Cold chain product whole-process wireless traceability system based on Internet of Things

Abstract

The invention relates to the technical field of cold chain product traceability, and discloses a cold chain product whole-process wireless traceability system based on the Internet of Things, and the system comprises an acquisition module which is used for collecting environment data of a cold chain product; the state matrix construction module is used for constructing an observation state matrix; the feature extraction module is used for extracting observation state matrix features and constructing transportation abnormity and risk grade indexes; the risk evaluation module is used for evaluating whether environment abnormity and state mutation exist or not at present; the data uploading control module is used for sending alarm information to the cloud; a state graph modeling and visualization module; and constructing a state graph based on the alarm information. State estimation is carried out through the state estimation unit, reconstruction compensation can still be carried out on key data when the nodes are in the dormant state, the integrity and continuity of a data chain are guaranteed, a data blind area is effectively filled, the integrity of data detection is improved, and the accuracy of cold chain product traceability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cold chain product traceability, and in particular to a full-process wireless traceability system for cold chain products based on the Internet of Things. Background Art

[0002] With the increasing circulation of environmentally sensitive goods such as fresh food, pharmaceuticals, and biological products throughout the supply chain, cold chain logistics systems are placing higher demands on the monitoring and management of environmental parameters during transportation. Cold chain traceability technology has emerged as a result. By deploying environmental sensing devices, it aims to collect and record key parameters such as temperature, humidity, location, and time during cold chain transportation in real time, establishing a full-process tracking information chain from source to destination. If product quality anomalies occur, historical environmental data can be used to quickly trace the source, identify the problematic link and responsible party, thereby effectively ensuring product safety, improving supply chain transparency, and meeting regulatory compliance requirements.

[0003] Existing cold chain traceability systems generally rely on sensor nodes to sample and upload environmental data in real time. Once a node enters a dormant state, the data chain is interrupted. These systems often assume that missing data is untraceable and lack effective mechanisms for addressing data discontinuities. Furthermore, while some systems have introduced data caching or delayed upload mechanisms, they still rely on nodes waking up to replenish data, failing to achieve continuous monitoring during dormant periods and, in particular, failing to capture sudden changes in risk. Furthermore, frequent dormancy of sensor nodes is common during cold chain transportation due to energy management requirements. Most existing technologies focus solely on triggering data uploads, but fail to address the state uncertainty caused by observation gaps, resulting in unpredictable information blind spots during transportation. Furthermore, current mainstream solutions lack intelligent compensation mechanisms for missing data. Instead, they often resort to "discarding" or "inserting zeros" to address data gaps, which is not only imprecise but also undermines the effectiveness of subsequent risk assessment models. Especially at critical moments (such as temperature change sections during long-distance transport), these simplistic approaches often mask the true risk of sudden state changes. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a full-process wireless traceability system for cold chain products based on the Internet of Things, which solves the problem that the existing cold chain traceability system cannot compensate for missing data during node dormancy, resulting in monitoring interruption and unknown status, thereby affecting data integrity and traceability accuracy.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a full-process wireless traceability system for cold chain products based on the Internet of Things, comprising:

[0006] A collection module is used to collect environmental data of cold chain products during cold chain transportation, wherein the data includes at least temperature, humidity, location information, vibration intensity and timestamp;

[0007] The state matrix construction module builds an observation state matrix based on environmental data to describe the dynamic changes of cold chain products during cold chain transportation;

[0008] The feature extraction module extracts features based on the observed state matrix to describe the key feature information of the transportation state change trend and construct transportation anomaly and risk level indicators;

[0009] The risk assessment module evaluates whether there are environmental anomalies and sudden changes in status based on historical trends in transportation anomalies, risk level indicators, and environmental data;

[0010] The data upload control module determines the safety risk of cold chain products based on the assessment results. When the safety risk exceeds the set threshold, an alarm message is sent to the cloud;

[0011] State graph modeling and visualization module; builds state graph based on alarm information.

[0012] Preferably, the acquisition module includes a plurality of wireless sensing nodes, which are deployed in transport containers to collect environmental data of cold chain products, wherein the environmental data includes temperature, humidity, geographic location and vibration intensity.

[0013] Preferably, the state matrix construction module establishes a data matrix of multiple nodes and multiple moments based on the environmental data collected by each sensing node, which is used to characterize the environmental change state during the cold chain transportation process.

[0014] Preferably, the state matrix building module further includes a sparse matrix to adapt to the non-continuous observation data of the node sleep and provide a basis for data compensation.

[0015] Preferably, the feature extraction module uses the principal component analysis algorithm to perform dimensionality reduction analysis on the state matrix to extract the changing characteristics of the environmental data during transportation; the risk assessment module uses the random forest classification algorithm to model and evaluate the extracted feature information and historical environmental data to determine whether there are currently environmental anomalies and state mutations, and output the corresponding risk level indicators.

