Big data-based intelligent cargo supervision data analysis processing method and system

Abstract

The application relates to the technical field of data processing, and discloses a smart cargo supervision data analysis processing method and system based on big data. The method comprises the following steps: constructing a cargo space-time correlation graph, performing graph neural network embedding representation processing, and generating a cargo graph embedding vector; inputting a multi-modal space-time risk propagation perception algorithm to perform diffusion equation risk modeling processing, and outputting a triple risk identification result; performing differential privacy federated learning collaborative decision processing to generate a cross-domain supervision instruction set; performing self-evolution knowledge graph updating and strategy gradient reinforcement learning optimization processing, and outputting smart supervision decision data. The application solves the technical problems of data islands, inaccurate risk identification, cross-domain collaboration difficulties and insufficient strategy optimization in traditional cargo supervision, and significantly improves the intelligent level and risk identification accuracy of cargo supervision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a smart cargo supervision data analysis processing method and system based on big data. BACKGROUND

[0002] The existing cargo supervision technology mainly adopts the traditional manual inspection and sampling verification mode, relies on the experience judgment of customs officers and on-site inspection equipment to evaluate the risk of goods. The traditional supervision method identifies potential risks through checking cargo documents, physically inspecting the appearance of the package, sampling and detecting the composition of the goods, and determines the inspection level and treatment measures according to the preset supervision rules and historical experience. Some supervision departments have begun to introduce information means, establish a cargo declaration database and a risk early warning system, and screen high-risk goods through keyword matching and simple statistical analysis methods, but overall it is still dominated by static rule matching and manual decision-making.

[0003] The existing technology has the problems of serious data island phenomenon, low risk identification accuracy, and low supervision efficiency. The data of each supervision link is not effectively integrated, and the state information of the goods at different stages such as production, transportation and storage cannot be tracked throughout the process, resulting in insufficient risk assessment basis. The traditional rule matching method cannot adapt to complex and variable risk patterns, and has limited ability to identify new types of irregularities and hidden risks, which is prone to missed reports and false reports. At the same time, the manual inspection method is low in efficiency and cannot meet the demand of fast customs clearance in the face of increasing cargo flow, and the unreasonable allocation of supervision resources leads to repeated inspection and waste of resources. SUMMARY

[0004] The present application provides a smart cargo supervision data analysis processing method and system based on big data, which solves the technical problems of data island, inaccurate risk identification, difficulty in cross-domain cooperation and insufficient strategy optimization in traditional cargo supervision, and significantly improves the intelligent level and risk identification accuracy of cargo supervision.

[0005] In a first aspect, the present application provides a smart cargo supervision data analysis processing method based on big data, which comprises:

[0006] Step S1: performing blockchain hash verification and space-time four-tuple standardization coding processing on multi-dimensional supervision sensing data to generate a blockchain verification supervision data set with a unique cargo identifier and a state parameter vector;

[0007] Step S2: constructing a cargo space-time association graph according to the blockchain verification supervision data set, performing graph neural network embedding representation processing on cargo nodes and multiple types of associated edges to generate a cargo graph embedding vector that fuses risk propagation characteristics;

[0008] Step S3: input the cargo graph embedding vector into a multi-modal spatio-temporal risk propagation perception algorithm for diffusion equation risk modeling processing, and output a triple risk identification result;

[0009] Step S4: perform differential privacy federated learning collaborative decision processing according to the triple risk identification result, and generate a cross-domain supervision instruction set;

[0010] Step S5: perform self-evolution knowledge graph updating and policy gradient reinforcement learning optimization processing on the cross-domain supervision instruction set, and output wisdom supervision decision data.

[0011] In a second aspect, the present application provides a big data-based wisdom cargo supervision data analysis processing system, which comprises:

[0012] An encoding module is configured to perform blockchain hash verification and spatio-temporal four-tuple standardization encoding processing on multi-dimensional supervision sensing data, and generate a blockchain verification supervision data set with cargo unique identification and state parameter vector;

[0013] An embedding module is configured to construct a cargo spatio-temporal association graph according to the blockchain verification supervision data set, perform graph neural network embedding representation processing on cargo nodes and multiple types of associated edges, and generate a cargo graph embedding vector fused with risk propagation characteristics;

[0014] A modeling module is configured to input the cargo graph embedding vector into a multi-modal spatio-temporal risk propagation perception algorithm for diffusion equation risk modeling processing, and output a triple risk identification result;

[0015] A decision module is configured to perform differential privacy federated learning collaborative decision processing according to the triple risk identification result, and generate a cross-domain supervision instruction set;

[0016] An output module is configured to perform self-evolution knowledge graph updating and policy gradient reinforcement learning optimization processing on the cross-domain supervision instruction set, and output wisdom supervision decision data.

[0017] In a third aspect, a big data-based wisdom cargo supervision data analysis processing device is provided, which comprises a memory and at least one processor, the memory has instructions stored therein; the at least one processor invokes the instructions in the memory, so that the big data-based wisdom cargo supervision data analysis processing device executes the above-mentioned big data-based wisdom cargo supervision data analysis processing method.

[0018] In a fourth aspect, a computer readable storage medium is provided, which has instructions stored therein, when running on a computer, causes the computer to execute the above-mentioned big data-based wisdom cargo supervision data analysis processing method.

[0019] In the technical solutions provided in the application, the construction of the cargo space-time correlation graph and the graph neural network embedded representation processing break through the limitations of traditional single-point risk assessment. By establishing multiple types of correlation between cargos, the accurate modeling of risk propagation characteristics is realized, so that the supervisory department can identify hidden risk propagation paths and potential cluster risks. The multi-modal space-time risk propagation perception algorithm processes the risk modeling through the diffusion equation, converts the abstract risk concept into a quantifiable density distribution, overcomes the subjectivity and uncertainty of traditional experience judgment, and significantly improves the scientificity and accuracy of risk identification. The differential privacy federated learning collaborative decision processing realizes cross-domain collaboration under the premise of protecting the sensitive information of each participant, solves the contradiction between information silos and privacy leakage in traditional supervision, and enables customs, ports, logistics enterprises and production enterprises to make collaborative decisions under a unified framework. The self-evolution knowledge graph updating and strategy gradient reinforcement learning optimization process establishes a continuous improvement mechanism, so that the supervision strategy can be dynamically adjusted according to the actual effect, avoiding the rigid problem of traditional static rules, and realizing the spiral improvement of supervision ability.

[0020] In the specific application field of cargo supervision, the multi-modal space-time risk propagation perception algorithm MSTRPA as the core innovation of the application, its algorithm characteristics contribute particularly to the scheme. The algorithm integrates graph structure information, sensor data, historical behavior and environmental factors through a multi-modal feature fusion mechanism, overcoming the problem of limited information from a single data source, making risk assessment more comprehensive and accurate. The space-time attention double-branch calculation process optimizes the modeling of spatial correlation and temporal dependence respectively, solving the defect of traditional methods ignoring the space-time correlation, and is particularly suitable for application scenarios in cargo supervision that need to consider geographical location and time sequence at the same time. The diffusion equation risk modeling introduces the diffusion theory in physics into risk propagation analysis, describing the propagation dynamics of risk in the cargo network through mathematical modeling. This innovative modeling method enables the algorithm to predict the propagation direction, speed and influence range of risk, providing forward-looking decision support for supervisory departments. The adaptive threshold determination mechanism dynamically adjusts the determination criteria according to different cargo types and supervision environments, enabling the algorithm to adapt to the complex and variable risk characteristics in the field of cargo supervision, avoiding the misjudgment problem caused by fixed threshold, and ensuring the flexibility and effectiveness of the supervision strategy. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are some embodiments of the application, and those skilled in the art can obtain other drawings based on these drawings without creative labor.

