Intelligent monitoring, regulating and controlling method and system for temperature and humidity in silo

Abstract

The invention relates to the technical field of silo temperature and humidity monitoring, in particular to an intelligent monitoring, regulating and controlling method and system for the temperature and humidity in a silo, and the method comprises the steps: constructing a space coordinate system, and determining the position coordinates of each temperature and humidity sensor in a three-dimensional space; taking each temperature and humidity sensor as a node, and constructing a three-dimensional dynamic topological graph based on spatial correlation parameters of adjacent nodes; constructing a space-time joint feature set to form a comprehensive feature vector; and inputting the comprehensive feature vector into an abnormal heat source prediction model, and outputting a heat source abnormality judgment result of the center node. According to the invention, by constructing the three-dimensional space coordinate system of the temperature and humidity in the silo and combining the dynamic data processing and intelligent anomaly identification mechanism of the multi-point distributed temperature and humidity sensor, the problems of large temperature sampling spacing and delayed local thermal anomaly response in the prior art are solved; and early recognition and targeted regulation and control of the local latent heat source in the silo are realized.

Description

Technical Field

[0001] This invention relates to the field of silo temperature and humidity monitoring technology, specifically to a method and system for intelligent monitoring and control of temperature and humidity inside silos. Background Technology

[0002] In existing silo temperature and humidity monitoring and control systems, distributed temperature and humidity sensors are typically used to collect environmental parameters inside the silo in real time, and threshold control or fuzzy control strategies are combined to achieve automatic start and stop of ventilation, dehumidification and cooling equipment.

[0003] However, most existing systems are based on fixed-point monitoring and overall averaging control logic. Their temperature sampling intervals and time sampling periods are insufficient to cover subtle thermal changes within the grain pile. When a local grain pile area experiences continuous heat release due to moisture accumulation, poor ventilation, or microbial metabolism, this heat is often confined to a small volume area, forming a micro-area temperature rise with a spatial scale of only tens of centimeters. Due to the low heat conduction rate and strong insulation of the upper grain layer in this area, temperature changes fail to propagate to nearby measuring points in a short time, thus failing to be captured by the system in a timely manner. Moreover, existing temperature and humidity data processing algorithms generally employ moving averages or outlier smoothing mechanisms to suppress measurement noise, causing the initial weak temperature rise signal to be judged as random disturbance and ignored. Simultaneously, the control system often relies on overall environmental trends in determining the moisture and heat balance, lacking adaptive sensitivity to local anomalies, leading to the continued spread of hot spots without identification. When local heat accumulates to a certain level, the superposition of moisture evaporation and oxidation reactions within the grain pile forms a latent spontaneous combustion zone characterized by high temperature, oxygen deficiency, but still active reactions. Once external airflow disturbances or ventilation are restored, the area may be instantly drawn in oxygen, potentially leading to spontaneous combustion or even an internal combustion accident inside the silo. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a method and system for intelligent monitoring and control of temperature and humidity inside silos.

[0005] This invention employs the following technical solution: a method for intelligent monitoring and control of temperature and humidity inside a silo, comprising:

[0006] A preset time window is used to collect raw temperature and humidity data from multiple points inside the silo based on distributed temperature and humidity sensors.

[0007] With the center point at the bottom of the silo as the origin, any two mutually perpendicular directions on the horizontal plane as the X and Y axes, and the vertical center line of the silo center as the Z axis, a spatial coordinate system is constructed to determine the position coordinates of each temperature and humidity sensor in three-dimensional space;

[0008] Calculate the spatial correlation parameters between any two adjacent temperature and humidity sensors;

[0009] Using each temperature and humidity sensor as a node, a three-dimensional dynamic topology map is constructed based on the spatial correlation parameters of adjacent nodes. The location coordinates are marked on the nodes, and the spatial correlation parameters are marked on the edges, forming a dynamic heat and humidity network that evolves over time.

[0010] Extract the temporal feature parameters of each node within a preset time window;

[0011] Any node is marked as the center node. Based on the topology graph, the edge spatial correlation parameters and temporal feature parameters between the node and its neighboring nodes are extracted. Combined with the temporal feature parameters of the center node, a spatial-temporal joint feature set is constructed to form a comprehensive feature vector.

[0012] The comprehensive feature vector is input into the abnormal heat source prediction model, and the heat source anomaly determination result of the central node is output. When the output result is "yes", the corresponding node is marked as an abnormal heat source node.

[0013] Traverse all nodes, summarize all nodes marked as abnormal heat sources, divide adjacent abnormal heat source nodes into the same abnormal node cluster, forming M mutually independent abnormal node clusters.

[0014] The thermal gradient change rate and humidity diffusion trend within each abnormal node cluster are calculated. The heat source center point is extracted based on a multi-scale spatiotemporal clustering algorithm. The local air supply and dehumidification paths of the control system are updated based on the heat source center point.

