Supply chain environmental risk early warning method and device, equipment and storage medium

Abstract

The invention discloses a supply chain environment risk early warning method, device and equipment and a storage medium, and relates to the technical field of supply chain management, and the method comprises the steps: obtaining pollution data of production, transportation and storage links, and carrying out the standardization processing and Hash encryption uplink storage, thereby guaranteeing the credibility of the data. A graph neural network model is utilized to evaluate pollution data, a time sequence prediction model is combined to dynamically adjust a threshold value and calculate a pollution residual error, and based on a comparison result of the residual error and the threshold value, multi-stage early warning response is triggered, so that the accuracy and the automation level of supply chain environmental risk management early warning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of supply chain management, and in particular to a supply chain environmental risk early warning method, device, equipment and storage medium. Background Art

[0002] Traditional supply chain environmental risk assessment and early warning methods rely primarily on manual data acquisition, single-source data analysis tools, and inefficient emergency response mechanisms. These methods typically involve the following steps: first, environmental data is collected manually or through centralized equipment; then, the collected data is processed using relatively simple analytical tools; and finally, when environmental risks occur, manual intervention measures based on preset rules are implemented. However, this model hinders data sharing across nodes, creating "data silos" and hindering overall risk management efficiency, making it particularly difficult to cope with the complex and ever-changing global supply chain. Therefore, a supply chain environmental risk assessment and dynamic early warning method based on blockchain and artificial intelligence is urgently needed to improve the efficiency of supply chain environmental risk management and early warning. Summary of the Invention

[0003] The main purpose of this application is to provide a supply chain environmental risk early warning method, device, equipment and storage medium, aiming to solve the technical problem of how to improve the efficiency of supply chain risk early warning.

[0004] To achieve the above objectives, this application proposes a supply chain environmental risk early warning method, including: Obtaining pollution data, including pollution data from production, transportation, and storage; performing standardization processing on the pollution data to obtain processed pollution data; The processed pollution data is hashed and stored on the chain to obtain a trusted supply chain network; Convert each node and corresponding relationship in the trusted supply chain network into a graph structure, and construct a pollution contribution assessment model based on a graph neural network; Inputting the processed pollution data into the pollution contribution assessment model for calculation to obtain the pollution contribution; Calculating the pollution contribution through a preset time series prediction model, and outputting a dynamic adaptive threshold and a pollution residual, wherein the preset time series prediction model includes a long short-term memory network and a Transformer encoder; Executing a corresponding warning operation according to a comparison result of the pollution residual and the dynamic adaptive threshold; The step of performing a corresponding warning operation according to a comparison result between the pollution residual and the dynamic adaptive threshold comprises: Comparing the node pollution residual with the dynamic adaptive threshold to obtain the warning level; Calculate the normalized residual and spatiotemporal weight of each node according to the specific parameters corresponding to the warning level; Calculating based on the normalized residual amount and the spatiotemporal weight to generate an overall supply chain risk value; According to the overall risk value, a multi-level warning and response mechanism is matched to execute corresponding warning operations.

[0005] In one embodiment, the step of hashing and encrypting the processed contamination data and storing it on-chain to obtain a trusted supply chain network includes: Get the blockchain network; The processed contaminated data is distributed and calculated through IPFS to obtain a hash summary. The specific formula is: in Using the SHA-256 algorithm, Represents a hash digest, which is a unique content identifier; Obtaining a node identity public key based on the hash summary and the blockchain network; The node identity is obtained by calculation based on the node identity public key. The specific formula is: in, Representation node The public key of represents a hash function; The node identity and the hash summary are aggregated in combination with the timestamp and stored on the blockchain network to obtain a trusted supply chain network.