[0016] Preferably, the data upload control module determines whether the current environment is abnormal based on the risk level index and the change trend of historical environmental data, and decides whether to trigger the node to upload data.

[0017] Preferably, the system also includes a node scheduling and energy consumption management module, which is used to build a multi-factor evaluation model based on the current risk level and data freshness, and control the working state selection of the wireless sensing node, and the working state includes sleep, sampling and activation upload.

[0018] Preferably, the data upload control module further includes a state estimation unit for performing data compensation and data reconstruction based on the collected environmental data and historical environmental data when the wireless sensing node is in a dormant state to make up for the lack of environmental data.

[0019] Preferably, the data compensation and data reconstruction are based on time series regression interpolation and bidirectional long short-term memory network algorithms to predict and reconstruct missing environmental data.

[0020] Preferably, the state graph modeling and visualization module further includes: constructing a state graph based on alarm information and compensation and data reconstruction data, the graph is composed of a node set and an edge set, the nodes represent time points, geographic locations and product batches, the edges represent transportation paths and data transmission relationships, and the cold chain transportation status is displayed through a graphical interface.

[0021] The present invention provides a full-process wireless traceability system for cold chain products based on the Internet of Things. It has the following beneficial effects:

[0022] 1. This invention uses a state estimation unit to perform state estimation, allowing reconstruction and compensation of key data even when a node is dormant, thus ensuring the integrity and continuity of the data chain. Unlike the passive processing of data loss or breakpoints in existing technologies, this invention effectively fills data blind spots, improves the integrity of data detection, and enhances the accuracy of cold chain product traceability.

[0023] 2. This invention significantly enhances the system's understanding of cold chain dynamics by employing a sparse matrix structure to process discontinuously sampled data, combining principal component analysis and random forest models to extract key features and assess risk. Compared to traditional simple statistical or single-variable monitoring mechanisms, it better captures environmental trends and potential anomalies, resolving the current bottleneck of excessive data redundancy and insufficient effective information.

[0024] 3. This invention dynamically schedules the operating status of wireless sensing nodes by building a multi-factor decision model based on risk level and data freshness, effectively improving node responsiveness and overall energy consumption control during critical periods. Compared with traditional strategies that use fixed sampling periods, this approach avoids unnecessary energy consumption caused by high-frequency operation and addresses the issue of insufficient monitoring coverage under energy budget constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Schematic diagram of the system architecture of the present invention. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the specification of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0027] In order to better understand the present invention, the above contents are described in detail below in conjunction with specific embodiments.

[0028] Please see the attached Figure 1 The embodiment of the present invention provides a full-process wireless traceability system for cold chain products based on the Internet of Things, including:

[0029] A collection module is used to collect environmental data of cold chain products during cold chain transportation, wherein the data includes at least temperature, humidity, location information, vibration intensity and timestamp;

[0030] In this embodiment, in order to achieve all-round and dynamic perception of the environmental status of the product during the entire cold chain product transportation process, the present invention provides a collection module with multi-node wireless data perception capabilities. This module serves as the data source entrance of the system, and its collection accuracy and data dimension directly affect the accuracy and response efficiency of subsequent state modeling, risk assessment, and graph visualization modules.

[0031] Cold chain transport typically responds to changes in environmental conditions with nonlinear, discontinuous, and regionally heterogeneous characteristics. Therefore, traditional single-point, single-parameter sensing methods are no longer sufficient to meet the current demand for fine-grained monitoring of transport conditions. To enhance the cold chain product traceability system's ability to respond to multidimensional environmental changes, this paper introduces a network of multiple sensing nodes with wireless communication capabilities at the environmental data acquisition layer. Combined with a multi-parameter integrated sampling design, this approach enables the stable collection and simultaneous calibration of key environmental factors.

[0032] In this embodiment, the acquisition module includes multiple wireless sensing nodes, each of which integrates a temperature sensor, a humidity sensor, a vibration sensor, and a positioning module. The positioning module uses a Global Positioning System (GPS) chip or a Beidou navigation terminal to continuously collect geographic location information. The sampling frequency can be flexibly adjusted between 1 and 30 minutes based on the data scheduling strategy.

[0033] Specifically, sensing nodes are deployed within cold chain transport containers, forming a spatial coverage network within the refrigerated truck compartment based on the stacking pattern and heat conduction characteristics of cold chain products. In some embodiments, node deployment is based on simulations based on thermal distribution, prioritizing coverage of high-fluctuation areas such as doorways, air duct exits, and areas far from the cold source.

[0034] As an option, the temperature sensor uses a high-precision DS18B20 single-bus temperature acquisition chip with digital output function, with a measurement accuracy of ±0.5℃ and a measurement range of -55℃ to +125℃, suitable for both refrigeration and freezing scenarios. The humidity sensor uses an integrated SHT3x series device, and the output mode supports I 2 C protocol, response time is less than 2s, and long-term drift error is less than ±1.5%RH.