[0015] As a further description of the above technical solution: the spatial correlation parameters include temperature and humidity difference values ​​and relative spatial distance gradient information;

[0016] The method for obtaining the temperature and humidity difference is as follows: obtain the temperature and humidity values ​​of two adjacent temperature and humidity sensors, and obtain the difference between the temperature values ​​and the difference between the humidity values ​​of the two temperature and humidity sensors, thus obtaining the temperature and humidity difference.

[0017] As a further description of the above technical solution: the method for obtaining the spatial distance gradient information includes:

[0018] Obtain the position coordinates of two adjacent temperature and humidity sensors, denoted as follows: and ,in, and Adjacent temperature and humidity sensors Temperature and humidity sensors -1 spatial coordinates;

[0019] The spatial distance is calculated using the Euclidean distance formula between three-dimensional coordinates and marked as the target spatial distance.

[0020] Calculate and obtain temperature and humidity sensor With temperature and humidity sensor +1 spatial distance and temperature and humidity sensor -1 and temperature and humidity sensor The spatial distance of -2 is marked as the first spatial distance and the second spatial distance, respectively.

[0021] The first spatial distance gradient is obtained by subtracting the first spatial distance from the target distance and then dividing the result by the target distance. The second spatial distance gradient is obtained by subtracting the second spatial distance from the target distance and then dividing the result by the target distance.

[0022] The average of the first spatial distance gradient and the second spatial distance gradient is obtained to obtain the values ​​of adjacent temperature and humidity sensors. Temperature and humidity sensors -1 spatial distance gradient.

[0023] As a further description of the above technical solution: the time characteristic parameters include the temperature change rate and humidity change rate of the node.

[0024] As a further description of the above technical solution: the method for constructing a joint spatial-temporal feature set to form a comprehensive feature vector includes:

[0025] By combining the spatial correlation parameters and temporal feature parameters of each adjacent node, a high-dimensional comprehensive feature vector is formed, representing the spatial-temporal relationship between the central node and adjacent nodes. This comprehensive feature vector is expressed as: In the formula, For nodes The corresponding feature vector, This is the temperature difference value. This is the humidity difference. The spatial distance gradient, For nodes rate of temperature change For nodes The rate of change in humidity.

[0026] As a further description of the above technical solution: the training method of the abnormal heat source prediction model includes:

[0027] Q sets of training data are collected in advance, where Q is an integer greater than 0. The training data includes the comprehensive feature vector and whether the corresponding center node is an abnormal heat source node. If the center node corresponds to a real abnormal heat source, the label is set to 1; otherwise, it is set to 0.

[0028] Gradient boosting tree classifier is used as the abnormal heat source prediction model. Initial hyperparameters are set, and the collected training data is divided into training set, validation set and test set according to a preset ratio.

[0029] The model is trained using a training set, employing a binary cross-entropy loss function. Leaf node weights are optimized using gradient descent, and model parameters are updated based on the negative gradient of the training set loss. Hyperparameters are tuned using Bayesian optimization. The validation set F1 score is calculated every 20 trees. Training is stopped when the F1 score improves by less than 0.01 for three consecutive iterations. The trained model is then evaluated using a test set. Once the model performance meets the evaluation criteria, it is deployed and applied.

[0030] As a further description of the above technical solution: the method of dividing adjacent abnormal heat source nodes into the same abnormal node cluster to form M mutually independent abnormal node clusters includes: summarizing the spatial coordinates of all nodes marked as abnormal heat source nodes, constructing an abnormal heat source node set, mapping the obtained abnormal heat source nodes to the topology graph, obtaining adjacent abnormal heat source nodes based on the constructed topology graph, and constructing multiple adjacent abnormal heat source nodes into an independent abnormal node cluster.

[0031] As a further description of the above technical solution: the method for extracting the center point of the heat source includes:

[0032] The temperature change rate sequence and humidity change rate sequence of each node in the abnormal node cluster are arranged side by side along the time axis and concatenated to form the joint time series feature vector of the node. A node sequence matrix is ​​formed between the nodes, where the rows correspond to the nodes and the columns correspond to the time series feature vectors.

[0033] The mean of each column of the joint feature matrix is ​​removed and scaled by the standard deviation to obtain the standardized matrix, which is the standardized time series feature vector.

[0034] Construct a comprehensive spatiotemporal similarity score for any two nodes;

[0035] Based on the comprehensive spatiotemporal similarity score, the comprehensive thermal potential score of each node in the abnormal node cluster is calculated. Based on the comprehensive thermal potential scores of all nodes in the abnormal node cluster, the weighted center position of the abnormal cluster is calculated and marked as the center point of the heat source.

[0036] As a further description of the above technical solution: the method for constructing a comprehensive spatiotemporal similarity score for any two nodes includes: first calculating the directional weighted spatial distance between the nodes to obtain spatial similarity, then calculating temporal similarity based on standardized temporal vectors, and finally fusing spatial similarity and temporal similarity according to weights to obtain a comprehensive spatiotemporal similarity score.