[0022] Figure 1 FIG. 1 is a schematic diagram of an embodiment of the big data-based smart cargo supervision data analysis processing method in the present application;

[0023] Figure 2 FIG. 5 is a schematic diagram of the risk propagation density distribution of different cargo types calculated by the multi-modal spatio-temporal risk propagation perception algorithm MSTRPA in the present application;

[0024] Figure 3 FIG. 6 is a schematic diagram of an embodiment of the big data-based smart cargo supervision data analysis processing system in the present application;

[0025] Figure 4 FIG. 7 is a structural schematic block diagram of the big data-based smart cargo supervision data analysis processing device in the present application. DETAILED DESCRIPTION

[0026] The present application provides a big data-based smart cargo supervision data analysis processing method and system. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0027] For the sake of understanding, the specific process of the embodiments of the present application is described below. Please refer to Figure 1 An embodiment of the big data-based smart cargo supervision data analysis processing method in the present application includes the following steps:

[0028] Step S1: Perform blockchain hash verification and spatio-temporal four-tuple standardization encoding processing on the multi-dimensional supervision sensing data to generate a blockchain verification supervision data set with cargo unique identification and state parameter vector;

[0029] Step S2: Construct a cargo spatio-temporal association graph according to the blockchain verification supervision data set, perform graph neural network embedding representation processing on the cargo nodes and multi-type association edges, and generate a cargo graph embedding vector that fuses risk propagation characteristics;

[0030] Step S3: Input the cargo graph embedding vector into the multi-modal spatio-temporal risk propagation perception algorithm for diffusion equation risk modeling processing, and output a three-tuple risk identification result;

[0031] Step S4: According to the triple risk identification result, the differential privacy federated learning collaborative decision processing is carried out, and the cross-domain supervision instruction set is generated;

[0032] Step S5: The cross-domain supervision instruction set is subjected to self-evolution knowledge graph updating and strategy gradient reinforcement learning optimization processing, and the intelligent supervision decision data is output.

[0033] It can be understood that the execution subject of the present application can be a big data-based intelligent cargo supervision data analysis processing system, and can also be a terminal or a server, which is not limited here. The server is taken as an example for illustration in the embodiments of the present application.

[0034] Specifically, through the deployment of Internet of Things sensor networks at the cargo production line, warehouse center, transportation vehicle and customs supervision point, radio frequency tag data, positioning trajectory information, temperature and humidity environment parameters and video monitoring images are collected. The blockchain hash verification adopts SHA-256 algorithm to perform hash calculation on the original data, generates data block header information containing time stamp, cargo identification, sensor type, data hash value and previous hash value, and then ensures data integrity and non-tamperability through the Byzantine fault tolerance consensus verification mechanism. The spatio-temporal four-tuple standardization coding converts the verified data into a four-dimensional vector containing spatial coordinates x, y, z and time coordinate t, and assigns a unique identification code to each cargo, generating a state parameter vector recording temperature, humidity, weight and other key information.

[0035] Based on the blockchain verification supervision data set, a cargo spatio-temporal association graph is constructed. Each cargo is identified as a node in the graph through cargo node extraction. The association relationship identification includes multiple types such as same batch association, same path association, same enterprise association and supply chain upstream and downstream association. The dynamic spatio-temporal graph structure organizes the cargo nodes and associated edges according to the time sequence, forming a graph structure with time sequence evolution characteristics. The graph neural network embedding representation calculates the neighbor feature mean of each node through the neighbor node feature aggregation mechanism, and then propagates the risk information between nodes through the message passing mechanism, finally generating a cargo graph embedding vector containing risk propagation characteristics.

[0036] The cargo graph is embedded into a vector input multi-modal spatio-temporal risk propagation perception algorithm MSTRPA for processing. Multi-modal feature fusion performs cross-modal attention calculation on the graph embedding features, real-time sensor data and historical behavior patterns, and calculates the attention weights between different modalities through a softmax function. The spatio-temporal attention double branch processes the features in the spatial and temporal dimensions respectively. The spatial attention calculates the spatial similarity between cargos, and the temporal attention controls the influence degree of historical information through an LSTM gating mechanism. The diffusion equation risk modeling uses a partial differential equation to describe the propagation dynamics of risk in the cargo network, calculates the speed, direction and intensity of risk propagation, and generates a risk propagation density distribution. The adaptive threshold judgment dynamically adjusts the risk level division standard according to historical supervision data, divides the risk into low, medium and high levels, and outputs a triple risk identification result containing the risk level, propagation path and confidence score.

[0037] According to the triple risk identification result, a differential privacy federated learning collaborative decision processing is performed. First, the cargos are classified according to the types of dangerous goods, ordinary goods, flammable and explosive goods and prohibited and restricted goods, and the sensitivity vectors of the risk levels, value intervals, transportation path types and enterprise credit levels of the cargos are calculated. The differential privacy noise addition adds appropriate noise to the risk assessment data to protect privacy through a Laplace noise generation mechanism according to the data sensitivity and allocates privacy budgets. Local training uses a gradient descent algorithm to iteratively optimize the noisy data and update the risk decision parameters. Multi-agent collaborative decision integrates the decision results of customs, ports, logistics enterprises and production enterprises, generates a global decision scheme through weighted aggregation, and forms a cross-domain supervision instruction set containing inspection levels, resource allocation and processing priorities.

[0038] The cross-domain supervision instruction set is updated by a self-evolution knowledge graph and optimized by a policy gradient reinforcement learning. The supervision knowledge triple extraction extracts entity relationships from dangerous goods inspection instructions, cargo release instructions, article detention instructions and enterprise credit update instructions to form a supervision knowledge entity set. The customs regulation ontology modeling constructs a knowledge graph containing cargo categories, regulation provisions, supervision measures and other entities, and the supervision case relationship reasoning discovers the implicit supervision rules through graph reasoning algorithms. The reinforcement learning mechanism takes the supervision accuracy and the clearance efficiency as reward signals, gives a positive reward when a high-risk cargo is accurately identified, and gives a negative reward when a mistake is made. The policy gradient algorithm adjusts the parameter weights of the MSTRPA algorithm according to the reward signals, optimizes the accuracy and efficiency of risk identification, and outputs the intelligent supervision decision data containing the risk warning threshold, the inspection priority weight and the resource allocation scheme.

[0039] In a specific embodiment, step S1 comprises:

[0040] The radio frequency tag data, positioning trajectory information, temperature and humidity environment parameters and video monitoring images collected by the Internet of Things sensor network deployed in the cargo production line, warehouse center, transportation vehicle and customs supervision point are subjected to data type identification processing to obtain classified and labeled multi-dimensional supervision sensor data.

[0041] The multi-dimensional supervision sensor data is subjected to hash algorithm calculation processing to obtain data block header information.

[0042] Based on the data block header information, Byzantine fault tolerance consensus verification processing is performed to obtain trusted supervision data.