[0035] The vibration sensor uses a MEMS triaxial accelerometer (such as the ADXL345) to capture strong vibrations or sustained oscillations caused by bumps during transportation or improper handling. Vibration data is recorded as a triaxial acceleration vector. The sampling window is set to 5 seconds, and the recorded values ​​within the window are recorded as a vector:

[0036] a t =[a x (t),a y (t),a z (t)];

[0037] And calculate the modulus value by:

[0038]

[0039] Among them, a x (t) is the acceleration vector of the cold chain product displacement along the X-axis at time t;

[0040] a y (t) is the acceleration vector of the cold chain product displacement along the Y axis at time t;

[0041] a z (t) is the acceleration vector of the cold chain product displacement along the Z axis at time t;

[0042] a t is the displacement vector of the cold chain product at time t;

[0043] In one possible implementation, all sensing nodes are equipped with Bluetooth Low Energy (BLE) or LoRa wireless communication capabilities, capable of transmitting collected environmental data at preset intervals to an edge gateway device or to a central server via a relay network. A dynamic topology can be constructed between nodes to achieve communication redundancy and fault-tolerant scheduling when some nodes are dormant.

[0044] Each collected record is accompanied by an accurate timestamp. The timestamp is uniformly formatted in UTC by the system and supports millisecond-level accuracy to meet the needs of high-frequency data fusion.

[0045] In actual use, the node sampling period, upload frequency, and power consumption status can be dynamically adjusted based on the strategies established by the data upload control module and the node scheduling module. The sensing node supports OTA remote parameter configuration and has a caching mechanism that supports local temporary storage during data interruptions. The cache capacity is no less than 200 complete data records.

[0046] Furthermore, to address data blind spots caused by node failure or obstructed areas, in some embodiments, the acquisition module further supports a collaborative compensation mechanism, whereby surrounding nodes approximate the data in the failed area through weighted averaging or covariance transfer. This mechanism provides support for state matrix construction and data compensation, enhancing overall system robustness.

[0047] In summary, the acquisition module described in the present invention not only realizes high-frequency and stable acquisition of key environmental factors such as temperature, humidity, location, vibration intensity and timestamp, but also improves the continuity and integrity of cold chain transportation environmental data through multi-node redundant configuration and wireless collaborative communication mechanism, providing an accurate and reliable data foundation for subsequent status modeling, risk analysis and visualization.

[0048] The state matrix construction module builds an observation state matrix based on environmental data to describe the dynamic changes of cold chain products during cold chain transportation;

[0049] In this embodiment, in order to realize continuous modeling and time-series analysis of environmental state changes during cold chain transportation, the present invention introduces a state matrix construction module on the basis of the acquisition module to systematically organize and express multidimensional environmental data collected from multiple nodes, thereby providing a structured input basis for subsequent feature extraction, risk assessment, and state graph modeling functions.

[0050] Typically, sensor data in cold chain transportation environments exhibits strong time series characteristics and spatial heterogeneity. Differences in geographic location, sampling frequency, and communication status between nodes often result in non-homogeneous and non-aligned raw data. To overcome this problem, this paper proposes an observation state matrix architecture based on the node-time dual dimension, which effectively describes the dynamic evolution of environmental states in cold chain systems.

[0051] In this embodiment, the state matrix construction module obtains raw data records from the acquisition module. Each data record contains key fields such as node ID, timestamp, temperature, humidity, vibration intensity, and location information. By structuring the observation results of each sensing node at different time points, a two-dimensional observation state matrix M is constructed. Its form is as follows:

[0052]

[0053] Where N represents the total number of wireless sensing nodes; T represents the total length of time sampling;

[0054] And the observation value is a multidimensional vector, including the following:

[0055] x ij =[Temp ij ,Hum ij ,|a ij |,Lat ij ,Lon ij ];

[0056] Among them, x ij Represents the observation value of the i-th node at the j-th time point, which is a multidimensional vector;

[0057] Specifically, each data item of the node time series is accompanied by an observation identification bit. If there is a valid value in the multidimensional vector, the missing position in the state matrix is ​​left blank to reduce storage space and provide a basic mark for subsequent data compensation and reconstruction.

[0058] As an option, the state matrix construction module further integrates time normalization and node alignment mechanisms. Time normalization discretizes the original timestamps using a uniform sampling interval Δt, ensuring that all observation nodes are comparable at the same time scale. Node alignment is performed based on the node's unique identifier and the deployment mapping table during system initialization, ensuring that the data of each node has consistent meaning along the matrix row dimension.

[0059] In some embodiments, to enhance the ability to model correlations between multi-node data, the state matrix also supports an additional metadata matrix to record static characteristics of each node, such as the node's initial deployment position, height, and relative distance to the cold source within the container. This metadata is used as an auxiliary parameter in the feature extraction module to participate in anomaly detection and risk scoring.