[0037] A silo-based intelligent temperature and humidity monitoring and control system, used to implement the aforementioned intelligent temperature and humidity monitoring and control method within the silo, the system comprising:

[0038] The data acquisition module, with a preset time window, collects raw temperature and humidity data from multiple points inside the silo based on distributed temperature and humidity sensors;

[0039] The coordinate construction module uses the center point at the bottom of the silo as the origin, any two mutually perpendicular directions on the horizontal plane as the X and Y axes, and the vertical center line of the silo center as the Z axis to construct a spatial coordinate system and determine the position coordinates of each temperature and humidity sensor in three-dimensional space.

[0040] The first parameter calculation module calculates the spatial correlation parameters between any two adjacent temperature and humidity sensors.

[0041] The topology construction module uses each temperature and humidity sensor as a node and constructs a three-dimensional dynamic topology graph based on the spatial correlation parameters of adjacent nodes. The position coordinates are marked on the nodes and the spatial correlation parameters are marked on the edges to form a dynamic thermal and humidity network that evolves over time.

[0042] The second parameter calculation module extracts the time feature parameters of each node within a preset time window;

[0043] The vector construction module marks any node as the center node, extracts the edge spatial correlation parameters and temporal feature parameters between the node and its neighboring nodes based on the topology graph, and constructs a spatial-temporal joint feature set by combining the temporal feature parameters of the center node to form a comprehensive feature vector.

[0044] The node marking module inputs the comprehensive feature vector into the abnormal heat source prediction model and outputs the heat source anomaly determination result of the central node. When the output result is "yes", the corresponding node is marked as an abnormal heat source node.

[0045] The node cluster generation module traverses all nodes, summarizes all nodes marked as abnormal heat sources, divides adjacent abnormal heat source nodes into the same abnormal node cluster, and forms M mutually independent abnormal node clusters.

[0046] The heat source center extraction module calculates the rate of change of thermal gradient and humidity diffusion trend within each abnormal node cluster, extracts the heat source center point based on a multi-scale spatiotemporal clustering algorithm, and updates the local air supply and dehumidification paths of the control system based on the heat source center point.

[0047] The beneficial effects of this invention are as follows:

[0048] The intelligent monitoring and control method for temperature and humidity inside silos provided by this invention overcomes the problems of large temperature sampling interval, uneven spatial coverage, delayed response to local thermal anomalies, and control homogenization in the prior art by constructing a three-dimensional spatial coordinate system for temperature and humidity inside the silo and combining dynamic data processing and intelligent anomaly identification mechanism of multi-point distributed temperature and humidity sensors. It realizes early identification and targeted control of local latent heat sources inside the silo.

[0049] Secondly, this invention uses the temperature change rate and humidity change rate of nodes as time feature parameters, and combines them with spatial correlation parameters to form a spatial-temporal joint feature vector, thereby realizing multi-dimensional fusion of thermal and humidity data. By introducing this comprehensive feature, the system can identify potential abnormal areas with a slow heat accumulation trend, avoiding the problem of traditional moving average or noise smoothing algorithms misjudging initial thermal anomalies as random disturbances, thus improving the system's sensitivity to weak heat source signals. Attached Figure Description

[0050] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0051] Figure 1 A flowchart of the intelligent monitoring and control method for temperature and humidity inside a silo provided in Embodiment 1 of the present invention;

[0052] Figure 2 This is a method for obtaining abnormal node clusters provided in Embodiment 1 of the present invention;

[0053] Figure 3 This is a method for extracting the center point of a heat source provided in Embodiment 1 of the present invention;

[0054] Figure 4 This is a module connection diagram of the intelligent monitoring and control system for temperature and humidity inside the silo provided in Embodiment 2 of the present invention. Detailed Implementation

[0055] To make the technical means, creative features, objectives, and effects of this invention readily understandable, the invention is further described below with reference to specific illustrations. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0056] Example 1

[0057] Please see Figures 1-3 This invention provides a technical solution: an intelligent monitoring and control method for temperature and humidity inside a silo, which is based on distributed temperature and humidity sensors installed inside the silo. The method includes:

[0058] A preset time window is used to collect raw temperature and humidity data from multiple points inside the silo based on distributed temperature and humidity sensors.

[0059] With the center point at the bottom of the silo as the origin, any two mutually perpendicular directions on the horizontal plane are the X and Y axes, and the vertical center line of the silo center is the Z axis. A spatial coordinate system is constructed to determine the position coordinates of each temperature and humidity sensor in three-dimensional space, providing structural constraints for subsequent spatial modeling.