[0043] The trusted supervision data is subjected to spatiotemporal four-tuple standardization encoding conversion processing to obtain a blockchain verification supervision data set with a unique cargo identifier and a state parameter vector.

[0044] Specifically, the original data collected by the Internet of Things sensor network is classified and labeled. The radio frequency tag data includes the RFID chip number, reading timestamp and signal strength of the cargo, the positioning trajectory information records the longitude and latitude of the GPS coordinates and the altitude, the temperature and humidity environment parameters monitor the real-time temperature value and relative humidity percentage around the cargo, and the video monitoring image captures the visual information of the cargo state and converts it into a digital image format. Different sources of data are classified and labeled by a data type identifier to form classified and labeled multi-dimensional supervision sensor data containing data source identification, data type code and timestamp. The hash algorithm calculation processing uses the SHA-256 encryption hash function to calculate the digest of the classified and labeled supervision sensor data. This algorithm converts input data of any length into a fixed-length 256-bit hash value. The calculation process includes four stages: message preprocessing, message grouping, compression function operation and final hash value output. The message preprocessing fills the original data to an integer multiple of 512 bits, the message grouping divides the preprocessed data into multiple 512-bit groups, the compression function performs 64 rounds of iteration operation on each group, and each round of operation uses different constants and logical functions. Finally, the data block header information containing the timestamp, cargo identifier code, sensor type code, current data hash value and previous block hash value is output.

[0045] The Byzantine fault-tolerant consensus verification process solves the consistency problem in the distributed regulatory network. The algorithm can tolerate no more than one-third of malicious nodes or faulty nodes in the network. The verification process is divided into three stages: pre-preparation, preparation and submission. In the pre-preparation stage, the master node broadcasts the pre-preparation message containing the block header information to all backup nodes. After the backup nodes verify the legality of the message, they enter the preparation stage. In the preparation stage, each node exchanges preparation messages with each other and collects a sufficient number of valid preparation messages. When a node receives more than two-thirds of the consensus preparation messages, it enters the submission stage. In the submission stage, each node confirms each other again and finally reaches a consensus. The data block that passes the verification is recognized by the network and forms trusted regulatory data.

[0046] The spatiotemporal four-tuple standardized encoding conversion process converts the trusted regulatory data into a unified data model format. The spatiotemporal four-tuple includes three spatial dimension coordinates x, y, z and a time dimension coordinate t. The spatial coordinates are standardized using the WGS84 geodetic coordinate system, which converts different sources of location information into numerical representations of longitude, latitude and elevation. The time coordinate is standardized using UTC coordinated universal time, which ensures that time information from different time zones has a unified reference basis. The unique identifier of the goods is generated using the UUID algorithm to generate a 128-bit globally unique identifier. The state parameter vector records the physical state information of the goods, including weight, volume, temperature, humidity, pressure and other key parameters. The encoding conversion process uniformly maps heterogeneous raw data into the field structure of the standard data model, and finally generates a blockchain verification regulatory data set with a unique identifier of the goods and a state parameter vector.

[0047] In a specific embodiment, step S2 comprises:

[0048] Based on the blockchain verification regulatory data set, the goods node extraction and association relationship identification process is performed to obtain a goods node set and a multi-type association edge set.

[0049] The goods node set and the multi-type association edge set are subjected to dynamic spatiotemporal graph structure construction processing to obtain a goods spatiotemporal association graph.

[0050] The neighbor node feature aggregation and message passing mechanism processing are performed on the goods spatiotemporal association graph to obtain a node embedding feature vector.

[0051] The node embedding feature vector is subjected to graph neural network embedding representation conversion processing to obtain a goods graph embedding vector that fuses the risk propagation characteristics.

[0052] Specifically, the cargo node extraction and association relationship identification process first extracts each cargo from the blockchain verified supervision dataset as an independent node in the graph structure, each cargo node contains basic attribute information such as cargo unique identifier, cargo type, weight, volume, etc., and the association relationship identification establishes edge connections by analyzing various association modes between cargos, same batch association refers to establishing edge connections between cargos from the same production batch, same path association refers to establishing edge connections between cargos passing through the same transportation path, same enterprise association refers to establishing edge connections between cargos belonging to the same production enterprise or logistics enterprise, and upstream and downstream association in the supply chain refers to establishing edge connections between cargos with direct or indirect business relationships in the supply chain. The association relationship is identified by traversing all cargo records and calculating the similarity matrix between cargos, and when the similarity exceeds the preset threshold, the corresponding type of association edge is established, and finally a cargo node set and a multi-type association edge set are formed.

[0053] The dynamic spatio-temporal graph structure construction process organizes static node and edge information into a graph structure with time evolution characteristics, the time dimension is divided according to the key time nodes of cargo supervision including production time, packaging time, loading time, transportation time, arrival time and inspection time, each time slice corresponds to a graph snapshot recording the state and association relationship of all cargos at that moment, and the spatial dimension constructs spatial proximity relationship according to the geographical location information of the cargo. When two cargos are in the same geographical area or have intersection in the transportation path, a spatial association edge is established, and a dynamic update mechanism adjusts the graph structure according to the real-time changes of the cargo state. When the cargo position moves, update the spatial association relationship, when the cargo state parameter changes, update the node attribute, when a new association relationship is found, add a new association edge. Through the time window sliding mechanism, the graph structure history within a fixed time range is maintained, forming a cargo spatio-temporal association graph containing time series and spatial topology information.

[0054] The neighbor node feature aggregation and message passing mechanism process realizes the propagation and fusion of information in the graph structure, the neighbor node feature aggregation calculates the weighted average value of the first-order neighbor node feature vector of each cargo node, the weight is determined according to the type and strength of the association edge, the weight of the same batch association is the highest, reflecting the close relationship in the production process, the weight of the same path association is the second, reflecting the risk propagation in the transportation process, the weight of the same enterprise association is medium, reflecting the influence of management level, and the weight of the supply chain association is lower, reflecting the influence of indirect business relationship. The aggregation process obtains the aggregated features by summing and normalizing the weighted feature vectors of all neighbor nodes, and the message passing mechanism propagates information between multiple levels of the graph, the first layer transmits the feature information of the direct neighbor, the second layer transmits the feature information of the second-order neighbor, and the third layer transmits the feature information of the third-order neighbor. After each layer transmission, the feature expression ability is enhanced through the nonlinear activation function processing, and the multi-layer transmission enables each node to perceive the graph structure information in a larger range, and finally generates a node embedding feature vector containing local and global features.

[0055] The graph neural network embedding representation conversion process maps the node embedding feature vector into a high-dimensional feature space and fuses the risk propagation characteristics. The embedding conversion projects the original feature vector into a higher-dimensional representation space through a multi-layer perceptron. The first layer performs linear transformation by multiplying the input feature vector with a weight matrix and adding a bias term. The second layer applies a ReLU activation function for nonlinear transformation to enhance the model's expression ability. The third layer performs linear transformation again and outputs the final embedding vector. The risk propagation characteristic fusion calculates the centrality index of each node in the graph to measure its importance in risk propagation. The degree centrality calculates the number of direct connections of a node to reflect its influence range. The betweenness centrality calculates the frequency of a node appearing in the shortest path to reflect its bridge role. The closeness centrality calculates the average distance from a node to all other nodes to reflect its propagation efficiency. These centrality indexes are integrated into the embedding vector as additional features to enhance the risk propagation modeling capability. The cargo graph embedding vector that fuses the risk propagation characteristics is generated.