[0060] In practical applications, this state matrix building module, as part of the edge data preprocessing module, can be deployed in gateway devices or edge servers, supporting a real-time update mechanism. As new data is continuously collected, the matrix can dynamically expand the time dimension and sparsely fill in the newly added data items.

[0061] Furthermore, in some extended implementations, the state matrix construction module also includes pre-configured data consistency verification logic. For example, this logic can determine if sensor drift, sudden errors, or other issues exist by comparing the rate of change of data at the same node within adjacent time windows to see if it exceeds a system-defined threshold. Cells with data anomalies are marked with a quality indicator and fed back to the data upload control module for further processing and decision-making.

[0062] In summary, the state matrix construction module described in the present invention realizes the cold chain environment observation data organization structure based on multiple nodes, multiple time periods, and multiple parameters, supports the effective modeling of sparse observation data, and can provide a unified, structured, and scalable input basis for subsequent modules, with strong data adaptability and dynamic adjustment capabilities.

[0063] The feature extraction module extracts features based on the observed state matrix to characterize key feature information of the transportation state change trend and construct transportation anomaly and risk level indicators. The risk assessment module assesses whether there are environmental anomalies and state mutations based on the historical change trends of transportation anomaly and risk level indicators and environmental data.

[0064] In this embodiment, in order to effectively capture and characterize the evolution trend of environmental states during cold chain transportation, after completing the structured organization of the original environmental data, the present invention further introduces a feature extraction module to extract key change patterns from the constructed observation state matrix, thereby providing efficient and identifiable feature representation for subsequent risk identification and assessment.

[0065] Typically, the original observation state matrix contains multi-node, multi-dimensional, high-frequency environmental data. Its dimensions are highly redundant and complexly correlated. Directly using this matrix for state judgment will lead to the curse of dimensionality, which is detrimental to model generalization and real-time performance. Therefore, this paper proposes a feature extraction process based on a dimensionality reduction algorithm. By performing principal component analysis (PCA) on the observation matrix, a low-dimensional feature space is constructed to characterize transportation state changes, extracting the key information components with discriminative capabilities.

[0066] In this embodiment, before feature extraction, the two-dimensional observation state matrix M needs to be flattened into a sample dimension matrix X as input to the PCA algorithm.

[0067] Specifically, the steps of principal component analysis include the following stages:

[0068] Centralization: Perform zero-mean normalization on each column in X to obtain the centralized matrix X c ;

[0069] Covariance matrix calculation: construct the covariance matrix between variables;

[0070] Eigendecomposition: Perform eigenvalue decomposition on the covariance matrix, calculate the eigenvalues ​​and corresponding eigenvectors, and arrange them in descending order of eigenvalue size;

[0071] Principal component selection: Select the first k principal component vectors whose cumulative contribution rate reaches the set threshold (such as 90%) and construct the dimension reduction projection matrix W k =[V1,V2,...,V k ];

[0072] Feature transformation: Calculate the dimension-reduced feature matrix F = X c W k , which is the key feature set used to characterize the changing trend of transportation status.

[0073] In one possible implementation, the feature dimension can be dynamically set, specifically based on the spectral distribution of the covariance matrix of the input data, to prevent excessive information loss or the retention of redundant features. Furthermore, to enhance the model's ability to perceive spatial differences, some embodiments introduce node position embedding vectors, which are mapped to node IDs and geographic locations and then concatenated into the feature vector.

[0074] The extracted feature set F will serve as the input for the risk assessment module, which is used to further construct a recognition model for environmental anomalies and state mutations. The present invention preferably uses a random forest classifier as the evaluation basis, which has the advantages of strong nonlinear discrimination ability, high noise resistance, and adaptability to multi-feature scenarios.

[0075] Specifically, the risk assessment model uses the extracted feature F as input and the historically labeled environmental status (whether it is abnormal, risk level) as a supervisory signal to construct a binary or multi-classification model. Assume that the random forest model contains n decision trees and the overall output is the risk level scoring function R(F). Its output is:

[0076]

[0077] Among them, h i (F) represents the prediction result of the i-th decision tree for the input feature F (such as output risk level 0 / 1 / 2, etc.).

[0078] In certain embodiments, the random forest model training phase utilizes cross-validation and a feature importance scoring mechanism to identify the optimal feature subset and eliminate redundant features to improve model generalization. Key features such as temperature change rate, humidity fluctuation frequency, and vibration periodicity intensity have high importance scores in the evaluation model and are highly correlated with transportation risk.

[0079] To further improve the interpretability of the system, the module can also output the main characteristic factors and their contribution rates involved in the judgment in each anomaly assessment, assisting the regulatory side in tracing decisions and risk positioning.

[0080] In an actual deployment example, the feature extraction and evaluation module is deployed in the edge computing unit, supporting online updates and rapid responses. The typical processing cycle does not exceed 5 seconds, which is suitable for high-frequency transportation status monitoring needs.

