Real-time monitoring method and system for port cargo hazard sources based on the Internet of Things

Abstract

The present invention discloses a real-time monitoring method and system for port cargo hazard sources based on the Internet of Things, which is applied to the quality and safety monitoring of cold chain cargo storage areas in ports. The method comprises: configuring a temperature sensor grid, a humidity sensor grid and a thermal imaging device to collect the temperature distribution matrix, the humidity distribution matrix and the surface temperature image of the cold chain cargo area; using a recursive neural network to extract the time series variation characteristics of temperature and humidity, adopting a deep residual network to extract the spatial temperature gradient characteristics of the thermal imaging image, and performing feature fusion through an adaptive weighting mechanism; constructing the fused features into a time window sequence, and using a gradient boosting tree model to evaluate the abnormal risk level of the cargo status and locate the hazard source. The present invention can timely discover and accurately locate potential hazard sources, effectively improving the intelligence level and early warning capability of port cargo safety management.

Description

Technical Field

[0001] The present invention belongs to the technical field of port cargo monitoring, and in particular relates to a real-time monitoring method and system for port cargo hazard sources based on the Internet of Things. Background Art

[0002] Existing port cargo monitoring systems primarily rely on traditional point sensors for isolated data collection, which suffers from large blind spots and limited sensing capabilities. This isolated point-based monitoring approach is particularly difficult to detect localized temperature and humidity anomalies in a timely manner for temperature-sensitive cold chain cargo, resulting in the inability to effectively manage potential cargo quality and safety risks. Traditional monitoring systems typically analyze temperature and humidity as independent variables, ignoring their complex interactions and their correlation with cargo characteristics. This significantly reduces monitoring accuracy and the timeliness of early warnings.

[0003] Existing port cargo monitoring technologies lack the ability to integrate and intelligently analyze multi-source data, with most systems still stuck at the stage of simple threshold judgment and single-indicator monitoring. When cold chain cargo is stored in complex environments, single-indicator monitoring struggles to identify potential compound risk factors and accurately locate sources of danger, preventing managers from taking targeted measures when abnormalities occur. Furthermore, due to a lack of deep mining and learning capabilities for historical data, existing systems have limited ability to identify abnormal patterns, often leading to false alarms or missed alerts. While IoT and AI technologies are widely used in various fields, their integrated application in port cargo safety monitoring remains in its infancy.

[0004] Existing port cargo monitoring systems suffer from issues such as insufficient real-time performance, a limited early warning mechanism, and low intelligence, making them unable to meet the demands of modern ports for refined and intelligent management of cold chain cargo. When multiple batches and types of cold chain cargo are stored simultaneously, the lack of precise spatial positioning and risk grading significantly increases management complexity and potential loss risks. Summary of the Invention

[0005] The present invention achieves the purpose of real-time and accurate monitoring and early warning of temperature and humidity anomalies in the cold chain cargo storage area of the port by constructing an intelligent monitoring system that combines a multimodal sensor network with a deep learning algorithm.

[0006] The present invention provides a real-time monitoring method for hazardous sources of port cargo based on the Internet of Things, which is applied to the quality and safety monitoring of cold chain cargo storage areas in ports. The method comprises the following steps:

[0007] Step S1: Configuring the sampling intervals of the temperature sensor grid, the humidity sensor grid, and the thermal imaging device to be consistent, using the temperature sensor grid and the humidity sensor grid to collect the temperature distribution matrix and the humidity distribution matrix of the cold chain cargo area, and using the thermal imaging device to obtain the cargo surface temperature image;

[0008] Step S2: Using a recursive neural network to extract the temporal variation characteristics of temperature and humidity, a deep residual network to extract the spatial temperature gradient characteristics of the thermal image, and a variational autoencoder to perform feature dimensionality reduction; the obtained multimodal features are fused through a dynamic weight mechanism to construct a comprehensive feature vector;

[0009] In step S3, the fused comprehensive features are constructed into a time window sequence. Based on the pre-trained gradient boosting tree model, the abnormal risk level of the cargo status in the current monitoring period is evaluated, and the location of the hazard source is identified and given based on the abnormal risk level.

[0010] Furthermore, in step S2, the recurrent neural network adopts a gated recurrent unit GRU structure, comprising: two bidirectional GRU layers, each followed by a layer normalization and a Dropout layer, and a fully connected output layer; the input is The temperature and humidity distribution matrix sequence at each time point is output as a 128-dimensional feature vector;

[0011] The mathematical expression of the GRU unit is:

[0012] ; ;

[0013] ; ;

[0014] in, represents the update gate, Represents the reset gate, represents the candidate hidden state, Indicates the hidden state at the current moment, Represents the input sequence at time The value of 、 、 are the weight matrices of the corresponding gates, 、 、 is the corresponding bias term, is the sigmoid activation function, represents the Hadamard product.

[0015] Furthermore, in step S2, the deep residual network adopts the DenseNet-121 structure, which includes: 1 7x7 convolutional layer, 4 dense blocks, each dense block has a transition layer, 1 global average pooling layer and 1 fully connected layer; the input is the thermal imaging temperature map, and the output is a 256-dimensional feature vector;

[0016] The mathematical expression of the dense block is: ,in Represents a composite transformation function, including batch normalization, ReLU activation and convolution operations, Indicates the The output of the layer, Represents the concatenation of the outputs of all previous layers.

