced allocation of monitoring resources; simulating the data acquisition process under balanced allocation to obtain data acquisition efficiency indicators; determining whether the efficiency indicators are better than a preset standard, and if so, confirming that the current layout is an optimized dot matrix layout configuration; otherwise, iteratively adjusting the sensor node positions until a configuration scheme that meets the efficiency requirements is obtained.

2. The method according to claim 1, characterized in that, Step 1 includes: A sensor array is deployed in the workshop based on a three-dimensional spatial grid, and the sensor nodes collect temperature and humidity data, microbial concentration data, and pollutant concentration data in real time. The collected data is organized using three-dimensional spatial coordinates as an index to construct an initial environmental parameter matrix; For the initial environmental parameter matrix, a time series smoothing algorithm is applied to remove noise and obtain a smoothed environmental parameter matrix; Based on the smoothed environmental parameter matrix, the gradient distribution of environmental parameters in three-dimensional space is calculated, and the local rate of change of environmental parameters is determined. Identify high-variability regions where the local rate of change exceeds a preset rate of change threshold and mark them as priority monitoring points; The correspondence between priority monitoring points and sensor arrays is analyzed, and the data acquisition frequency of high-variability areas is dynamically adjusted based on the analysis results. New environmental parameter data are collected, and the initial environmental parameter matrix is ​​updated accordingly to obtain the real-time environmental parameter matrix.

3. The method according to claim 1, characterized in that, Step 2 includes: Based on the real-time environmental parameter matrix, identify irregular distribution points caused by missing or uneven data intervals; The Kriging spatial interpolation algorithm is applied to estimate the data of irregularly distributed points. The Kriging spatial interpolation algorithm estimates the environmental parameter values ​​of irregularly distributed points by weighted average based on the environmental parameter values ​​of adjacent known points. The weights are determined by spatial distance and variogram. The estimated values ​​are used to fill the real-time environmental parameter matrix to obtain the complete parameter matrix, and an initial continuous spatial fluctuation field is constructed based on this matrix using a surface fitting method. Calculate the gradient of the initial continuous spatial wave field, determine the wave direction and intensity, and obtain the wave vector field; Based on the analysis of the wave vector field, spatial connectivity is identified to identify potential wave propagation channels; The data on the wave propagation channel is compared with the original measured data at the corresponding position in the real-time environmental parameter matrix, and the deviation value is calculated. If the deviation value exceeds the preset deviation threshold, the parameters of the Kriging space interpolation algorithm are adjusted to correct the initial continuous space wave field and obtain the final continuous space wave field.

4. The method according to claim 1, characterized in that, Step 3 includes: The continuous spatial wave field is divided into regions based on a preset three-dimensional spatial grid; For each pair of adjacent regions, calculate the Pearson correlation coefficient between their environmental parameter sequences; Based on all Pearson correlation coefficients, a correlation coefficient matrix indexed by regional units is constructed. The correlation coefficient matrix is ​​filtered according to a preset significance threshold, and significant correlation pairs with coefficient values ​​exceeding the threshold are retained to obtain a simplified matrix. An initial fluctuation correlation graph is constructed based on a simplified matrix, with regions as nodes and significantly correlated pairs as edges. Calculate the degree of each node in the initial fluctuation correlation graph. The degree represents the number of other nodes associated with that node. The node with the highest degree is determined as the central region. The environmental parameter sequence of the central region is compared with the environmental parameter sequences of adjacent points in the continuous spatial fluctuation field. Based on the comparison results, the edge weights of the initial fluctuation correlation graph are updated to obtain the fluctuation correlation graph, which reflects the correlation strength of environmental parameter fluctuations.

5. The method according to claim 1, characterized in that, Step 4 includes: Based on historical environmental data, a preset reference threshold is set to identify nodes corresponding to areas where the correlation coefficient exceeds the preset reference threshold, and these nodes are marked as associated source nodes. For any associated source node, traverse all its connected edges in the oscillating association graph and construct an initial path starting from any associated source node; Based on path length and edge weight, calculate the path score for each initial path and filter out high-scoring paths with scores higher than a preset score threshold. Perform similarity analysis on all high-scoring paths, merge similar paths, and obtain an initial path set; Based on the initial path set, a depth-first search algorithm is applied to expand the potential branch paths to obtain an expanded path set. The extended path set is compared with the fluctuation correlation graph, cyclic paths are removed, and the effective path set is determined. Calculate the path score for each path in the set of valid paths, sort the set of valid paths according to the path scores, and obtain the initial propagation path set.

6. The method according to claim 1, characterized in that, Step 6 includes: The predicted propagation trajectory is compared with the real-time environmental parameter matrix, and the difference between the predicted value at the trajectory point and the measured value at the corresponding position in the matrix is ​​calculated. The difference value is compared with a preset difference threshold to identify outliers where the difference value exceeds the threshold. For outliers, the weights of adjacent edges in the fluctuation correlation graph are updated by merging the difference values ​​into the original edge weights in a weighted average manner. Based on the fluctuation correlation graph with updated weights, the correlation coefficients between regions are recalculated to obtain the corrected correlation coefficient matrix. Based on the modified correlation coefficient matrix, the connected component analysis method is used to extract the preliminary anomaly diffusion boundary; Morphological operations are performed on the initial abnormal diffusion boundary to smooth the boundary line and obtain a smooth boundary. The smooth boundary is compared with the predicted propagation trajectory to calculate the boundary deviation and determine the boundary confidence. High-confidence boundaries with a confidence level higher than a preset confidence threshold are selected and used as anomalous diffusion boundaries.

7. An intelligent environmental monitoring system for meat processing, used to implement the method as described in any one of claims 1 to 6, characterized in that, The system includes: The matrix construction module is used to collect environmental parameter data in real time through sensor arrays deployed in meat processing workshops and construct a real-time environmental parameter matrix. The preprocessing module is used to complete and smooth the data of irregularly distributed points by applying spatial interpolation algorithms based on the real-time environmental parameter matrix, so as to obtain a continuous spatial fluctuation field. The region division module is used to divide regions based on the continuous spatial fluctuation field, calculate the correlation coefficient matrix between adjacent regions, and generate a fluctuation correlation diagram accordingly. The path generation module is used to mark regions with correlation coefficients exceeding a preset reference threshold as associated source nodes in the fluctuation correlation diagram, and to obtain an initial set of propagation paths starting from the associated source nodes through a path search algorithm. The associated source nodes are potential sources of pollutants or other environmental changes. The trajectory prediction module combines the initial set of propagation paths with ventilation flow data, applies graph convolutional networks to simulate the diffusion dynamics of microorganisms and / or pollutants in the workshop, and then obtains the predicted propagation trajectory. The boundary determination module is used to compare the predicted propagation trajectory with the real-time environmental parameter matrix, update the edge weights in the fluctuation correlation graph according to the differences, and determine the abnormal propagation boundary based on the updated correlation graph. The configuration generation module is used to dynamically adjust the spatial distribution of the sensor array in the workshop based on the abnormal diffusion boundary, and optimize the dot matrix layout configuration.

Images