[0006] In one embodiment, the step of converting the nodes and corresponding relationships in the trusted supply chain network into a graph structure and constructing a pollution contribution assessment model based on a graph neural network includes: The supply chain network is transformed and modeled as a directed attribute graph, where nodes represent enterprises and edges represent logistics relationships; Based on the directed attribute graph, a node attribute matrix and an edge attribute matrix are defined, wherein the node attribute matrix includes pollution data, emission frequency, operation level, business scale, industry category, and historical compliance rate, and the edge attribute matrix includes logistics relationship type, supply frequency, cooperation intensity, and physical distance; Initialization is performed based on the node attribute matrix and the edge attribute matrix to obtain the initial node features and the corresponding neighbor node features. The formula for the initial node features is: in, represents the node feature map weight matrix, is a bias term, and the formula for the corresponding neighbor node feature is: in, represents the adjacency matrix; The attention weights between nodes are calculated based on the initial node features and the neighbor node features through the multi-channel graph attention mechanism. The specific formula is: in, Indicates the number of layers of message passing, For all neighbor nodes The normalization function of is a nonlinear activation function, represents the weight vector of the attention mechanism, For nodes of The set of neighbor nodes of order , Representation node exist In the link order features, Represents neighbor nodes exist In the link order features, Represents the edge attribute The mapping transformation, Represents vector concatenation operation; Based on the attention weight, the initial node features and the corresponding neighbor node features are iteratively updated to obtain the feature representation of each node. The specific formula is: in, for In the link layer weight matrices, Indicates the total number of layers of message passing, Representation node The degree, Representation node The degree, Representation node of The set of neighbor nodes of order , is the adjacency matrix, Represents neighbor nodes exist In the link Layer characteristics; Calculate the total pollution contribution of the node by using the characteristic representation; Constructing a pollution contribution assessment model based on the total pollution contribution; The step of calculating the total pollution contribution of the node by using the characteristic representation includes: representing the direct pollution contribution, conducted pollution contribution, and historical violation correction term of the computing node according to the characteristics; The direct pollution contribution, the conducted pollution contribution and the historical violation correction item are weighted and summed using a preset weight coefficient to obtain the total pollution contribution of the node.

[0007] In one embodiment, the step of calculating the pollution contribution using a preset time series prediction model and outputting a dynamic adaptive threshold and a pollution residual includes: Obtain historical pollution contribution data and construct a time series sample set; Based on the time series sample set, a long short-term memory network of a preset time series prediction model is used to capture short-term dependency features and generate a hidden state sequence; Input the hidden state sequence into the Transformer encoder of a preset time series prediction model, and extract long-term trend features through a multi-head attention mechanism; Performing a prediction based on the long-term trend characteristics to obtain a predicted pollution contribution; Calculating based on the predicted pollution contribution and the pollution contribution to obtain a pollution residual; Based on the pollution residual and the node collaborative weight matrix, the warning threshold multiple is dynamically adjusted to generate a dynamic adaptive threshold.

[0008] In one embodiment, the step of dynamically adjusting the warning threshold multiple based on the prediction residual in combination with the node collaborative weight matrix to generate a dynamic adaptive threshold includes: Obtain the logistics relationship between nodes; Performing exponentially weighted moving average on the contamination residuals to obtain a mean and a variance; Constructing a collaborative weight matrix based on the logistics relationship between the nodes; Based on the collaborative weight matrix, the basic threshold multiple is modified to generate an early warning threshold, where the basic threshold multiple is a preset constant value used to adjust the size of the early warning threshold; A dynamic adaptive threshold is generated according to the warning threshold in combination with the mean and variance.

[0009] In one embodiment, the step of performing corresponding warning operations according to the multi-level warning and response mechanism matching the overall risk value includes: When the overall risk value exceeds the preset first-level risk threshold and the overall risk value does not exceed the preset second-level risk threshold, a first-level warning is triggered and an internal risk warning is issued; When the overall risk value exceeds the preset second-level risk threshold and the overall risk value does not exceed the preset third-level risk threshold, a second-level warning is triggered, cross-node linkage is initiated, and existing production, transportation, and warehousing processes are optimized; When the overall risk value exceeds the preset third-level risk threshold and the overall risk value does not exceed the preset fourth-level risk threshold, a third-level warning is triggered to adjust the supply chain resource allocation and reduce the business volume with high-risk nodes; When the overall risk value exceeds the preset level 4 risk threshold, a level 4 warning is triggered, the supply chain reconstruction procedure is executed and business dealings with high-risk nodes are suspended.