[0056] In a specific embodiment, step S3 comprises:

[0057] The cargo graph embedding vector is subjected to multi-modal feature fusion and cross-modal attention mechanism processing to obtain a fused feature vector.

[0058] The fused feature vector is subjected to spatio-temporal attention double branch calculation processing to obtain a spatio-temporal weight feature.

[0059] The spatio-temporal weight feature is input into a multi-modal spatio-temporal risk propagation perception algorithm for diffusion equation modeling and calculation processing to obtain a risk propagation density distribution.

[0060] Based on the risk propagation density distribution, adaptive threshold determination and risk level classification processing are performed to obtain a triple risk identification result.

[0061] Specifically, the multimodal feature fusion and cross-modal attention mechanism process integrates the cargo graph embedding vector with other data modalities, including cargo graph embedding vectors, sensor real-time data streams, historical behavior pattern data, and external environmental factor data. The sensor real-time data stream records the current temperature, humidity, pressure, vibration, and other physical parameters of the cargo. The historical behavior pattern data contains historical information such as past transportation paths, stay times, and inspection records. The external environmental factor data includes environmental variables such as weather conditions, traffic conditions, and port congestion. The cross-modal attention mechanism achieves feature fusion by calculating the correlation weights between different modalities. First, each modality data is mapped to a unified feature space. Then, the dot product of the query vector and the key vector is calculated to obtain the attention score. Next, the attention score is normalized by the softmax function to obtain the attention weight. Finally, the attention weight is multiplied by the value vector and summed to obtain the fused feature vector, which integrates graph structure information, real-time state information, historical behavior information, and environmental background information.

[0062] The spatio-temporal attention dual-branch computation process handles the spatial and temporal dimension information of the fused feature vector respectively. The spatial attention branch calculates the spatial correlation between goods in different geographical locations by constructing a spatial similarity matrix based on the Euclidean distance between goods, the shortest path distance in the transportation network, and the administrative division relationship of geographical areas. The spatial attention weight is assigned according to the spatial proximity between goods, with closer goods receiving higher attention weights. The spatial branch outputs a spatial feature vector containing location-related information. The temporal attention branch handles the temporal correlation between goods at different time points by controlling the influence of historical information on the current state through the LSTM gating mechanism. The forget gate determines which outdated historical information to discard, the input gate determines which new state information to retain, and the output gate controls the information content output at the current time. The temporal branch outputs a temporal feature vector containing temporal dependency relationships. The dual-branch computation results are merged into spatio-temporal weight features through a weighted fusion mechanism, which contains both the spatial distribution pattern and the temporal evolution law of the goods.

[0063] The diffusion equation modeling and calculation process is the core link of the multi-modal spatio-temporal risk propagation perception algorithm MSTRPA. The algorithm uses a partial differential equation to describe the propagation dynamics of risks in the cargo network. The diffusion equation modeling considers four key processes of risk generation, propagation, growth, and decay. The risk generation term describes the relationship between the risk generation ability of the cargo itself and its type, state, and environmental conditions. Hazardous cargo has a higher risk generation rate, and package damage or temperature abnormalities can significantly increase the risk generation speed. The risk propagation term uses a diffusion operator to describe the transfer process of risk between adjacent cargos. The propagation speed is inversely proportional to the correlation strength and spatial distance between cargos. The risk growth term uses a logistic growth model to describe the self-amplification process of risk under favorable conditions. When environmental conditions deteriorate or supervision is relaxed, the risk growth rate increases. The risk decay term describes the inhibitory effect of regulatory measures and safety measures on risk. Effective inspection and protection measures can reduce risk density. The diffusion equation is numerically solved by finite difference method, which discretizes the time and space domain into grid points. The rate of change of risk density is calculated at each grid point. The risk propagation density distribution at different times and spatial locations is obtained through iterative calculation.

[0064] The adaptive threshold determination and risk level classification process determines the risk level of the cargo based on the risk propagation density distribution. The adaptive threshold mechanism dynamically adjusts the risk determination standard according to historical supervision data and the current supervision environment. The basic threshold is set to an initial value according to different cargo types. The basic threshold of hazardous cargo is lower, reflecting its high-risk characteristics. The basic threshold of ordinary cargo is higher, reflecting its relative safety. The threshold adjustment factor considers factors such as the adequacy of current regulatory resources, port congestion, and seasonal risk changes. When regulatory resources are tight, the threshold is appropriately increased to reduce the number of cargos that need to be focused on. When new risk patterns are found, the threshold of related types of cargo is lowered. The risk level classification maps the risk density value to three levels of low risk, medium risk, and high risk. Low-risk cargos use the green channel for fast clearance. Medium-risk cargos are subject to intensive inspection and sampling detection. High-risk cargos are subject to comprehensive inspection and detailed audit. The triple risk identification result includes risk level identification, risk propagation path information, and confidence score. The risk level identification uses numerical coding to represent the severity of risk. The risk propagation path records the propagation trajectory and impact range of risk in the cargo network. The confidence score quantifies the credibility of the risk identification result, reflecting the accuracy of the algorithm's prediction.

[0065] Figure 2A schematic diagram of the risk propagation density distribution of different cargo types calculated by the multi-modal spatio-temporal risk propagation perception algorithm MSTRPA in the embodiments of the present application. Through the diffusion equation modeling and calculation processing of the multi-modal spatio-temporal risk propagation perception algorithm MSTRPA, the risk propagation density distribution results of four main cargo types are obtained. Among them, the dangerous goods show the highest high-risk density (0.82) and the lowest low-risk density (0.03), which reflects its inherent high-risk characteristics; the high-risk density of flammable and explosive goods is 0.75, and the medium-risk density is 0.20; the risk distribution of restricted and prohibited goods is relatively balanced, with a high-risk density of 0.68; ordinary goods show the characteristics of low-risk density dominance (0.50), with a high-risk density of only 0.15. The risk propagation density distribution results provide a quantitative basis for subsequent adaptive threshold judgment and risk level classification processing, verifying that the MSTRPA algorithm can accurately identify the risk propagation characteristics of different cargo types and achieve accurate risk modeling based on the diffusion equation.

[0066] In a specific embodiment, step S4 comprises:

[0067] Classifying the triple risk identification results according to the cargo types of dangerous goods, ordinary goods, flammable and explosive goods, and restricted and prohibited goods to obtain classified risk assessment data;

[0068] Adding differential privacy noise to the classified risk assessment data and performing local training processing to obtain privacy-protected risk decision parameters;

[0069] Based on the risk decision parameters, performing multi-agent collaborative decision processing of customs, ports, logistics enterprises and production enterprises to obtain cross-domain collaborative decision results;

[0070] Performing supervision resource allocation and inspection level matching processing on the cross-domain collaborative decision results to obtain a cross-domain supervision instruction set.