[0017] Furthermore, in step S2, the feature fusion adopts an adaptive weighting mechanism, and the feature fusion method includes: calculating the uncertainty weight of each modal feature, integrating the multimodal features using an attention gated network, and generating a context-related fusion representation;

[0018] The mathematical expression of the adaptive weighting mechanism is:

[0019]

[0020] The calculation expression of the uncertainty weight is: ;in, Indicates the Class features, Corresponding to the characteristics of different modes; represents the adaptive weight based on uncertainty; Representation characteristics The variance of , reflects the uncertainty of the feature; represents the attention gating coefficient, represents the global context vector; represents the weight matrix of the gating network, represents the bias vector; Represents the final fused feature representation.

[0021] Furthermore, in step S3, the gradient boosting tree model adopts the LightGBM framework, including 100 decision trees, each tree has a maximum depth of 6, a minimum number of leaf node samples of 20, and a learning rate of 0.05; the prediction function of the decision tree ensemble can be expressed as:

[0022] ; ;in, Representation sample The predicted value of Indicates the A decision tree, represents the input feature vector, represents the loss function, represents the training error term, represents the regularization term, which is used to control the complexity of the model. Representation sample The true label of

[0023] The model outputs anomaly risk probability distribution, which is divided into four levels: normal, slight anomaly, moderate anomaly, and severe anomaly. When the anomaly risk level reaches moderate anomaly or above for three consecutive time windows, the system will use cluster analysis to determine the spatial distribution characteristics of temperature and humidity anomalies and, combined with the cargo storage layout, accurately locate the potential hazard source.

[0024] Hazard Source Positioning is based on the DBSCAN density clustering algorithm, and its mathematical expression is:

[0025] , where D is the set of sensor data points, for point and point The Euclidean distance function, is the neighborhood radius.

[0026] Furthermore, in step S1, the data acquisition frequency of the temperature sensor grid, the humidity sensor grid and the thermal imaging device is set to 0.5 Hz; the spatial resolution of the temperature sensor grid is 12x12 dot matrix, the spatial resolution of the humidity sensor grid is 10x10 dot matrix, and the temperature image resolution collected by the thermal imaging device is 384x288 pixels.

[0027] Furthermore, in step S2, the input of the recursive neural network is the temperature and humidity data of m consecutive time points, the value range of m is 15 to 45, and the latent space dimension of the variational autoencoder is 64.

[0028] Furthermore, in step S3, the length of the time window sequence is K, the value range of K is 8 to 20, and the overlapping rate of the sliding window is 50%.

[0029] The classification of the abnormal cargo status risk level includes: when the temperature deviation does not exceed ±1.0°C of the set threshold and the humidity deviation does not exceed ±5%RH, and the temperature difference of the thermal imaging image does not exceed 3.0°C, it is a normal state.

[0030] A minor abnormality occurs when the temperature deviation is between ±1.0°C and ±2.0°C of the set threshold, or the humidity deviation is between ±5%RH and ±10%RH, or the detected temperature and humidity change rate in a local area exceeds 0.5°C / hour or 2%RH / hour, and the temperature difference in the thermal image does not exceed 3.0°C.

[0031] A moderate abnormality occurs when the temperature deviation is between ±2.0°C and ±4.0°C of the set threshold, or the humidity deviation is between ±10%RH and ±15%RH, or the detected temperature and humidity change rate in a local area exceeds 1.0°C / hour or 4%RH / hour, or the temperature difference in the thermal image is between 3.0°C and 6.0°C.

[0032] A serious anomaly occurs when the temperature deviation exceeds the set threshold of ±4.0°C or the humidity deviation exceeds ±15%RH, or the detected temperature and humidity change rate in a local area exceeds 2.0°C / hour or 8%RH / hour, or the temperature and humidity anomaly indicators show an accelerated deterioration trend within three consecutive time windows, or the temperature difference in the thermal imaging image exceeds 6.0°C.

[0033] Based on the same inventive concept, the present invention provides a real-time monitoring system for port cargo hazards based on the Internet of Things, which is used to implement the real-time monitoring method for port cargo hazards of the present invention. The system includes: a data acquisition module, a feature extraction module, a feature fusion module and a risk assessment module;

[0034] The data acquisition module is used to obtain the temperature distribution matrix, humidity distribution matrix and surface temperature image of the cold chain cargo area. The data acquisition module includes a temperature sensor grid, a humidity sensor grid and a thermal imaging device.

[0035] The feature extraction module is used to extract time series features and spatial temperature gradient features from the collected data, and the feature extraction module includes a recursive neural network and a deep residual network.

[0036] The feature fusion module is used to perform adaptive weighted fusion on the extracted features to generate a comprehensive feature representation. The feature fusion module adopts an adaptive weighting mechanism based on uncertainty.