[0081] At the same time, as an option, the risk assessment module also introduces a dynamic baseline mechanism. For some transport nodes that have been operating stably for a long time, their normal environment range will gradually learn during operation to form a node-individualized assessment benchmark. This mechanism can form a probability model by statistically analyzing the distribution interval of historical data for each node, which is used to compare whether the current features deviate significantly from the expected area, and to assist in generating anomaly scores accordingly. In some embodiments, the output of the risk assessment module includes not only the final risk level label, but also an evaluation probability vector, which represents the normalized probability estimate of the corresponding level. This output form helps downstream modules to perform soft decisions and adaptive threshold control.

[0082] To enhance model transparency and auditability, some implementations incorporate a feature importance analysis module. This module dynamically outputs the most critical input features for the current decision, based on metrics such as the Gini coefficient or information gain. For example, in certain refrigerated pharmaceutical transport tasks, the weight of the temperature fluctuation frequency feature is significantly higher than the vibration parameter, indicating that this transport task is more sensitive to temperature control stability.

[0083] Furthermore, during model execution, the system supports a rule fusion mechanism, which weights and fuses the machine learning results with those of the preset hard threshold. If a conflict arises, manual review or administrator intervention is prioritized.

[0084] Therefore, in summary, the feature extraction module of the present invention performs dimensionality reduction processing on multi-node and multi-dimensional environmental data through principal component analysis to extract key change trend features. The risk assessment module combines the random forest assessment method to construct a transportation risk level index, forming an efficient recognition capability of abnormal conditions in the cold chain transportation process, and providing reliable criteria for subsequent upload control, alarm response and status map modeling.

[0085] The data upload control module determines the safety risk of cold chain products based on the assessment results. When the safety risk exceeds the set threshold, an alarm message is sent to the cloud;

[0086] In this embodiment, after completing the output of the risk level indicator, to achieve dynamic control and resource optimization of data transmission in the cold chain transportation system, the present invention further provides a data upload control module. Based on the output of the risk assessment module, this module combines environmental data trends and node operating status to determine whether there is a security risk and decide whether to trigger the node to upload data or enter an alarm state. This module not only improves system energy efficiency and bandwidth utilization, but also ensures information accessibility and data integrity at critical risk moments.

[0087] Typically, sensing nodes in cold chain transportation environments employ low-power designs, operating in intermittent sleep and wakeup states. Data transmission is constrained by energy budgets and network stability. Without an effective risk-driven mechanism, redundant data will be uploaded in low-risk scenarios. In high-risk scenarios, however, dormant nodes prevent timely reporting of critical data, impacting system reliability.

[0088] In this embodiment, the data upload control module receives the overall output risk level scoring function R(F) output by the aforementioned risk assessment module, combines it with the historical evolution trajectory E(t) of the environmental data, and constructs a state judgment function U(t) to determine in real time whether to upload data. The upload control logic is expressed as follows:

[0089]

[0090] Among them, θ r The threshold for triggering upload at risk level; θ e The threshold value of the rate of change controls the sensitivity of the system to abnormal mutations. Indicates the rate of change of current environmental data, which can be approximated by the maximum value of the derivative of each dimension indicator;

[0091] That is, U(t)=1 indicates that uploading is triggered, and U(t)=0 indicates that uploading is not required.

[0092] Specifically, when the risk level and environmental disturbance meet the set conditions at the same time, the control module generates an upload trigger signal and pushes the key data packets to the cloud server in real time through the edge node for the supervision system to respond quickly.

[0093] Optionally, the data upload control module further includes a state estimation unit. This unit is used to dynamically compensate for missing segments based on the collected data sequence and the historical environmental change model when the sensing node is in an inactive state (such as sleep or signal interruption), thereby preventing the overall evaluation process from being affected by data interruptions.

[0094] In one possible implementation, the state estimation unit is reconstructed based on a combination of a time series regression interpolation algorithm and a bidirectional long short-term memory network (BiLSTM). Suppose the temperature data sequence of a node is:

[0095] T={T1,T2,…,T t-1 ,□,T t+1};

[0096] Where □ indicates that the data at time t is missing, the estimation module relies on the previous and next contexts and estimates the missing value through the regression model

[0097] The estimation process can be expressed as:

[0098]

[0099] Among them, f BiLSTM (.) represents the bidirectional LSTM network function after training;

[0100] The input sequence includes the complete context before and after the missing point;

[0101] The output is the estimated value of the reconstructed point.

[0102] In actual deployment, to improve estimation accuracy, a weighted loss function is introduced during model training to give higher weight to reconstruction errors close to outliers. The form is as follows:

[0103]

[0104] Where L is the weighted mean square error of the prediction model; w t is the weight coefficient, which can be adjusted according to the historical risk distribution. For example, the weight of the high-risk segment is set to more than twice that of the normal segment to improve the prediction accuracy at critical moments; t represents the actual environmental data value at time point t, such as the actual collected parameters such as temperature, humidity or vibration intensity; represents the environmental data value predicted or estimated by the model at time point t, that is, the value of T t The estimated value of , which comes from the BiLSTM network as described above.