[0071] Specifically, according to the cargo attribute information in the triple risk identification result, accurate classification is carried out, the dangerous chemical category includes toxic chemicals, corrosive substances, radioactive materials and other chemical substances with potential hazards, the classification is based on the physical and chemical characteristics of the chemical, such as flash point, toxicity grade, corrosion intensity, the ordinary cargo category covers daily consumer goods, mechanical equipment, textiles and other goods without special danger, the classification standard mainly considers the use, material and safety of the goods, the flammable and explosive category includes gasoline, alcohol, fireworks and other substances that are easy to burn or explode, the classification is based on key indicators such as ignition temperature, explosion limit concentration and burning speed, the prohibited and restricted goods category includes goods that are clearly prohibited or restricted for import and export by laws and regulations, including ivory products, endangered plants and animals, drug raw materials, etc., the classification process determines the legal status of the goods by querying the customs commodity code database and the international control list, and the classification result reorganizes the risk level, transmission path and confidence score of each cargo according to the four categories, forming a classification risk assessment data containing category identification, risk characteristics and regulatory requirements.

[0072] Differential privacy noise addition and local training processing solve the privacy protection problem in multi-agent collaborative decision-making. Differential privacy is a mathematical framework for quantifying and controlling the risk of privacy leakage in data publishing. This technology protects the privacy of individual information by adding carefully designed random noise to the original data. The noise addition process first calculates the global sensitivity of the classification risk assessment data. Global sensitivity represents the maximum impact of a single data record change on the query result. Sensitivity calculation considers the variation range of dangerous chemical risk level, the fluctuation range of cargo value, the deviation degree of transportation path and the adjustment range of enterprise credit rating. Laplace noise generation uses double exponential distribution to generate random noise that meets the requirements of differential privacy. The noise amplitude is proportional to the sensitivity and inversely proportional to the privacy budget. Privacy budget allocation is differentiated according to the sensitivity of different categories of goods. Dangerous chemicals are allocated a smaller privacy budget to achieve stronger privacy protection, and ordinary goods are allocated a larger privacy budget to maintain data usability. Local training processing uses the gradient descent algorithm to iteratively optimize the noise-added risk assessment data. The training goal is to learn the decision preferences and constraints of each agent. The training process calculates the gradient of the loss function through the back propagation algorithm and updates the decision parameter weight. Finally, the privacy-protected risk decision parameters containing the decision tendencies of each agent are generated.

[0073] The multi-agent collaborative decision-making process integrates the decision-making opinions of four subjects of customs, port, logistics enterprise and production enterprise. The customs subject focuses on the compliance inspection, tax collection and smuggling prevention of goods, and the decision-making focuses on risk level determination and inspection depth determination. The port subject focuses on the efficiency of goods handling, berth resource allocation and safety protection measures, and the decision-making considers the goods handling priority and safety isolation requirements. The logistics enterprise subject values the transportation time efficiency, cost control and service quality, and the decision-making content involves transportation path selection and delivery time arrangement. The production enterprise subject emphasizes product quality, supply chain stability and customer satisfaction, and the decision-making range includes production plan adjustment and quality control measures. The collaborative decision-making algorithm integrates the decision-making opinions of each subject by using a weighted voting mechanism. The weight distribution is determined according to the professional field and responsibility range of the subject. For dangerous chemical supervision, the weight of customs is the highest, reflecting its legal supervision responsibility. For transportation efficiency, the weight of logistics enterprise is the largest, reflecting its professional advantage. The voting process collects the handling suggestions of each subject for each goods, including inspection level, handling time limit and special requirements. The comprehensive decision score is calculated by weighted average, and the final decision scheme is generated according to the preset rules. The cross-domain collaborative decision-making result records the consistent opinions and disagreement points of each subject and the unified action scheme after coordination.

[0074] The supervision resource allocation and inspection level matching process allocates limited supervision resources and determines specific inspection measures according to the cross-domain collaborative decision-making result. The supervision resources include tangible and intangible resources such as inspection personnel, detection equipment, inspection site and time window. The resource allocation algorithm considers the availability, professionalism and cost-effectiveness of the resources. Personnel allocation matches inspectors with corresponding professional background according to the type of goods. Dangerous chemical inspection requires inspectors with chemical professional background, and food inspection requires inspectors with food safety professional background. Equipment allocation calls for corresponding detection instruments according to inspection needs. X-ray machine is used for internal structure inspection of goods, and gas chromatograph is used for chemical composition analysis. Site allocation arranges appropriate inspection areas according to the characteristics of goods. Dangerous chemicals need to be inspected in special areas with explosion-proof facilities, and perishable goods need to be inspected in temperature-controlled environments. Inspection level matching maps abstract risk level to specific inspection measures. Low-risk goods perform document audit and appearance inspection, including goods identification check, package integrity inspection and document consistency verification. Medium-risk goods implement sampling detection and key inspection, expanding the inspection range to goods composition analysis, weight and volume verification and source tracing investigation. High-risk goods are subjected to comprehensive inspection and detailed audit, including unboxing, item-by-item counting, composition detection, background investigation and risk assessment. The cross-domain supervision instruction set integrates resource allocation scheme, inspection execution plan and coordination requirements, forming a common action guide for each subject.

[0075] In a specific embodiment, differential privacy noise is added to the classified risk assessment data and local training processing is performed to obtain privacy-protected risk decision parameters, including:

[0076] The classification risk assessment data is processed for sensitivity calculation according to the dangerous chemical risk level, the cargo value interval, the transportation path type and the enterprise credit level, to obtain a data sensitivity vector;

[0077] The data sensitivity vector is processed for Laplace noise generation and privacy budget allocation to obtain a differential privacy noise parameter;

[0078] The differential privacy noise parameter is processed for noise superposition and privacy protection with the classification risk assessment data to obtain the noisy local training data;

[0079] The local training data is processed for gradient descent iteration and parameter update to obtain the privacy-protected risk decision parameter.

[0080] Specifically, the dangerous chemical risk level sensitivity calculation considers the toxicity level, the explosive equivalent, the corrosion intensity and the environmental pollution degree of the chemical, the toxicity level is divided into five levels from mild irritation to fatal toxicity, the sensitivity weight of each level increases in turn, the explosive equivalent calculates the influence of explosive power on sensitivity according to TNT equivalent, the corrosion intensity determines the sensitivity coefficient through pH value and metal corrosion rate, the environmental pollution degree evaluates the long-term impact of chemical leakage on soil, water and atmosphere, the cargo value interval sensitivity calculation divides the cargo value into four intervals of low value, medium value, high value and extra high value, the value sensitivity is proportional to the potential loss of tax evasion, the extra high value cargo has the highest sensitivity because it involves huge tax and luxury goods control, the transportation path type sensitivity analysis includes four types of direct path, transit path, multi-country path and special path, the direct path has the lowest sensitivity because the risk is controllable with simple links, the multi-country path has the highest sensitivity because it involves multiple jurisdictions and complex supervision coordination, the enterprise credit level sensitivity is evaluated according to the enterprise historical violation record, the financial situation, the management level and the industry reputation, the lower the credit level of the enterprise, the higher the sensitivity of the related data, the sensitivity calculation combines the sensitivity scores of the four dimensions into a comprehensive sensitivity vector through weighted summation, the weight allocation is dynamically adjusted according to the key attention areas of different regulatory scenarios.