[0037] The risk assessment module is used to derive the abnormal risk level of the cargo status in the current monitoring period based on the fused feature sequence using a gradient boosting tree model, and identify and provide the location of the hazard source based on the abnormal risk level.

[0038] Furthermore, the system also includes an early warning module for triggering an early warning mechanism when a moderate abnormality or a serious abnormality is detected.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] This invention achieves comprehensive perception, intelligent analysis, and precise early warning of cold chain cargo in ports by constructing an IoT architecture that integrates a multimodal sensor network and combines it with advanced deep learning and ensemble learning algorithms. The system uses a recursive neural network to extract the temporal variation characteristics of temperature and humidity, a deep residual network to extract the spatial temperature gradient characteristics of thermal imaging, and an adaptive weighting mechanism to fuse multimodal features. This effectively improves the accuracy and sensitivity of anomaly detection, accurately identifies abnormal areas, and significantly reduces manual investigation time and disposal costs. It provides comprehensive, real-time, and intelligent technical support for the safe management of cold chain cargo in ports, effectively reducing cargo damage rates and improving logistics efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of the method for real-time monitoring of dangerous sources of port cargo based on the Internet of Things of the present invention;

[0042] Figure 2 This is a schematic diagram of the composition of the real-time monitoring system for port cargo hazards based on the Internet of Things of the present invention. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only part of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] Example 1

[0045] like Figure 1 FIG. 1 is a flow chart of a method for real-time monitoring of hazardous sources of port cargo based on the Internet of Things according to the present invention, which is applied to the quality and safety monitoring of cold chain cargo storage areas in ports. The method includes the following steps:

[0046] Step S1: Configure the sampling intervals of the temperature sensor grid, humidity sensor grid and thermal imaging equipment to be consistent, use the temperature sensor grid and humidity sensor grid to collect the temperature distribution matrix and humidity distribution matrix of the cold chain cargo area, and use the thermal imaging equipment to obtain the cargo surface temperature image.

[0047] The data acquisition frequency of the temperature sensor grid, humidity sensor grid, and thermal imaging device was set to 0.5 Hz; the spatial resolution of the temperature sensor grid was 12x12 dots, the spatial resolution of the humidity sensor grid was 10x10 dots, and the resolution of the temperature image collected by the thermal imaging device was 384x288 pixels.

[0048] In the present invention, the sampling interval consistency design of the sensor grid is to ensure the alignment of multimodal data in the time dimension, which is crucial for subsequent feature fusion and anomaly detection. The sampling frequency of 0.5Hz is the optimal frequency verified by a large number of experiments. It can capture subtle trends in temperature and humidity changes without generating excessive data storage and transmission burdens. The 12x12 and 10x10 dot matrix resolution designs enable the system to monitor small spatial changes in temperature and humidity within the cold chain cargo storage area, especially for the temperature and humidity stratification phenomenon inside large containers. The thermal imaging resolution of 384x288 pixels can provide sufficiently detailed surface temperature distribution information, which helps to detect internal temperature anomalies that cannot be directly observed in the outer packaging of the goods.

[0049] In step S2, a recursive neural network is used to extract the temporal variation characteristics of temperature and humidity, a deep residual network is used to extract the spatial temperature gradient characteristics of the thermal image, and a variational autoencoder is used to perform feature dimensionality reduction; the obtained multimodal features are fused through a dynamic weight mechanism to construct a comprehensive feature vector.

[0050] In step S2, the recurrent neural network adopts a gated recurrent unit GRU structure, which includes: 2 bidirectional GRU layers, each followed by a layer normalization and Dropout layer, and a fully connected output layer; the input is The temperature and humidity distribution matrix sequence at each time point is output as a 128-dimensional feature vector;

[0051] The mathematical expression of the GRU unit is:

[0052] ; ;

[0053] ; ;

[0054] in, represents the update gate, Represents the reset gate, represents the candidate hidden state, Indicates the hidden state at the current moment, Represents the input sequence at time The value of 、 、 are the weight matrices of the corresponding gates, 、 、 is the corresponding bias term, is the sigmoid activation function, represents the Hadamard product.

[0055] The design concept of the GRU unit is to control the flow of information through update gates and reset gates. Compared to the traditional LSTM structure, the GRU has the advantages of fewer parameters and higher computational efficiency, making it particularly suitable for resource-constrained edge computing environments. In the processing of temperature and humidity time series data, the update gate $z_t$ determines the ratio of retaining historical information to accepting new information, while the reset gate $r_t$ controls the degree to which historical information affects the current state. When abnormal changes in temperature and humidity are detected, the update gate automatically adjusts its weight, making the model more focused on the abnormal signal of the current input. The bidirectional GRU design enables the model to consider both past and future information simultaneously, making it more comprehensive and accurate in judging abnormal conditions at the current moment.