[0105] In some embodiments, in order to enhance the real-time compensation capability of the module, the state estimation unit will preload a trend prediction model before the node goes into sleep mode, and combine it with the short-term prediction strategy to estimate the environmental changes in the next few cycles to form a "preparatory upload package" to be quickly completed and reported when the risk is activated.

[0106] Furthermore, the upload control module supports an asynchronous retransmission mechanism. If a node is determined to be in a signal blind spot during upload, the module caches the data locally and automatically retransmits it once the network is restored, ensuring data integrity and timeliness. A priority queue mechanism is incorporated into cache management, prioritizing the removal of data from high-risk time periods.

[0107] In a typical scenario, this module is deployed in conjunction with the edge computing framework. Even with limited hardware resources on a single node, it can still maintain a decision response time of less than 1s, thus realizing an efficient, flexible, and controllable upload scheduling system.

[0108] To sum up, the data upload control module described in the present invention uses risk level and environmental dynamics as triggering basis, introduces state estimation and data compensation mechanism, and ensures the complete accessibility of key data while taking into account transmission efficiency and risk identification capabilities. It is suitable for cold chain logistics scenarios with high timeliness and high stability requirements.

[0109] State graph modeling and visualization module; builds state graph based on alarm information.

[0110] In this embodiment, after completing the risk level index calculation of the risk assessment module and the data upload control module's management of the transmission behavior of key data, in order to further realize the global modeling and in-depth visualization analysis of the key states in the cold chain transportation process, the present invention proposes a state map modeling and visualization module. This module takes the alarm information uploaded to the cloud as the input basis, and further combines the environmental perception data that has been compensated and reconstructed by data to construct a transportation state map structure that reflects the three-dimensional fusion of time, space and product batches. This state map is not only used for the intuitive display of the cold chain logistics status, but also serves as an important data basis for subsequent scheduling optimization, abnormal evolution tracking and batch risk tracing.

[0111] In general, cold chain transportation scenarios involve multiple transportation nodes, multiple routes, data records for different time periods, and multiple product batches. Due to the complex dynamics of the environment, intermittent loss of sensor data, and potential path switching and node interruptions during transportation, traditional linear time series or tabular data structures are difficult to effectively describe the evolution trend of transportation status in the time and space dimensions, and it is also difficult to express the potential correlation relationship between multiple points as a whole. To this end, the present invention constructs a state map of cold chain transportation through a graph structure to better adapt to the multi-source heterogeneous information integration needs in actual scenarios, and supports further data analysis and information display at the graph level.

[0112] In this embodiment, the state map modeling and visualization module extracts the corresponding timestamp, geographic location information and product batch identification based on the historically uploaded alarm information as the core fields that constitute the map nodes. At the same time, the module further calls the complete environmental data sequence generated by the state estimation unit. In the case of missing original data, the reconstructed estimated value is used as the state description feature of each node in the map. Each map node represents an identifiable transport state unit in the cold chain transportation system, and its core attributes include at least: the specific time of the node occurrence, the geographical location (such as GPS coordinates), the batch number of the product, and the temperature, humidity, vibration amplitude, risk level score and other completed environmental characteristics at that time point.

[0113] Specifically, when constructing the graph, the module also identifies the associations between nodes in different states based on constraints such as spatiotemporal proximity, transport path continuity, and product batch consistency. If two nodes are within an acceptable temporal distance, have geographical accessibility, and belong to the same logistics batch, a graph edge relationship can be established between them, representing an actual transport path or a logical data transmission path. In this graph structure, the information carried by the edge includes not only the path start and end node identifiers, but can also be expanded to include attributes such as the transport path type, risk information transmission direction, and state evolution trend corresponding to the edge, thereby enhancing the dynamic expression capabilities of the graph.

[0114] Alternatively, the state graph modeling module can be expanded into a multi-layered graph system. In the upper-level graph, the system aggregates multiple state nodes into transport task-level aggregation nodes, displaying the overall state trend of a batch of products. In the lower-level graph, the system retains the node granularity of the original sensor level and constructs a detailed time-segment-level graph to support more refined data analysis and state backtracking. This layered graph architecture can effectively address the granularity requirements of different user scenarios. It can be used for macro-situation display in the regulatory system and for analysis and tracking of local state evolution paths in the risk diagnosis module.