[0060] Calculate the spatial correlation parameters between any two adjacent temperature and humidity sensors to characterize the intensity of local fluctuations in the silo's internal environment;

[0061] The spatial correlation parameters include temperature and humidity differences and relative spatial distance gradient information;

[0062] The method for obtaining the temperature and humidity difference is as follows: obtain the temperature and humidity values ​​of two adjacent temperature and humidity sensors, and obtain the difference between the temperature values ​​and the difference between the humidity values ​​of the two temperature and humidity sensors, thus obtaining the temperature and humidity difference.

[0063] The method for obtaining the spatial distance gradient information includes:

[0064] Obtain the position coordinates of two adjacent temperature and humidity sensors, denoted as follows: and ,in, and Adjacent temperature and humidity sensors Temperature and humidity sensors -1 spatial coordinates;

[0065] The spatial distance is calculated using the Euclidean distance formula between three-dimensional coordinates and marked as the target spatial distance.

[0066] Optionally, the formula for calculating the target spatial distance is:

[0067] ;

[0068] In the formula, Temperature and humidity sensor Temperature and humidity sensors Spatial distance;

[0069] Calculate and obtain temperature and humidity sensor With temperature and humidity sensor +1 spatial distance and temperature and humidity sensor -1 and temperature and humidity sensor The spatial distance of -2 is marked as the first spatial distance and the second spatial distance, respectively.

[0070] The first spatial distance gradient is obtained by subtracting the first spatial distance from the target distance and then dividing the result by the target distance. The second spatial distance gradient is obtained by subtracting the second spatial distance from the target distance and then dividing the result by the target distance.

[0071] The average of the first spatial distance gradient and the second spatial distance gradient is obtained to obtain the values ​​of adjacent temperature and humidity sensors. Temperature and humidity sensors -1 spatial distance gradient.

[0072] It should be noted that if there are no adjacent sensors connected to each other, a smooth interpolation method is used to predict the missing spatial gradient using spatial location or distance information from other adjacent sensors.

[0073] In this embodiment, by calculating the temperature and humidity difference and spatial distance gradient between adjacent temperature and humidity sensors, a spatial correlation parameter is formed, which realizes a quantitative description of the spatial propagation characteristics of heat and humidity changes. This parameter can comprehensively reflect the changing trend of temperature and humidity and the relative spatial distribution relationship, breaking through the limitations of traditional fixed threshold judgment, and enabling the system to identify micro-areas that have not yet shown obvious temperature rise but have potential heat accumulation risks in advance.

[0074] Using each temperature and humidity sensor as a node, a three-dimensional dynamic topology map is constructed based on the spatial correlation parameters of adjacent nodes. The location coordinates are marked on the nodes, and the spatial correlation parameters are marked on the edges, forming a dynamic heat and humidity network that evolves over time.

[0075] In this embodiment, each temperature and humidity sensor is used as a node, and a three-dimensional dynamic topology graph is constructed based on spatial correlation parameters. Spatial coordinates and correlation information are marked on the nodes and edges respectively to achieve structured modeling of temperature and humidity distribution.

[0076] This topology can be dynamically updated over time, reflecting the evolution of the thermal and humid field inside the silo. It makes up for the shortcomings of existing technologies, such as isolated data and inability to reflect spatial correlation, and provides spatiotemporal structural features to support the anomaly identification model.

[0077] Extract the temporal feature parameters of each node within a preset time window to form a temporal feature matrix that reflects local dynamics;

[0078] The time characteristic parameters include the node's temperature change rate and humidity change rate;

[0079] It should be noted that the method for obtaining the temperature change rate includes: obtaining the temperature value at the beginning and end of the preset time window, obtaining the temperature difference, placing the temperature difference within the preset time window length, and obtaining its temperature change rate.

[0080] The method for obtaining the humidity change rate includes: obtaining the humidity value at the beginning and end of a preset time window, obtaining the humidity difference, placing the humidity difference within the preset time window length, and obtaining its humidity change rate.

[0081] Any node is marked as the center node. Based on the topology graph, the edge spatial correlation parameters and temporal feature parameters between the node and its neighboring nodes are extracted. Combined with the temporal feature parameters of the center node, a spatial-temporal joint feature set is constructed to form a comprehensive feature vector.

[0082] Specifically, the method for constructing a joint spatial-temporal feature set to form a comprehensive feature vector includes:

[0083] By combining the spatial correlation parameters and temporal feature parameters of each adjacent node, a high-dimensional comprehensive feature vector is formed, representing the spatial-temporal relationship between the central node and adjacent nodes. This comprehensive feature vector is expressed as: In the formula, For nodes The corresponding feature vector, This is the temperature difference value. This is the humidity difference. The spatial distance gradient, For nodes rate of temperature change For nodes The rate of change in humidity.

[0084] The comprehensive feature vector is input into the abnormal heat source prediction model, and the heat source anomaly determination result of the central node is output. When the output result is "yes", the corresponding node is marked as an abnormal heat source node.