[0056] The deep residual network uses the DenseNet-121 architecture, consisting of one 7x7 convolutional layer, four dense blocks with transition layers between each dense block, one global average pooling layer, and one fully connected layer. The input is a thermal image temperature map, and the output is a 256-dimensional feature vector. The DenseNet-121 network was chosen for processing thermal images because its unique dense connection structure offers excellent feature reuse and gradient flow properties. Compared to traditional convolutional neural networks, DenseNet maximizes feature utilization by directly connecting each layer to all subsequent layers, significantly reducing the number of parameters and improving computational efficiency. In processing thermal image temperature maps, the combination of shallow features (such as edges and textures) and deep features (such as semantic information) is particularly important. The dense connections enable the model to simultaneously utilize multi-scale features, effectively identifying temperature anomalies of varying sizes and shapes. In practical applications, this network architecture has achieved an accuracy rate of over 95% for detecting surface temperature anomalies on cargo.

[0057] The mathematical expression of the dense block is: ,in Represents a composite transformation function, including batch normalization, ReLU activation and convolution operations, Indicates the The output of the layer, Represents the concatenation of the outputs of all previous layers.

[0058] Feature fusion uses an adaptive weighting mechanism. The feature fusion method includes: calculating the uncertainty weight of each modal feature, integrating multimodal features using an attention gating network, and generating a context-dependent fusion representation;

[0059] The mathematical expression of the adaptive weighting mechanism is:

[0060]

[0061] The calculation expression of the uncertainty weight is: ;in, Indicates the Class features, Corresponding to the characteristics of different modes; represents the adaptive weight based on uncertainty; Representation characteristics The variance of , reflects the uncertainty of the feature; represents the attention gating coefficient, represents the global context vector; represents the weight matrix of the gating network, represents the bias vector; Represents the final fused feature representation. In cold chain cargo monitoring, data from different modalities may be affected by noise to varying degrees. For example, temperature sensors may be disturbed by local air flow, humidity sensors may be affected by condensation, and thermal imaging equipment may be affected by surface reflections and environmental factors. By calculating the variance of features to estimate their uncertainty, the system can automatically reduce the weights of noisy or unstable features and increase the contribution of more reliable features, thereby improving the overall quality of fused features and the accuracy of anomaly detection. The introduction of the attention gated network further enhances the context-awareness of feature fusion, enabling the system to dynamically adjust feature weights based on the currently observed environmental state.

[0062] The input to the recurrent neural network is temperature and humidity data from m consecutive time points, where m ranges from 15 to 45. The latent space dimension of the variational autoencoder is 64. The recurrent neural network processes temperature and humidity data from 15 to 45 consecutive time points, which is equivalent to monitoring the state changes of the goods for 7.5 to 22.5 minutes. The setting of this time window is based on the physical characteristics of the temperature and humidity changes of cold chain goods. Experience shows that the temperature and humidity anomalies of most cold chain goods will show a clear trend of change within this time range. A time window that is too short may miss slowly developing abnormal patterns, while a time window that is too long will increase the computational burden and reduce the real-time response capability of the system. The latent space dimension of the variational autoencoder is set to 64, which significantly reduces the feature dimension while maintaining sufficient information, reducing the complexity of the subsequent classification model and improving the generalization ability of the model.

[0063] In step S3, the fused comprehensive features are constructed into a time window sequence. Based on the pre-trained gradient boosting tree model, the abnormal risk level of the cargo status in the current monitoring period is evaluated, and the location of the hazard source is identified and given based on the abnormal risk level.

[0064] In cold chain cargo monitoring, certain abnormal patterns may not be apparent at a single point in time, but will exhibit specific development trajectories over time. The gradient boosting tree model, a powerful ensemble learning method capable of handling nonlinear feature relationships and unbalanced datasets, is particularly well-suited for scenarios such as cold chain cargo anomaly detection, where normal samples far outnumber abnormal ones. The pre-trained model utilizes a large amount of historical data, including data on normal operating conditions and various abnormalities, ensuring the model's ability to identify various abnormal patterns. The hazard source location function, based on abnormal risk levels, provides managers with precise intervention guidance, significantly reducing troubleshooting and resolution time.

[0065] In step S3, the gradient boosting tree model adopts the LightGBM framework. The selection of the LightGBM framework is based on its advantages in processing high-dimensional features and large-scale data sets. Compared with traditional gradient boosting tree algorithms such as XGBoost, LightGBM adopts a histogram-based decision tree algorithm and a leaf-first growth strategy, which greatly improves the training speed and memory efficiency. The ensemble setting of 100 decision trees showed the best balance between performance and efficiency in experimental verification. The maximum depth of the tree is 6 to prevent overfitting while maintaining sufficient model complexity. The setting of the minimum number of leaf node samples of 20 ensures the statistical stability of each leaf node and improves the generalization ability of the model. The learning rate of 0.05 enables the model to gradually learn the complex patterns of the data and avoid falling into local optimality. Contains 100 decision trees, each tree has a maximum depth of 6, a minimum number of leaf node samples of 20, and a learning rate of 0.05; the prediction function of the decision tree ensemble can be expressed as:

[0066] ; ;in, Representation sample The predicted value of Indicates the A decision tree, represents the input feature vector, represents the loss function, represents the training error term, represents the regularization term, which is used to control the complexity of the model. Representation sample The true label of