[0115] In order to improve the usability and interactivity of the system, the state graph modeling module also integrates a front-end graphical interface display mechanism. Users can intuitively view the attribute information of each state node in the graphical interface, including but not limited to reconstructed environmental data, original collected values ​​(if any), alarm records, GPS track points and sensor numbers, etc. The graphical presentation of the node can change dynamically according to the risk level, for example, using colors to identify different risk levels (green for normal, yellow for slight abnormalities, and red for serious alarms), and the node shape can be used to distinguish different product batches. The thickness or style of the graph edge can be used to express the severity of the state change on the path, helping users quickly identify high-risk sections that may exist during transportation.

[0116] In one possible implementation, to improve the efficiency and consistency of graph data updates, the module also designs an incremental graph structure maintenance mechanism. When new alarm information or compensation data is uploaded to the system, the graph automatically detects whether the newly added status node already exists in the graph. If so, the node information is merged. If not, a new node is added and new edges are established based on relationship rules. This mechanism effectively avoids the generation of redundant nodes and ensures the maintainability of the graph structure in the long run.

[0117] In addition, in some embodiments, the graph modeling module also supports comparative analysis with historical state graphs to identify common high-risk path patterns during transportation. By introducing graph structure encoding methods, such as those based on graph neural networks (such as GCN) or random walk embedding (such as Node2Vec), the module can encode the structural relationship between nodes into vector form and perform similarity comparisons on graphs of multiple batches. This method can be used to discover potential abnormal infection paths or structural weaknesses, and assist managers in risk avoidance and structural optimization of future transportation routes.

[0118] In summary, the state graph modeling and visualization module described in this invention fully integrates alarm information with reconstructed environmental data, combining spatiotemporal information from multiple times, locations, and batches within the cold chain transportation system to construct a transportation state graph system with structured expression capabilities and interactive visualization capabilities. This module not only enhances the system's overall visibility and controllability of the transportation process but also provides a complete, clear, and dynamically updated data foundation for subsequent anomaly propagation analysis, situation replay, and scheduling planning. It is suitable for cold chain logistics monitoring systems requiring high visualization, traceability, and multi-dimensional decision-making.

[0119] Furthermore, in the cold chain logistics full-process monitoring system described in the present invention, after the risk assessment module calculates the environmental risk level of each time node and the data upload control module manages the key data transmission behavior, in order to further improve the resource utilization efficiency and node response intelligence during the system operation, the system also includes a node scheduling and energy consumption management module. This module combines the current environmental risk level assessment results of the perception node, the data freshness index, and the node's own status history information to construct a multi-factor decision model for the scheduling of perception node tasks, so as to dynamically control its operating mode in different time periods, thereby achieving optimized management of the perception node's energy consumption while ensuring data integrity.

[0120] In general, wireless sensing nodes deployed inside cold chain transport containers usually rely on battery power, and the deployment environment is often difficult to achieve instant energy replenishment or replacement. Therefore, how to maintain sufficient data sampling frequency and upload frequency under a limited energy budget has become a key issue in system operation design. Traditional methods mostly use fixed sampling and transmission strategies, which make it difficult to take into account the high-frequency data collection needs in high-risk periods and the energy conservation goals in low-risk stages. To solve this problem, the present invention proposes a node scheduling mechanism based on risk perception and data timeliness, and constructs a set of dynamic energy consumption management strategies to improve the overall operating time and monitoring efficiency of the system.

[0121] In this embodiment, the node scheduling and energy consumption management module mainly controls the switching of node working states based on two core evaluation factors:

[0122] On the one hand, the module receives real-time output from the risk assessment module, which corresponds to the current or short-term environmental risk level. If the risk level is high (such as when the ambient temperature fluctuates drastically or when acceleration changes frequently), the system prioritizes maintaining the node in an active "sampling + uploading" state to improve the system's data density and anomaly detection capabilities during critical periods.

[0123] The module also evaluates the "data freshness" of the data collected by the current node relative to the last valid upload, a measure of data timeliness. If the system determines that the data on the current node has changed significantly, or that the time since the last upload has exceeded a set threshold, it prioritizes triggering sampling or uploading. Conversely, if the data changes are minimal or the environment is stable, the system will control the node to enter sleep mode to reduce power consumption.

[0124] Specifically, the node can dynamically switch between the following three working states:

[0125] Sleep state: The node is in a low-power standby state, with only the wake-up monitoring mechanism retained. It is suitable for time periods when the environment is stable or the risk level is low, and is used to reduce the overall energy consumption of the system.

[0126] Sampling state: The node activates the sensor module to collect data such as temperature, humidity, and acceleration, but does not immediately upload the data. This state is suitable for situations where the data is not fresh enough but the risk level is low, and is used to save communication energy consumption.

[0127] Activate upload status: The node uploads data immediately after sampling. This is suitable for situations with high risk levels or severe data lags, ensuring that critical data can be fed back to the central system in real time.

[0128] In one possible implementation, the system also incorporates a sliding time window mechanism and a local mutation detection algorithm to help determine whether significant environmental changes have occurred within a sampling period. If multiple alarm events occur within a given time window, or if the magnitude of the change in consecutive observations exceeds a threshold, a node status upgrade is triggered.