[0081] The Laplace noise generation and privacy budget allocation process calculates a differential privacy noise parameter based on the data sensitivity vector. The Laplace noise is a kind of random noise with a double exponential distribution, whose probability density function is symmetric about zero and has a sharp peak shape. The noise generation process first determines the noise amplitude parameter, which is equal to the global sensitivity divided by the privacy budget. The global sensitivity represents the maximum impact of the change of a single data record on the query result. The global sensitivity value is determined by analyzing the maximum component in the sensitivity vector. The privacy budget is a core parameter in the differential privacy framework that controls the strength of privacy protection. The smaller the budget value, the stronger the privacy protection but the lower the data availability. The budget allocation adopts an adaptive strategy to differentiate the setting according to the type of goods and regulatory requirements. Dangerous goods are allocated a smaller privacy budget because they involve public safety sensitive information. Ordinary goods are allocated a larger privacy budget to maintain regulatory efficiency. Restricted goods are allocated the smallest privacy budget because they involve strict legal regulations. The Laplace noise is generated by an inverse transform sampling method. First, a uniformly distributed random number is generated, and then the uniform random number is converted to Laplace noise by the inverse cumulative distribution function of the Laplace distribution. The noise generation considers the positive and negative symmetry to ensure that it does not systematically bias in one direction. Finally, a differential privacy noise parameter vector consistent with the dimension of the sensitivity vector is formed.

[0082] The noise superposition and privacy protection process performs arithmetic operations on the differential privacy noise parameter and the classification risk assessment data. The noise superposition adds the noise vector and the original data vector at the corresponding positions by element-wise addition. The addition operation ensures that the randomness of the noise can mask the true value of the original data. The superposition process needs to maintain the semantic reasonableness of the data. When the risk level after adding noise appears negative, it is truncated to zero. When the risk score exceeds the theoretical upper limit, it is normalized. The privacy protection mechanism destroys the correlation pattern between data records through the randomness of the noise, making it impossible for attackers to infer the sensitive information of specific enterprises or goods through data analysis. The quantitative evaluation of the protection effect is calculated by the privacy loss function, which measures the similarity of the data distribution before and after adding noise. The lower the similarity, the better the privacy protection effect. Noise superposition also needs to consider data integrity constraints to ensure that the noise-added data still meets the basic requirements of business logic, such as the weight of goods cannot be negative, the risk level must be within the valid range, and the timestamp must comply with the time sequence logic. The local training data generated by the noise addition process maintains the statistical properties and distribution rules of the original data, while effectively protecting the sensitive information contained in the data.

[0083] The gradient descent iteration and parameter update process trains the machine learning model on the noisy local training data. Gradient descent is an optimization algorithm based on first-order derivative information, which minimizes the target loss function by iteratively updating parameters. The training process defines a loss function to measure the difference between the model's predicted results and the true labels. The loss function uses mean squared error to calculate the squared difference between the predicted risk level and the actual risk level. Gradient calculation obtains the direction and magnitude of parameter update by taking the partial derivative of the loss function. The backpropagation algorithm calculates the gradient value of each parameter in the neural network layer by layer. The parameter update uses a fixed learning rate strategy. The learning rate controls the step size of each parameter adjustment. A large learning rate leads to unstable training or even divergence, while a small learning rate leads to slow convergence. The iteration process repeatedly performs four steps: forward propagation, loss calculation, gradient calculation, and parameter update. After each iteration, the performance indicators on the validation set are calculated to evaluate the model training effect. When the validation performance does not improve for consecutive iterations, the training is terminated to avoid overfitting. The model parameters obtained through training include the weight coefficients and bias terms of each feature, which reflect the influence of different risk factors on the final decision. The privacy-protected risk decision parameters retain the core information of the original data for accurate decision-making, while the sensitive information is protected by differential privacy technology to avoid privacy leakage.

[0084] In a specific embodiment, step S5 comprises:

[0085] The dangerous chemical inspection instruction, the cargo release instruction, the article detention instruction and the enterprise credit update instruction in the cross-domain supervision instruction set are subjected to supervision knowledge triple extraction processing to obtain a supervision knowledge entity set;

[0086] The supervision knowledge entity set is subjected to customs regulation ontology modeling and supervision case relationship reasoning processing to obtain a self-evolution knowledge graph;

[0087] Based on the self-evolution knowledge graph, supervision accuracy evaluation and customs clearance efficiency reward signal calculation processing are performed to obtain a reinforcement learning reward value;

[0088] The reinforcement learning reward value is subjected to strategy gradient calculation and algorithm parameter optimization processing to obtain optimized decision strategy parameters;

[0089] According to the decision strategy parameters, parameter update processing is performed on the cargo supervision decision-making process to obtain intelligent supervision decision-making data containing risk warning threshold, inspection priority weight and resource allocation scheme.

[0090] Specifically, the regulatory knowledge triple extraction process extracts structured knowledge from four types of instructions in the cross-domain regulatory instruction set. The dangerous chemical inspection instruction includes three core elements: cargo type, inspection item, and safety requirement. The extraction process identifies the chemical name as the subject entity, the inspection method as the relation predicate, and the compliance standard as the object entity, forming a triple structure such as "acrylic acid - needs to be - component detection". The triple extraction of the cargo release instruction identifies the cargo number, release condition, and time limit requirement, generating a knowledge triple such as "ordinary textiles - meet the conditions - fast clearance". The article detention instruction extracts three elements: illegal goods, detention reason, and disposal measures, constructing a relation triple such as "ivory products - because of violation - endangered species protection regulations". The enterprise credit update instruction extracts the enterprise name, credit change, and change reason, forming a knowledge representation such as "chemical enterprise A - credit rating reduction - due to non-standard packaging". The triple extraction uses named entity recognition and dependency syntax analysis techniques. Named entity recognition identifies entity types such as cargo name, enterprise name, and legal provisions through pre-trained models. Dependency syntax analysis determines the syntactic relationship and semantic role between entities. The extraction results are processed to remove duplicates and standardized to form a regulatory knowledge entity set.

[0091] The customs regulations ontology modeling and regulatory case relation reasoning process constructs a domain-specific knowledge organization structure. The customs regulations ontology modeling organizes regulatory knowledge using a hierarchical classification system. The top-level concepts include cargo categories, regulatory measures, legal provisions, and inspection procedures. The cargo category branch includes subcategories such as dangerous chemicals, ordinary goods, flammable and explosive goods, and prohibited and restricted goods. Each subcategory is further divided into specific commodity types and customs codes. The regulatory measures branch includes inspection methods, handling procedures, and resource allocation. The legal provisions branch records the content and scope of application of relevant laws and regulations. The inspection procedure branch describes the standard operating procedures for different types of goods. The ontology modeling constructs a knowledge framework by defining semantic associations such as hierarchical relationships, equivalence relationships, and disjoint relationships. The regulatory case relation reasoning uses a rule-based reasoning engine to discover implicit knowledge. The reasoning rules include transitive reasoning, inheritance reasoning, and combination reasoning. Transitive reasoning discovers new risk associations based on the logic "if A is dangerous and B is of the same class as A, then B is also dangerous". Inheritance reasoning transfers attributes from superior concepts to subordinate concepts using the hierarchical structure. Combination reasoning derives complex regulatory conclusions by combining multiple simple rules. The reasoning results are checked for consistency and conflict resolution to ensure the logical consistency of the knowledge graph. The self-evolving knowledge graph dynamically updates and expands knowledge content through a continuous learning mechanism.