[0067] The model outputs anomaly risk probability distribution, which is divided into four levels: normal, minor, moderate, and severe. When the system detects anomaly risk levels reaching moderate or above for three consecutive time windows, it uses cluster analysis to determine the spatial distribution of temperature and humidity anomalies and, combined with the cargo storage layout, pinpoints the potential source of danger. The anomaly risk classification is based on the sensitivity of cold chain cargo to temperature and humidity fluctuations and the actual damage threshold. For example, for high-value pharmaceuticals or fresh food, even minor temperature deviations can lead to quality degradation or shortened shelf life. This four-level risk classification system provides sufficient granularity to avoid management fatigue caused by excessive alerts while ensuring timely response to major risks. The temperature deviation thresholds of ±1.0°C, ±2.0°C, and ±4.0°C are based on international cold chain standards and actual business needs. The humidity deviation thresholds of ±5%RH, ±10%RH, and ±15%RH account for the varying humidity sensitivities of different cargo types. The temperature and humidity change rate indicator is used to capture rapidly deteriorating anomalies, even if the absolute values remain within safe limits.

[0068] Hazard Source The positioning is based on the DBSCAN density clustering algorithm. The DBSCAN density clustering algorithm is suitable for hazard source positioning because it does not require the pre-specification of the number of clusters. It can automatically discover high-density areas of any shape and identify noise points. In actual port cold chain cargo monitoring, temperature and humidity anomalies often show local concentration characteristics, which may be caused by a single heat source or cold source. The DBSCAN algorithm calculates the Euclidean distance between data points and classifies density-connected points into the same cluster, thereby identifying the spatial distribution pattern of abnormal temperature and humidity. The neighborhood radius Eps is a key parameter that determines the maximum distance for points to be classified into the same cluster. Its mathematical expression is:

[0069] , where D is the set of sensor data points, for point and point The Euclidean distance function, is the neighborhood radius.

[0070] The length of the time window sequence is K, the value range of K is 8 to 20, and the overlap rate of the sliding window is 50%; the classification of the abnormal risk level of the cargo status includes:

[0071] The normal state is when the temperature deviation does not exceed ±1.0°C of the set threshold, the humidity deviation does not exceed ±5%RH, and the temperature difference of the thermal imaging image does not exceed 3.0°C.

[0072] A minor abnormality occurs when the temperature deviation is between ±1.0°C and ±2.0°C of the set threshold, or the humidity deviation is between ±5%RH and ±10%RH, or the detected temperature and humidity change rate in a local area exceeds 0.5°C / hour or 2%RH / hour, and the temperature difference in the thermal image does not exceed 3.0°C.

[0073] A moderate abnormality occurs when the temperature deviation is between ±2.0°C and ±4.0°C of the set threshold, or the humidity deviation is between ±10%RH and ±15%RH, or the detected temperature and humidity change rate in a local area exceeds 1.0°C / hour or 4%RH / hour, or the temperature difference in the thermal image is between 3.0°C and 6.0°C.

[0074] A serious anomaly occurs when the temperature deviation exceeds the set threshold of ±4.0°C or the humidity deviation exceeds ±15%RH, or the detected temperature and humidity change rate in a local area exceeds 2.0°C / hour or 8%RH / hour, or the temperature and humidity anomaly indicators show an accelerated deterioration trend within three consecutive time windows, or the temperature difference in the thermal imaging image exceeds 6.0°C.

[0075] The following is an actual application example of the present invention in a cold chain logistics center of a large coastal port. The logistics center covers an area of 30,000 square meters and is mainly used for storing temperature-sensitive goods such as imported and exported frozen foods, medicines, and biological products.

[0076] At this cold chain logistics center, the system deployed a total of 10 monitoring units, each covering approximately 500 square meters of cold storage area. The system features a 12×12 grid of temperature sensors (144 total sensor points), a 10×10 grid of humidity sensors (100 total sensor points), two high-precision thermal imaging devices (384×288 pixel resolution), and an edge computing unit (integrated data acquisition, preliminary processing, and transmission). All monitoring units are connected to a central control system via a low-power wide area network (LPWAN). The data acquisition frequency is uniformly set to 0.5Hz. All sensors are calibrated to ISO standards, achieving temperature measurement accuracy of ±0.1°C and humidity measurement accuracy of ±2%RH. The system utilizes a layered architecture, with the edge computing unit responsible for data preprocessing and primary feature extraction, while the central server handles high-level feature fusion, model inference, and risk assessment.

[0077] On August 15, 2022, a batch of high-value vaccines stored in the pharmaceutical area experienced abnormal temperatures. At 9:23:45 AM, the temperature sensor grid of monitoring unit 5 detected a slow temperature increase in a local area (coordinates E6-F8) at approximately 0.3°C / hour, which was within the system's preset static threshold (±1.0°C). Simultaneously, thermal imaging equipment captured surface temperature images of the area, revealing a subtle temperature gradient.

[0078] A recursive neural network (GRU structure) processed the temperature and humidity sequence data of the last 30 time points (about 15 minutes) and extracted a 128-dimensional time series feature vector; the DenseNet-121 network processed the thermal imaging images and extracted a 256-dimensional spatial temperature gradient feature vector; and the variational autoencoder reduced these features to a 64-dimensional latent space.