[0129] As an option, the module also supports further adjustment of node priority and task allocation based on auxiliary factors such as historical node energy consumption levels, remaining battery estimates, and sensor health status. For example, low-battery nodes can have their sleep cycles appropriately extended or be scheduled to passively respond only to system requests to extend their service life. High-health nodes with sufficient energy can take on more active sampling and uploading tasks.

[0130] In some embodiments, node scheduling strategies are not limited to local decision-making but also support central remote control. This means that scheduling instructions are issued from the cloud, and nodes receive configuration updates via a regular heartbeat mechanism. This model is suitable for scenarios where large-scale policy adjustments are required during cold chain transportation, such as sudden weather impacts a region or a surge in cold chain density during holidays.

[0131] Furthermore, this module also achieves data linkage with the state graph modeling module. Specifically, the environmental data status and alarm levels recorded in the state graph nodes can be fed back to the node scheduling module as a basis for historical evolution, thereby achieving closed-loop optimization and gradual iterative updates of the scheduling strategy.

[0132] In summary, the node scheduling and energy management module described in this invention, by combining environmental risk levels and data freshness, constructs a multi-mode node operation mechanism adapted to the dynamic nature of cold chain transportation environments. This achieves an optimal balance between energy consumption control and data timeliness for sensing nodes, thereby improving the overall operational lifespan and monitoring coverage of the system. This module is particularly suitable for long-term cold chain transportation missions with limited sensor resources and high data reliability requirements, and is a key component for the efficient operation of cold chain monitoring systems.

[0133] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The full-process wireless traceability system for cold chain products based on the Internet of Things is characterized by: include: A collection module is used to collect environmental data of cold chain products during cold chain transportation, the data including at least temperature, humidity, location information, vibration intensity and timestamp; The state matrix construction module builds an observation state matrix based on environmental data to describe the dynamic changes of cold chain products during cold chain transportation; The feature extraction module extracts features based on the observed state matrix to describe the key feature information of the transportation state change trend and construct transportation anomaly and risk level indicators; The risk assessment module evaluates whether there are environmental anomalies and sudden changes in status based on historical trends in transportation anomalies, risk level indicators, and environmental data; The data upload control module determines the safety risk of cold chain products based on the assessment results. When the safety risk exceeds the set threshold, an alarm message is sent to the cloud; State graph modeling and visualization module; Build a status map based on alarm information; The acquisition module includes a variety of wireless sensing nodes, which are deployed in the transport container to collect environmental data of cold chain products, including temperature, humidity, geographical location and vibration intensity; The state matrix construction module establishes a data matrix of multiple nodes and multiple moments based on the environmental data collected by each sensing node, which is used to describe the environmental change state during the cold chain transportation process; The feature extraction module uses the principal component analysis algorithm to perform dimensionality reduction analysis on the state matrix to extract the change characteristics of the environmental data during the transportation process; The risk assessment module uses the random forest classification algorithm to model and evaluate the extracted feature information and historical environmental data to determine whether there are environmental anomalies and state mutations, and output the corresponding risk level indicators.

2. The full-process wireless traceability system for cold chain products based on the Internet of Things according to claim 1 is characterized in that: The state matrix building module further includes a sparse matrix to adapt to the non-continuous observation data of the node sleep and provide a basis for data compensation.

3. The full-process wireless traceability system for cold chain products based on the Internet of Things according to claim 1 is characterized in that: The data upload control module determines whether the current environment is abnormal based on the risk level index and the change trend of historical environmental data, and decides whether to trigger the node to upload data.

4. The full-process wireless traceability system for cold chain products based on the Internet of Things according to claim 1 is characterized in that: The system also includes a node scheduling and energy consumption management module, which is used to build a multi-factor evaluation model based on the current risk level and data freshness to control the working state selection of the wireless sensing node, and the working state includes sleep, sampling and activation upload.

5. The full-process wireless traceability system for cold chain products based on the Internet of Things according to claim 4 is characterized in that: The data upload control module also includes a state estimation unit for performing data compensation and data reconstruction based on the collected environmental data and historical environmental data when the wireless sensing node is in a dormant state to make up for the lack of environmental data.

6. The full-process wireless traceability system for cold chain products based on the Internet of Things according to claim 5 is characterized in that: The data compensation and data reconstruction are based on time series regression interpolation and bidirectional long short-term memory network algorithms to predict and reconstruct missing environmental data.

7. The full-process wireless traceability system for cold chain products based on the Internet of Things according to claim 6 is characterized in that: The state graph modeling and visualization module further includes: constructing a state graph based on alarm information and compensation and data reconstruction data, the graph is composed of a node set and an edge set, the nodes represent time points, geographic locations and product batches, the edges represent transportation paths and data transmission relationships, and the cold chain transportation status is displayed through a graphical interface.

Images