[0085] Specifically, the training method for the abnormal heat source prediction model includes:

[0086] Q sets of training data are collected in advance, where Q is an integer greater than 0. The training data includes a comprehensive feature vector and whether the corresponding center node is an abnormal heat source node. If the center node corresponds to a real abnormal heat source, the label is set to 1; otherwise, it is set to 0. The feature vector is standardized / normalized to remove abnormal samples. Stratified sampling is used to divide the training data into training set, validation set and test set according to a preset ratio (e.g., 6:3:1) to ensure that the class distribution of each dataset is consistent.

[0087] Gradient boosting tree classifier (GBDT classifier) ​​was used as the abnormal heat source prediction model. The initial hyperparameters were set as follows: the number of decision trees was 100-200, the maximum depth of a single tree was 5-8 (to control overfitting), the minimum number of samples for node splitting was 10, the maximum number of features considered during splitting was 3, and the Gini impurity criterion was adopted for splitting.

[0088] Model training and hyperparameter tuning: The model is trained using the training set, with a binary cross-entropy loss function as the loss function. The weights of the leaf nodes are optimized using gradient descent, and the model parameters are updated based on the negative gradient of the training set loss (the classification residual of the previous model is fitted to each new tree). The hyperparameters are tuned using Bayesian optimization, with the following optimization ranges: number of trees 120-180, learning rate 0.01-0.1, and regularization coefficient 0.05-0.2.

[0089] Early stopping mechanism and model selection: Calculate the validation set F1 score every 20 trees in each iteration; when the F1 score improves by <0.01 for 3 consecutive iterations (60 trees), stop training to avoid model overfitting; save the model parameters with the highest validation set F1 score (including the splitting rules of all decision trees) to ensure that there is still a stable ability to determine the types of abnormal heat sources that have not been seen before.

[0090] Model evaluation and deployment: The trained model is evaluated using a test set, and the precision, recall, F1 score and confusion matrix are calculated. If the precision is ≥90%, the recall is ≥95%, and the F1 score is ≥92%, the model performance evaluation is qualified and it can be deployed and applied.

[0091] Traverse all nodes, summarize all nodes marked as abnormal heat sources, divide adjacent abnormal heat source nodes into the same abnormal node cluster, forming M mutually independent abnormal node clusters.

[0092] Specifically, the method for dividing adjacent abnormal heat source nodes into the same abnormal node cluster to form M mutually independent abnormal node clusters includes: summarizing the spatial coordinates of all nodes marked as abnormal heat source nodes, constructing an abnormal heat source node set, mapping the obtained abnormal heat source nodes to the topology graph, obtaining adjacent abnormal heat source nodes based on the constructed topology graph, and constructing several adjacent abnormal heat source nodes into an independent abnormal node cluster.

[0093] In this embodiment, by using an abnormal heat source prediction model trained based on comprehensive feature vectors, this application can make real-time judgments on each node in the silo, and perform spatial aggregation and connectivity analysis on the nodes whose prediction results are abnormal, dividing adjacent abnormal nodes into the same abnormal node cluster. This clustering process effectively avoids false alarms or missed alarms caused by single-point fluctuations, and enhances the recognition accuracy of local abnormal distribution.

[0094] The thermal gradient change rate and humidity diffusion trend within each abnormal node cluster are calculated. The heat source center point is extracted based on a multi-scale spatiotemporal clustering algorithm. The local air supply and dehumidification paths of the control system are updated based on the heat source center point.

[0095] Specifically, methods for calculating the rate of change of thermal gradient and humidity diffusion trend within each anomalous node cluster, and for extracting the center point of the heat source based on a multi-scale spatiotemporal clustering algorithm, include:

[0096] The temperature change rate sequence and humidity change rate sequence of each node in the abnormal node cluster are arranged side by side along the time axis and concatenated to form the joint time series feature vector of the node. A node sequence matrix is ​​formed between the nodes, where the rows correspond to the nodes and the columns correspond to the time series feature vectors.

[0097] The mean of each column of the joint feature matrix is ​​removed and scaled by the standard deviation to obtain the standardized matrix, which is the standardized time series feature vector.

[0098] It should be noted that calculating the mean and standard deviation of each parameter and converting the raw data into Z-score standardized values ​​are existing techniques and will not be elaborated here.

[0099] Construct a comprehensive spatiotemporal similarity score for any two nodes;

[0100] The method for constructing the comprehensive spatiotemporal similarity score includes: first calculating the directional weighted spatial distance between nodes to obtain spatial similarity, then calculating temporal similarity based on standardized temporal vectors, and finally fusing spatial similarity and temporal similarity according to weights to obtain the comprehensive spatiotemporal similarity score;

[0101] Example: using nodes and nodes For example, the formula for calculating spatial similarity is:

[0102] ;

[0103] ;

[0104] In the formula, For nodes and nodes Spatial weighted distance, For nodes spatial coordinates, For nodes spatial coordinates, The weights are in the horizontal direction. Weights in the numerical direction; For nodes and nodes Spatial similarity, This is a spatial scale parameter, which can be taken as a multiple of the typical spacing of temperature and humidity sensors.