[0079] The system uses an adaptive weighting mechanism to calculate the uncertainty of each modal feature: the variance of the temperature time series feature is 0.023, the variance of the humidity time series feature is 0.085, and the variance of the heat map feature is 0.042. Based on these uncertainty values, the system calculates a weight of 0.52 for the temperature time series feature, 0.13 for the humidity time series feature, and 0.35 for the heat map feature. After feature fusion, a comprehensive feature vector is generated.

[0080] At 9:38:22 AM, based on a sequence analysis of a sliding time window (K=15, or 7.5 minutes), the gradient boosting tree model outputted the following anomaly risk probability distribution: normal state 0.12, minor anomaly 0.68, moderate anomaly 0.19, and severe anomaly 0.01. The system classified the risk level as "minor anomaly," generated a low-priority notification, and shortened the data processing interval to 15 seconds. At 10:05:47 AM, the system detected a continuous increase in anomaly probability for eight consecutive time windows. The latest risk probability distribution was: normal state 0.03, minor anomaly 0.24, moderate anomaly 0.63, and severe anomaly 0.10. The system raised the risk level to "moderate anomaly," triggering a Level 2 alert.

[0081] The system analyzed the spatial distribution of the temperature anomaly using the DBSCAN density clustering algorithm (with parameters Eps=2.5 and MinPts=4). It identified an anomaly cluster centered at coordinates E7-F7. Combined with the cargo storage layout, the system pinpointed the location of vaccine refrigerated containers with batch number VP22081203. The system automatically sent SMS, email, and mobile app push notifications to the responsible personnel and simultaneously marked the anomaly area in the monitoring center's digital twin system. Management arrived on-site within 10 minutes of receiving the notification and discovered that the refrigeration system of the vaccine refrigerated containers had malfunctioned, causing the internal temperature to slowly rise.

[0082] Comparative testing with traditional monitoring systems demonstrated the clear advantages of this invention: whereas the traditional fixed-threshold-based monitoring system did not trigger an alarm until the temperature deviation reached +1.2°C (at 11:10:35 a.m.), this system identified the abnormal trend when the temperature deviation was only +0.4°C (at 09:38:22 a.m.), discovering the potential problem approximately 90 minutes in advance.

[0083] During a six-month comparative test, the system achieved an anomaly detection accuracy rate of 94.7%, with a false alarm rate of only 3.2%. Traditional systems achieved an accuracy rate of 78.5%, with a false alarm rate as high as 15.6%. The system's advantages are particularly evident in detecting slow-moving temperature and humidity anomalies. The system can pinpoint the source of an anomaly to a specific cargo unit (with an accuracy of approximately 0.5 meters), while traditional systems can only indicate a general area (with an accuracy of approximately 5-10 meters). In the aforementioned case, the system directly pinpointed the location and batch number of the problematic refrigerated container, significantly reducing troubleshooting time. According to statistics from the logistics center, the use of this system has reduced cargo losses due to temperature and humidity anomalies by 68.4%.

[0084] Example 2

[0085] like Figure 2 The figure shows a schematic diagram of the composition of the real-time monitoring system for port cargo hazards based on the Internet of Things of the present invention, which is used to execute the real-time monitoring method for port cargo hazards of Example 1. The system includes: a data acquisition module, a feature extraction module, a feature fusion module and a risk assessment module. The system adopts a modular design architecture, and the functional modules exchange data and collaborate functions through standardized interfaces. This design not only improves the scalability and maintainability of the system, but also enables the system to adapt to port cargo storage scenarios of different scales and types. The modular architecture also supports dynamic upgrading and replacement of functions. For example, the algorithm parameters of the feature extraction module can be customized according to the requirements of specific cargo types, or the configuration of the data acquisition module can be adjusted according to the infrastructure conditions of different ports. The software part of the system adopts a cloud-edge collaborative architecture. Data processing tasks are dynamically allocated between edge devices and cloud servers according to real-time requirements and computing resource availability, which not only ensures the real-time response of key warning functions, but also can utilize the powerful computing resources of the cloud for more complex data analysis and model optimization.

[0086] The data acquisition module is used to obtain the temperature distribution matrix, humidity distribution matrix and surface temperature image of the cold chain cargo area. The data acquisition module includes a temperature sensor grid, a humidity sensor grid and a thermal imaging device.

[0087] The data acquisition module utilizes low-power wide area network (LPWAN) technology for sensor network networking and data transmission. Compared to traditional wired or short-range wireless solutions, LPWAN technology can cover the entire port area while maintaining low energy consumption, significantly reducing the complexity and cost of infrastructure deployment. The temperature sensor grid utilizes a high-precision semiconductor temperature sensor array with a temperature measurement accuracy of ±0.1°C, capable of detecting even small temperature changes. The humidity sensor grid utilizes capacitive humidity sensors with a measurement range of 0-100%RH and an accuracy of ±2%RH, suitable for various humidity environments. The thermal imaging device utilizes a combination of a miniature thermal imaging sensor and a wide-angle lens, capable of capturing surface temperature distribution images over a wide area without interfering with normal cargo handling. The data acquisition module also integrates self-calibration and fault detection capabilities, automatically identifying and compensating for sensor drift to ensure the accuracy and reliability of long-term monitoring data.