[0092] The supervision accuracy evaluation and the clearance efficiency reward signal calculation process establish a feedback mechanism of reinforcement learning. The supervision accuracy evaluation calculates the prediction accuracy by comparing the actual inspection result with the predicted risk level. The accuracy rate calculation formula is the number of correctly predicted goods divided by the total number of predicted goods. Correct prediction includes accurate identification of high-risk goods and correct judgment of low-risk goods. False positive rate and false negative rate are used as supplementary indicators of accuracy rate to measure false high-risk prediction and missed high-risk goods, respectively. The clearance efficiency evaluation measures the average processing time of goods from declaration to release. The efficiency indicator considers the processing time difference of goods of different risk levels and the resource occupation situation. The reward signal calculation adopts a multi-objective optimization strategy to balance the accuracy and efficiency. When the algorithm accurately identifies high-risk goods and timely discovers violations, a high positive reward is given. When the algorithm makes a false prediction, resulting in resource waste or safety hazards, a negative reward is given. The reward function design considers the relative severity of different types of errors. The negative reward for missing dangerous chemicals is much greater than the negative reward for false reporting of ordinary goods. The reward signal is obtained by weighted summation of the accuracy reward, efficiency reward and safety reward. The weight distribution reflects the basic principle that safety is prioritized over efficiency in supervision. The reinforcement learning reward value serves as a guide signal for algorithm optimization to drive parameter adjustment and strategy improvement.

[0093] The policy gradient calculation and algorithm parameter optimization process updates the decision strategy using a gradient-based optimization method. The policy gradient algorithm determines the parameter update direction by calculating the gradient of the policy function with respect to the parameters. The policy function is defined as the probability distribution of selecting a specific action in a given state. The state includes the characteristics of the goods, historical information and environmental conditions. The action corresponds to different supervision decision options, including release, inspection, detention, etc. The gradient calculation uses the REINFORCE algorithm to estimate the policy gradient. This algorithm samples action trajectories from the policy using the Monte Carlo sampling method. The value of each action is calculated based on the cumulative reward of the trajectory. The policy gradient is equal to the product of the action value and the logarithmic gradient of the policy function, and the expectation value of all sampled trajectories. The parameter update uses the stochastic gradient ascent method to adjust the parameters in the direction of increasing expected reward. The learning rate controls the step size of parameter update. A too large learning rate leads to unstable training, and a too small learning rate leads to slow convergence. The algorithm parameter optimization includes the adjustment of feature weights, threshold parameters and fusion coefficients in the multi-modal spatio-temporal risk propagation perception algorithm MSTRPA. The optimization process converges to the optimal policy parameter configuration through repeated iterations. The optimized decision strategy parameters can maximize the clearance efficiency while ensuring supervision safety.

[0094] The cargo supervision decision-making process parameter updating process applies the optimized decision-making strategy parameters to the actual supervision process. The parameter updating includes adjustment of three core components: risk early warning threshold, inspection priority weight, and resource allocation scheme. The risk early warning threshold is updated based on historical supervision data and the current threat situation to dynamically adjust the risk judgment criteria for different types of cargo. The threshold for dangerous goods is adjusted considering seasonal factors and changes in international security situation, and the threshold for ordinary goods is adjusted to reflect trade volume fluctuations and supervision resource availability. The inspection priority weight is updated to redistribute the importance of different risk factors. The weights of risk factors such as temperature anomalies, package damage, and poor enterprise credit are adjusted based on the latest violation case analysis results. The resource allocation scheme is updated to optimize personnel arrangement, equipment scheduling, and time allocation. During high-risk periods, the number of inspection personnel is increased, special inspection equipment is matched for special cargo types, and a rapid response mechanism is activated in emergency situations. The smart supervision decision-making data integrates all updated parameters and configuration information to form a complete decision support dataset to guide daily supervision work.

[0095] The above describes the data analysis and processing method for smart cargo supervision based on big data in the embodiments of the present application. The data analysis and processing system for smart cargo supervision based on big data in the embodiments of the present application is described below. Please refer to Figure 3 An embodiment of the data analysis and processing system for smart cargo supervision based on big data in the embodiments of the present application includes:

[0096] An encoding module for blockchain hash verification and spatio-temporal four-tuple standardization encoding processing of multi-dimensional supervision sensing data, generating a blockchain verification supervision dataset with a unique cargo identifier and a state parameter vector;

[0097] An embedding module for constructing a cargo spatio-temporal association graph based on the blockchain verification supervision dataset, performing graph neural network embedding representation processing on cargo nodes and multiple types of associated edges, and generating a cargo graph embedding vector that integrates risk propagation characteristics;

[0098] A modeling module for inputting the cargo graph embedding vector into a multi-modal spatio-temporal risk propagation perception algorithm for diffusion equation risk modeling processing, and outputting a three-tuple risk identification result;

[0099] A decision-making module for performing differential privacy federated learning collaborative decision-making processing based on the three-tuple risk identification result, and generating a cross-domain supervision instruction set;

[0100] An output module for performing self-evolution knowledge graph updating and strategy gradient reinforcement learning optimization processing on the cross-domain supervision instruction set, and outputting smart supervision decision-making data.

[0101] The above Figure 3The big data-based smart goods supervision data analysis processing system in the embodiment of the application is described in detail from the perspective of a modular functional entity. The big data-based smart goods supervision data analysis processing device in the embodiment of the application is described in detail from the perspective of hardware processing.

[0102] With reference to Figure 4 The embodiment of the application also provides a big data-based smart goods supervision data analysis processing device. The big data-based smart goods supervision data analysis processing device can be a server, and the internal structure thereof can be as shown in Figure 4 The big data-based smart goods supervision data analysis processing device comprises a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer is used to provide computing and control capabilities. The memory of the big data-based smart goods supervision data analysis processing device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The database of the big data-based smart goods supervision data analysis processing device is used to store corresponding data in the embodiment. The network interface of the big data-based smart goods supervision data analysis processing device is used to communicate with an external terminal through network connection. The computer program is executed by the processor to implement the above method.

[0103] Those skilled in the art can understand that Figure 4 The structure shown in the embodiment is only a block diagram of part of the structure related to the scheme of the application, and does not constitute a limitation on the big data-based smart goods supervision data analysis processing device to which the scheme of the application is applied.

[0104] The application also provides a computer readable storage medium. The computer readable storage medium can be a non-volatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium stores instructions. When the instructions are run on a computer, the computer executes the steps of the big data-based smart goods supervision data analysis processing method.

[0105] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the system, the system and the unit described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein.