[0105] It should be noted that when vertical heat transfer is weaker than lateral convection, it can be set as follows: > This is to adjust the influence of vertical distance.

[0106] Example: using nodes and nodes For example, the formula for calculating temporal similarity is:

[0107]

[0108] In the formula, For nodes and nodes Temporal similarity, and For standardized nodes and nodes The temporal feature vector;

[0109] The formula for calculating the comprehensive spatiotemporal similarity score by weighted fusion of spatial similarity and temporal similarity is as follows:

[0110]

[0111] In the formula, For nodes and nodes The comprehensive spatiotemporal similarity score, + =1, , These are the weighting coefficients;

[0112] It should be noted that the weighting coefficients mentioned above are set by those skilled in the art based on actual conditions or obtained through simulation with a large amount of data. The size of the weighting coefficient is a specific value obtained by quantifying each parameter to facilitate subsequent comparison. The size of the weighting coefficient depends on the amount of sample data and the processing coefficients initially set by those skilled in the art for each set of sample data. As long as it does not affect the proportional relationship between the parameter and the quantified value, it is acceptable.

[0113] Based on the comprehensive spatiotemporal similarity score, the comprehensive thermal potential score of each node in the abnormal node cluster is calculated and obtained;

[0114] ;

[0115] For nodes The overall thermal potential score, For nodes The average spatiotemporal similarity score with other nodes indicates that the smaller the average spatiotemporal similarity score, the stronger the local heat source clustering. For nodes Rate of temperature change For nodes Humidity change rate , and These are the weighting coefficients.

[0116] It should be noted that the formulas mentioned above are all numerical calculations after removing the dimensions. They are formulas that are closest to the real situation, obtained by software simulation based on a large amount of data. The weight coefficients in the formulas are set by those skilled in the art based on the actual situation or obtained by simulation based on a large amount of data.

[0117] in, N is the number of nodes in the abnormal node cluster. For nodes and nodes The comprehensive spatiotemporal similarity score;

[0118] Based on the comprehensive thermal potential score of all nodes within the anomalous node cluster, the weighted center position of the anomalous cluster is calculated, and the weighted center position is marked as the heat source center point.

[0119] Optionally, the formula for calculating the center point of the heat source is:

[0120]

[0121] In the formula, Let n be the center point of the heat source, and n be the number of nodes within the anomalous node cluster. For nodes The overall thermal potential score, For nodes The three-dimensional coordinates.

[0122] In this embodiment, within the cluster of anomalous nodes, the rate of change of thermal gradient and the humidity diffusion trend of each node are calculated. A multi-scale spatiotemporal clustering algorithm is used to identify the thermal potential kernel, and combined with a weighted evaluation of thermal potential, the location of the heat source center is determined. Through comprehensive analysis of the direction of heat diffusion and the trend of humidity migration at different spatial scales, this application can accurately identify the true heat source center, significantly improving the accuracy and stability of anomaly localization.

[0123] Based on the identified heat source center point, the local air supply and dehumidification paths are dynamically adjusted to achieve directional ventilation and dehumidification control of specific thermal anomaly areas inside the silo. This local adaptive control strategy can effectively improve the system's energy efficiency and response speed, prevent the continuous spread of local heat accumulation, and reduce the potential risk of spontaneous combustion inside the silo.

[0124] Example 2

[0125] Please see Figure 3 This invention provides a technical solution: an intelligent monitoring and control system for temperature and humidity inside a silo, used to implement the aforementioned intelligent monitoring and control method for temperature and humidity inside a silo. The system includes:

[0126] The data acquisition module, with a preset time window, collects raw temperature and humidity data from multiple points inside the silo based on distributed temperature and humidity sensors;

[0127] The coordinate construction module uses the center point at the bottom of the silo as the origin, any two mutually perpendicular directions on the horizontal plane as the X and Y axes, and the vertical center line of the silo center as the Z axis to construct a spatial coordinate system and determine the position coordinates of each temperature and humidity sensor in three-dimensional space.

[0128] The first parameter calculation module calculates the spatial correlation parameters between any two adjacent temperature and humidity sensors.

[0129] The topology construction module uses each temperature and humidity sensor as a node and constructs a three-dimensional dynamic topology graph based on the spatial correlation parameters of adjacent nodes. The position coordinates are marked on the nodes and the spatial correlation parameters are marked on the edges to form a dynamic thermal and humidity network that evolves over time.

[0130] The second parameter calculation module extracts the time feature parameters of each node within a preset time window;

[0131] The vector construction module marks any node as the center node, extracts the edge spatial correlation parameters and temporal feature parameters between the node and its neighboring nodes based on the topology graph, and constructs a spatial-temporal joint feature set by combining the temporal feature parameters of the center node to form a comprehensive feature vector.