[0088] The feature extraction module is used to extract time series features and spatial temperature gradient features from the collected data, and the feature extraction module includes a recursive neural network and a deep residual network.

[0089] The feature extraction module runs on the system's edge computing unit and utilizes a neural network inference framework optimized for low-power devices, enabling efficient feature extraction with limited computing resources. The parameters of the recursive neural network and deep residual network have been carefully optimized to ensure the model's expressiveness while controlling computational complexity. In actual deployment, the feature extraction module uses model quantization technology to convert raw floating-point operations into integer operations, further improving computational efficiency and energy efficiency. Furthermore, the feature extraction module implements incremental learning, enabling it to continuously optimize and update the model using newly collected data without interrupting system operation. This allows the system to adapt to changes in the environment and cargo characteristics, maintaining long-term monitoring effectiveness.

[0090] The feature fusion module is used to perform adaptive weighted fusion on the extracted features to generate a comprehensive feature representation. The feature fusion module adopts an adaptive weighting mechanism based on uncertainty.

[0091] The risk assessment module is used to derive the abnormal risk level of the cargo status in the current monitoring period based on the fused feature sequence using a gradient boosting tree model, and identify and provide the location of the hazard source based on the abnormal risk level.

[0092] The system further comprises an early warning module for triggering an early warning mechanism when a moderate anomaly or a severe anomaly is detected.

[0093] The early warning module adopts a multi-level, multi-channel early warning strategy. The system automatically selects the appropriate early warning method based on the risk level of the detected abnormality. For minor abnormalities, the system will generate a low-priority notification and prompt the management personnel through the message center of the monitoring platform. For moderate abnormalities, the system will simultaneously send SMS, email, and mobile application push notifications to the relevant responsible persons, and initiate more frequent data collection and analysis. For serious abnormalities, the system will trigger an audible and visual alarm, and notify the emergency management department and key decision makers through the preset emergency communication channels, and automatically generate preliminary emergency response recommendations. The early warning module also integrates early warning confirmation and response tracking functions, which can record the receipt, confirmation, and processing of early warning information, forming a complete early warning response closed loop, providing a basis for subsequent system optimization and event backtracking. In addition, the early warning module supports personalized early warning rule configuration. Management personnel can adjust the trigger conditions and notification methods of the early warning according to specific scenarios and needs, improving the flexibility and practicality of the system.

[0094] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time monitoring method for hazardous sources of port cargo based on the Internet of Things is applied to the quality and safety monitoring of cold chain cargo storage areas in ports. It is characterized by: The method comprises the following steps: Step S1: Configuring the sampling intervals of the temperature sensor grid, the humidity sensor grid, and the thermal imaging device to be consistent, using the temperature sensor grid and the humidity sensor grid to collect the temperature distribution matrix and the humidity distribution matrix of the cold chain cargo area, and using the thermal imaging device to obtain the cargo surface temperature image; Step S2: Using a recursive neural network to extract the temporal variation characteristics of temperature and humidity, a deep residual network to extract the spatial temperature gradient characteristics of the thermal image, and a variational autoencoder to perform feature dimensionality reduction; the obtained multimodal features are fused through a dynamic weight mechanism to construct a comprehensive feature vector; Step S3: The fused comprehensive features are constructed into a time window sequence. Based on the pre-trained gradient boosting tree model, the abnormal risk level of the cargo status in the current monitoring period is evaluated. Based on the abnormal risk level, the location of the hazard source is identified and given. The model outputs anomaly risk probability distribution, which is divided into four levels: normal, slight anomaly, moderate anomaly, and severe anomaly. When the anomaly risk level reaches moderate anomaly or above for three consecutive time windows, the system will use cluster analysis to determine the spatial distribution characteristics of temperature and humidity anomalies and, combined with the cargo storage layout, accurately locate the potential hazard source. The length of the time window sequence is K, the value range of K is 8 to 20, and the overlap rate of the sliding window is 50%; The classification of the abnormal cargo status risk level includes: when the temperature deviation does not exceed ±1.0°C of the set threshold and the humidity deviation does not exceed ±5%RH, and the temperature difference of the thermal imaging image does not exceed 3.0°C, it is considered normal; A minor abnormality occurs when the temperature deviation is between ±1.0°C and ±2.0°C of the set threshold, or the humidity deviation is between ±5%RH and ±10%RH, or the detected temperature and humidity change rate in a local area exceeds 0.5°C / hour or 2%RH / hour, and the temperature difference in the thermal image does not exceed 3.0°C. Moderate abnormality occurs when the temperature deviation is between ±2.0°C and ±4.0°C of the set threshold, or the humidity deviation is between ±10%RH and ±15%RH, or the detected temperature and humidity change rate in a local area exceeds 1.0°C / hour or 4%RH / hour, or the temperature difference in the thermal image is between 3.0°C and 6.0°C. A serious anomaly is identified when the temperature deviation exceeds the set threshold of ±4.0°C or the humidity deviation exceeds ±15%RH, or the detected temperature and humidity change rate in a local area exceeds 2.0°C / hour or 8%RH / hour, or the temperature and humidity anomaly indicators show an accelerated deterioration trend within three consecutive time windows, or the temperature difference in the thermal image exceeds 6.0°C. Hazard Source The positioning adopts the density clustering algorithm, and its mathematical expression is: , where D is the set of sensor data points, for point and point The Euclidean distance function, is the neighborhood radius.