[0106] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the entire or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a big data-based intelligent cargo supervision data analysis processing device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0107] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A big data-based intelligent cargo supervision data analysis processing method, characterized in that, The method comprises: Step S1: performing blockchain hash verification and space-time four-tuple standardization coding processing on multi-dimensional supervision sensing data to generate a blockchain verification supervision data set with a unique cargo identifier and a state parameter vector; Step S2: constructing a cargo space-time correlation graph according to the blockchain verification supervision data set, performing graph neural network embedding representation processing on cargo nodes and multiple types of associated edges, and generating a cargo graph embedding vector that fuses risk propagation characteristics; Step S3: inputting the cargo graph embedding vector into a multi-modal space-time risk propagation perception algorithm for diffusion equation risk modeling processing, and outputting a three-tuple risk identification result; Step S4: performing differential privacy federated learning collaborative decision processing according to the three-tuple risk identification result to generate a cross-domain supervision instruction set, including: performing cargo type classification processing on the three-tuple risk identification result according to dangerous goods, ordinary goods, flammable and explosive goods, and prohibited and restricted goods to obtain classified risk assessment data; performing differential privacy noise addition and local training processing on the classified risk assessment data to obtain privacy-protected risk decision parameters; performing multi-agent collaborative decision processing of customs, ports, logistics enterprises and production enterprises based on the risk decision parameters to obtain a cross-domain collaborative decision result; performing supervision resource allocation and inspection level matching processing on the cross-domain collaborative decision result to obtain a cross-domain supervision instruction set; The differential privacy noise addition and local training processing on the classified risk assessment data to obtain privacy-protected risk decision parameters comprises: performing sensitivity calculation processing on the classified risk assessment data according to dangerous goods risk level, cargo value interval, transportation path type and enterprise credit level to obtain a data sensitivity vector; performing Laplace noise generation and privacy budget allocation processing on the data sensitivity vector to obtain a differential privacy noise parameter; performing noise superposition and privacy protection processing on the differential privacy noise parameter and the classified risk assessment data to obtain noisy local training data; performing gradient descent iteration and parameter update processing on the local training data to obtain privacy-protected risk decision parameters; Step S5: constructing a self-evolving knowledge graph according to the cross-domain supervision instruction set; performing supervision accuracy evaluation and customs clearance efficiency reward signal calculation based on the self-evolving knowledge graph to obtain a reinforcement learning reward value; performing policy optimization based on the reinforcement learning reward value to update the parameters of the cargo supervision decision-making process, and obtaining intelligent supervision decision-making data. 2.The big data based smart cargo supervision data analysis processing method according to claim 1, characterized in that, The step S1 comprises: Performing data type identification processing on radio frequency tag data, positioning trajectory information, temperature and humidity environment parameters and video monitoring images collected by an Internet of Things sensor network deployed on cargo production lines, warehouse centers, transportation vehicles and customs supervision points to obtain multi-dimensional supervision sensing data with classification labels; Performing hash algorithm calculation processing on the multi-dimensional supervision sensing data to obtain data block header information; Performing Byzantine fault tolerance consensus verification processing based on the data block header information to obtain trusted supervision data; The trusted supervision data is subjected to spatio-temporal four-tuple standardization coding conversion processing to obtain a blockchain verification supervision data set with a cargo unique identifier and a state parameter vector. 3.The big data based smart cargo supervision data analysis processing method according to claim 1, characterized in that, The step S2 comprises: Based on the blockchain verification supervision data set, cargo node extraction and correlation relationship identification processing is performed to obtain a cargo node set and a multi-type correlation edge set; The cargo node set and the multi-type correlation edge set are subjected to dynamic spatio-temporal graph structure construction processing to obtain a cargo spatio-temporal correlation graph; The cargo spatio-temporal correlation graph is subjected to neighbor node feature aggregation and message passing mechanism processing to obtain a node embedding feature vector; The node embedding feature vector is subjected to graph neural network embedding representation conversion processing to obtain a cargo graph embedding vector fused with risk propagation characteristics. 4.The big data based smart cargo supervision data analysis processing method according to claim 1, wherein, The step S3 comprises: The cargo graph embedding vector is subjected to multi-modal feature fusion and cross-modal attention mechanism processing to obtain a fusion feature vector; The fusion feature vector is subjected to spatio-temporal attention double-branch calculation processing to obtain a spatio-temporal weight feature; The spatio-temporal weight feature is input into a multi-modal spatio-temporal risk propagation perception algorithm for diffusion equation modeling calculation processing to obtain a risk propagation density distribution; Based on the risk propagation density distribution, adaptive threshold judgment and risk level classification processing are performed to obtain a triple risk identification result. 5.The big data based smart cargo supervision data analysis processing method according to claim 1, wherein, The step S5 comprises: The dangerous chemical inspection instruction, the cargo release instruction, the article detention instruction, and the enterprise credit update instruction in the cross-domain supervision instruction set are subjected to supervision knowledge triple extraction processing to obtain a supervision knowledge entity set; The supervision knowledge entity set is subjected to customs regulation ontology modeling and supervision case relationship reasoning processing to obtain a self-evolution knowledge graph; Based on the self-evolution knowledge graph, supervision accuracy evaluation and customs clearance efficiency reward signal calculation processing are performed to obtain a reinforcement learning reward value; The reinforcement learning reward value is subjected to policy gradient calculation and algorithm parameter optimization processing to obtain optimized decision strategy parameters; According to the decision strategy parameters, parameter update processing is performed on the cargo supervision decision flow to obtain a smart supervision decision data containing a risk early warning threshold, an inspection priority weight, and a resource allocation scheme.

6. A big data-based intelligent cargo supervision data analysis processing system, characterized in that, The smart cargo supervision data analysis processing system based on big data comprises: An encoding module for performing blockchain hash verification and spatio-temporal four-tuple standardization coding processing on multi-dimensional supervision sensing data to generate a blockchain verification supervision data set with a cargo unique identifier and a state parameter vector; An embedding module for constructing a cargo spatio-temporal correlation graph according to the blockchain verification supervision data set, and performing graph neural network embedding representation processing on cargo nodes and multi-type correlation edges to generate a cargo graph embedding vector fused with risk propagation characteristics; A modeling module for inputting the cargo graph embedding vector into a multi-modal spatio-temporal risk propagation perception algorithm for diffusion equation risk modeling processing, and outputting a triple risk identification result; The decision module is configured to perform differential privacy federated learning collaborative decision processing according to the triplet risk identification result, and generate a cross-domain supervision instruction set, including: performing cargo type classification processing on the triplet risk identification result according to dangerous goods, ordinary goods, flammable and explosive goods, and prohibited and restricted goods, to obtain classification risk assessment data; performing differential privacy noise addition and local training processing on the classification risk assessment data to obtain privacy-protected risk decision parameters; performing multi-agent collaborative decision processing of customs, ports, logistics enterprises and production enterprises based on the risk decision parameters to obtain a cross-domain collaborative decision result; and performing supervision resource allocation and inspection level matching processing on the cross-domain collaborative decision result to obtain the cross-domain supervision instruction set; The differential privacy noise addition and local training processing on the classification risk assessment data to obtain privacy-protected risk decision parameters includes: performing sensitivity calculation processing on the classification risk assessment data according to dangerous goods risk levels, cargo value intervals, transportation path types and enterprise credit levels to obtain a data sensitivity vector; performing Laplace noise generation and privacy budget allocation processing on the data sensitivity vector to obtain differential privacy noise parameters; performing noise superposition and privacy protection processing on the differential privacy noise parameters and the classification risk assessment data to obtain local training data after noise addition; and performing gradient descent iteration and parameter update processing on the local training data to obtain privacy-protected risk decision parameters; The output module is configured to construct a self-evolution knowledge graph according to the cross-domain supervision instruction set; perform supervision accuracy evaluation and customs clearance efficiency reward signal calculation based on the self-evolution knowledge graph to obtain a reinforcement learning reward value; and perform strategy optimization based on the reinforcement learning reward value, update parameters of a cargo supervision decision process, and obtain intelligent supervision decision data.

7. A big data-based smart cargo supervision data analysis processing device, characterized in that, The computer program is stored in the memory and can be run on the processor, and the processor implements the big data-based intelligent cargo supervision data analysis processing method in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, causes the processor to perform the big data-based intelligent cargo supervision data analysis processing method in any one of claims 1 to 5.

Images