[0132] The node marking module inputs the comprehensive feature vector into the abnormal heat source prediction model and outputs the heat source anomaly determination result of the central node. When the output result is "yes", the corresponding node is marked as an abnormal heat source node.

[0133] The node cluster generation module traverses all nodes, summarizes all nodes marked as abnormal heat sources, divides adjacent abnormal heat source nodes into the same abnormal node cluster, and forms M mutually independent abnormal node clusters.

[0134] The heat source center extraction module calculates the rate of change of thermal gradient and humidity diffusion trend within each abnormal node cluster, extracts the heat source center point based on a multi-scale spatiotemporal clustering algorithm, and updates the local air supply and dehumidification paths of the control system based on the heat source center point.

[0135] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent monitoring and control of temperature and humidity inside a silo, characterized in that, include: A preset time window is used to collect raw temperature and humidity data from multiple points inside the silo based on distributed temperature and humidity sensors. With the center point at the bottom of the silo as the origin, any two mutually perpendicular directions on the horizontal plane as the X and Y axes, and the vertical center line of the silo center as the Z axis, a spatial coordinate system is constructed to determine the position coordinates of each temperature and humidity sensor in three-dimensional space; Calculate the spatial correlation parameters between any two adjacent temperature and humidity sensors; Using each temperature and humidity sensor as a node, a three-dimensional dynamic topology map is constructed based on the spatial correlation parameters of adjacent nodes. The location coordinates are marked on the nodes, and the spatial correlation parameters are marked on the edges, forming a dynamic heat and humidity network that evolves over time. Extract the temporal feature parameters of each node within a preset time window; Any node is marked as the center node. Based on the topology graph, the edge spatial correlation parameters and temporal feature parameters between the node and its neighboring nodes are extracted. Combined with the temporal feature parameters of the center node, a spatial-temporal joint feature set is constructed to form a comprehensive feature vector. The comprehensive feature vector is input into the abnormal heat source prediction model, and the heat source anomaly determination result of the central node is output. When the output result is "yes", the corresponding node is marked as an abnormal heat source node. Traverse all nodes, summarize all nodes marked as abnormal heat sources, divide adjacent abnormal heat source nodes into the same abnormal node cluster, forming M mutually independent abnormal node clusters. The thermal gradient change rate and humidity diffusion trend within each abnormal node cluster are calculated. The heat source center point is extracted based on a multi-scale spatiotemporal clustering algorithm. The local air supply and dehumidification paths of the control system are updated based on the heat source center point.

2. The intelligent monitoring and control method for temperature and humidity inside a silo according to claim 1, characterized in that, The spatial correlation parameters include temperature and humidity differences and relative spatial distance gradient information; The method for obtaining the temperature and humidity difference is as follows: obtain the temperature and humidity values ​​of two adjacent temperature and humidity sensors, and obtain the difference between the temperature values ​​and the difference between the humidity values ​​of the two temperature and humidity sensors, thus obtaining the temperature and humidity difference.

3. The intelligent monitoring and control method for temperature and humidity inside the silo according to claim 2, characterized in that, The method for obtaining the spatial distance gradient information includes: Obtain the position coordinates of two adjacent temperature and humidity sensors, denoted as follows: and ,in, and Adjacent temperature and humidity sensors Temperature and humidity sensors -1 spatial coordinates; The spatial distance is calculated using the Euclidean distance formula between three-dimensional coordinates and marked as the target spatial distance. Calculate and obtain temperature and humidity sensor With temperature and humidity sensor +1 spatial distance and temperature and humidity sensor -1 and temperature and humidity sensor The spatial distance of -2 is marked as the first spatial distance and the second spatial distance, respectively. The first spatial distance gradient is obtained by subtracting the first spatial distance from the target distance and then dividing the result by the target distance. The second spatial distance gradient is obtained by subtracting the second spatial distance from the target distance and then dividing the result by the target distance. The average of the first spatial distance gradient and the second spatial distance gradient is obtained to obtain the values ​​of adjacent temperature and humidity sensors. Temperature and humidity sensors -1 spatial distance gradient.

4. The intelligent monitoring and control method for temperature and humidity inside a silo according to claim 1, characterized in that, The time characteristic parameters include the node's temperature change rate and humidity change rate.

5. The intelligent monitoring and control method for temperature and humidity inside a silo according to claim 1, characterized in that, The method for constructing a joint spatial-temporal feature set to form a comprehensive feature vector includes: By combining the spatial correlation parameters and temporal feature parameters of each adjacent node, a high-dimensional comprehensive feature vector is formed, representing the spatial-temporal relationship between the central node and adjacent nodes. This comprehensive feature vector is expressed as: In the formula, For nodes The corresponding feature vector, This is the temperature difference value. This is the humidity difference value. The spatial distance gradient, For nodes rate of temperature change For nodes The rate of change in humidity.