2. The method for real-time monitoring of dangerous sources of port cargo based on the Internet of Things according to claim 1 is characterized in that: In step S2, the recurrent neural network adopts a gated recurrent unit GRU structure, which includes: 2 bidirectional GRU layers, each followed by a layer normalization and Dropout layer, and a fully connected output layer; the input is The temperature and humidity distribution matrix sequence at each time point is output as a 128-dimensional feature vector; The mathematical expression of the GRU unit is: ； ； ； ； in, represents the update gate, Represents the reset gate, represents the candidate hidden state, Indicates the hidden state at the current moment, Represents the input sequence at time The value of 、 、 are the weight matrices of the corresponding gates, 、 、 is the corresponding bias term, is the sigmoid activation function, Represents the Hadamard product.

3. The method for real-time monitoring of dangerous sources of port cargo based on the Internet of Things according to claim 2 is characterized in that: In step S2, the deep residual network adopts the DenseNet-121 structure, which includes: 1 7x7 convolutional layer, 4 dense blocks, each with a transition layer between dense blocks, 1 global average pooling layer and 1 fully connected layer; the input is the thermal imaging temperature map, and the output is a 256-dimensional feature vector; The mathematical expression of the dense block is: ,in Represents a composite transformation function, including batch normalization, ReLU activation and convolution operations, Indicates the The output of the layer, Represents the concatenation of the outputs of all previous layers.

4. The method for real-time monitoring of dangerous sources of port cargo based on the Internet of Things according to claim 3 is characterized in that: In step S2, feature fusion adopts an adaptive weighting mechanism, and the feature fusion method includes: calculating the uncertainty weight of each modal feature, integrating multimodal features using an attention gating network, and generating a context-related fusion representation; The mathematical expression of the adaptive weighting mechanism is: ； The calculation expression of the uncertainty weight is: ;in, Indicates the Class features, Corresponding to the characteristics of different modes; represents the adaptive weight based on uncertainty; Representation characteristics The variance of , reflects the uncertainty of the feature; represents the attention gating coefficient, represents the global context vector; represents the weight matrix of the gating network, represents the bias vector; Represents the final fused feature representation.

5. The method for real-time monitoring of dangerous sources of port cargo based on the Internet of Things according to claim 4 is characterized in that: In step S3, the gradient boosting tree model adopts the LightGBM framework, including 100 decision trees, each tree has a maximum depth of 6, a minimum number of leaf node samples of 20, and a learning rate of 0.05; the prediction function of the decision tree ensemble can be expressed as: ; ;in, Representation sample The predicted value of Indicates the A decision tree, represents the input feature vector, represents the loss function, represents the training error term, represents the regularization term, which is used to control the complexity of the model. Representation sample The true label.

6. The method for real-time monitoring of dangerous sources of port cargo based on the Internet of Things according to claim 5 is characterized in that: In step S1, the data acquisition frequency of the temperature sensor grid, the humidity sensor grid, and the thermal imaging device is set to 0.5 Hz; the spatial resolution of the temperature sensor grid is 12x12 dot matrix, the spatial resolution of the humidity sensor grid is 10x10 dot matrix, and the resolution of the temperature image collected by the thermal imaging device is 384x288 pixels.

7. The method for real-time monitoring of dangerous sources of port cargo based on the Internet of Things according to claim 6 is characterized in that: In step S2, the input of the recursive neural network is the temperature and humidity data of m consecutive time points, the value range of m is 15 to 45, and the latent space dimension of the variational autoencoder is 64.

8. A real-time monitoring system for port cargo hazards based on the Internet of Things, used to implement the real-time monitoring method for port cargo hazards according to any one of claims 1 to 7, characterized in that: The system includes: a data acquisition module, a feature extraction module, a feature fusion module and a risk assessment module; The data acquisition module is used to obtain the temperature distribution matrix, humidity distribution matrix and surface temperature image of the cold chain cargo area. The data acquisition module includes a temperature sensor grid, a humidity sensor grid and a thermal imaging device; The feature extraction module is used to extract time series features and spatial temperature gradient features from the collected data, and the feature extraction module includes a recursive neural network and a deep residual network; The feature fusion module is used to perform adaptive weighted fusion on the extracted features to generate a comprehensive feature representation, and the feature fusion module adopts an adaptive weighting mechanism based on uncertainty; The risk assessment module is used to derive the abnormal risk level of the cargo status in the current monitoring period based on the fused feature sequence using a gradient boosting tree model, and identify and provide the location of the hazard source based on the abnormal risk level.

9. The real-time monitoring system for port cargo hazards based on the Internet of Things according to claim 8 is characterized by: The system also includes an early warning module for triggering an early warning mechanism when a moderate abnormality or a serious abnormality is detected.

Images