6. The intelligent monitoring and control method for temperature and humidity inside a silo according to claim 5, characterized in that, The training method for the abnormal heat source prediction model includes: Q sets of training data are collected in advance, where Q is an integer greater than 0. The training data includes the comprehensive feature vector and whether the corresponding center node is an abnormal heat source node. If the center node corresponds to a real abnormal heat source, the label is set to 1; otherwise, it is set to 0. Gradient boosting tree classifier is used as the abnormal heat source prediction model. Initial hyperparameters are set, and the collected training data is divided into training set, validation set and test set according to a preset ratio. The model is trained using a training set, employing a binary cross-entropy loss function. Leaf node weights are optimized using gradient descent, and model parameters are updated based on the negative gradient of the training set loss. Hyperparameters are tuned using Bayesian optimization. The validation set F1 score is calculated every 20 trees. Training is stopped when the F1 score improves by less than 0.01 for three consecutive iterations. The trained model is then evaluated using a test set. Once the model performance meets the evaluation criteria, it is deployed and applied.

7. The intelligent monitoring and control method for temperature and humidity inside a silo according to claim 1, characterized in that, The method for dividing adjacent abnormal heat source nodes into the same abnormal node cluster to form M mutually independent abnormal node clusters includes: summarizing the spatial coordinates of all nodes marked as abnormal heat source nodes, constructing an abnormal heat source node set, mapping the obtained abnormal heat source nodes to a topology graph, obtaining adjacent abnormal heat source nodes based on the constructed topology graph, and constructing multiple adjacent abnormal heat source nodes into an independent abnormal node cluster.

8. The intelligent monitoring and control method for temperature and humidity inside a silo according to claim 1, characterized in that, The method for extracting the center point of the heat source includes: The temperature change rate sequence and humidity change rate sequence of each node in the abnormal node cluster are arranged side by side along the time axis and concatenated to form the joint time series feature vector of the node. A node sequence matrix is ​​formed between the nodes, where the rows correspond to the nodes and the columns correspond to the time series feature vectors. The mean of each column of the joint feature matrix is ​​removed and scaled by the standard deviation to obtain the standardized matrix, which is the standardized time series feature vector. Construct a comprehensive spatiotemporal similarity score for any two nodes; Based on the comprehensive spatiotemporal similarity score, the comprehensive thermal potential score of each node in the abnormal node cluster is calculated. Based on the comprehensive thermal potential scores of all nodes in the abnormal node cluster, the weighted center position of the abnormal cluster is calculated and marked as the center point of the heat source.

9. The intelligent monitoring and control method for temperature and humidity inside a silo according to claim 8, characterized in that, The method for constructing a comprehensive spatiotemporal similarity score for any two nodes includes: first calculating the directional weighted spatial distance between nodes to obtain spatial similarity; then calculating temporal similarity based on standardized temporal vectors; and finally fusing spatial similarity and temporal similarity according to weights to obtain a comprehensive spatiotemporal similarity score.

10. A silo internal temperature and humidity intelligent monitoring and control system, used to implement the silo internal temperature and humidity intelligent monitoring and control method according to any one of claims 1-9, characterized in that, The system includes: The data acquisition module, with a preset time window, collects raw temperature and humidity data from multiple points inside the silo based on distributed temperature and humidity sensors; The coordinate construction module uses the center point at the bottom of the silo as the origin, any two mutually perpendicular directions on the horizontal plane as the X and Y axes, and the vertical center line of the silo center as the Z axis to construct a spatial coordinate system and determine the position coordinates of each temperature and humidity sensor in three-dimensional space. The first parameter calculation module calculates the spatial correlation parameters between any two adjacent temperature and humidity sensors. The topology construction module uses each temperature and humidity sensor as a node and constructs a three-dimensional dynamic topology graph based on the spatial correlation parameters of adjacent nodes. The position coordinates are marked on the nodes and the spatial correlation parameters are marked on the edges to form a dynamic thermal and humidity network that evolves over time. The second parameter calculation module extracts the time feature parameters of each node within a preset time window; The vector construction module marks any node as the center node, extracts the edge spatial correlation parameters and temporal feature parameters between the node and its neighboring nodes based on the topology graph, and constructs a spatial-temporal joint feature set by combining the temporal feature parameters of the center node to form a comprehensive feature vector. The node marking module inputs the comprehensive feature vector into the abnormal heat source prediction model and outputs the heat source anomaly determination result of the central node. When the output result is "yes", the corresponding node is marked as an abnormal heat source node. The node cluster generation module traverses all nodes, summarizes all nodes marked as abnormal heat sources, divides adjacent abnormal heat source nodes into the same abnormal node cluster, and forms M mutually independent abnormal node clusters.

11. Heat source center extraction module: calculates the rate of change of thermal gradient and humidity diffusion trend in each abnormal node cluster, extracts the heat source center point based on multi-scale spatiotemporal clustering algorithm, and updates the local air supply and dehumidification paths of the control system based on the heat source center point.

Images