[0010] In one embodiment, the step of obtaining pollution data includes: Collect pollutant concentration data by deploying sensors; Obtain transport route and emission data through the positioning device of the transport vehicle; Collect pollution indicators in the storage process through storage environment monitoring equipment; Performing spatiotemporal alignment of the pollutant concentration data, the transport route and emission data, and the pollution index with regional environmental indicators acquired through satellite remote sensing to obtain multi-source data; The multi-source data is aggregated through federated learning technology to obtain contaminated data.

[0011] In addition, to achieve the above-mentioned purpose, the present application also proposes a supply chain environmental risk early warning device, which includes: An acquisition module is used to acquire pollution data, including pollution data from production, transportation, and storage; A processing module, configured to perform standardization processing on the pollution data to obtain processed pollution data; An encryption module, used to hash and encrypt the processed contamination data and store it on the chain to obtain a trusted supply chain network; A construction module is used to convert the nodes and corresponding relationships in the trusted supply chain network into a graph structure and build a pollution contribution assessment model based on a graph neural network; a calculation module, configured to input the processed pollution data into the pollution contribution assessment model for calculation to obtain a pollution contribution; A prediction module, configured to predict the pollution contribution through a preset time series prediction model and calculate a dynamic adaptive threshold and a pollution residual; An execution module is used to perform corresponding warning operations based on the comparison results of the pollution residual and the dynamic adaptive threshold; it is also used to compare the node pollution residual with the dynamic adaptive threshold to obtain the warning level; calculate the residual normalization amount and spatiotemporal weight of each node according to the specific parameters corresponding to the warning level; calculate based on the residual normalization amount and the spatiotemporal weight to generate the overall risk value of the supply chain; match the multi-level warning and response mechanism according to the overall risk value to perform corresponding warning operations.

[0012] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the supply chain environmental risk warning method as described above are implemented.

[0013] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the supply chain environmental risk early warning method as described above.

[0014] This application obtains pollution data from production, transportation, and warehousing, standardizes it, and stores it on-chain with hash encryption to ensure data credibility. It uses a graph neural network model to evaluate pollution data, and combines it with a time series prediction model to dynamically adjust thresholds and calculate pollution residuals. Based on the comparison of residuals with thresholds, it triggers a multi-level early warning response, improving the accuracy and automation of early warnings for supply chain environmental risk management. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0016] Figure 1 This is a flow chart of the first embodiment of the supply chain environmental risk early warning method of this application; Figure 2 This is a blockchain network architecture diagram of the first embodiment of the supply chain environmental risk early warning method of this application; Figure 3 This is a flow chart of the second embodiment of the supply chain environmental risk early warning method of this application; Figure 4 This is a flow chart of the third embodiment of the supply chain environmental risk early warning method of this application; Figure 5 This is a flow chart of the fourth embodiment of the supply chain environmental risk early warning method of this application; Figure 6 This is a schematic diagram of the module structure of the supply chain environmental risk early warning device of the first embodiment of the supply chain environmental risk early warning method of this application; Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the supply chain environmental risk early warning method in the embodiment of this application.

[0017] The purpose, features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0018] It should be understood that the specific embodiments described herein are merely used to explain the technical solutions of the present application and are not intended to limit the present application.

[0019] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0020] Amidst increasing globalization, supply chain networks are becoming increasingly complex, encompassing multiple links, from raw material sourcing to final product delivery to consumers. With this increasing complexity, environmental risks within the supply chain are becoming a significant challenge for both business operations and government regulation. In particular, pollutant emissions in key links such as production, transportation, and warehousing are increasingly impacting the environment. Traditional environmental risk management approaches, relying on manual data acquisition, single data analysis tools, and inefficient emergency response mechanisms, are no longer able to meet the demands of modern supply chain management. However, these approaches suffer from numerous deficiencies in data transparency, real-time monitoring capabilities, intelligent early warning, and multi-agent collaboration. For example, data authenticity and traceability are difficult to guarantee, static analysis methods cannot capture dynamic changes in the supply chain, and early warning mechanisms lack flexibility and personalization.

[0021] Therefore, this application proposes a supply chain environmental risk warning method to solve the above problems. The main solution of the embodiment of this application is: obtain pollution data; standardize the pollution data to obtain processed pollution data; hash the processed pollution data and store it on the chain to obtain a trusted supply chain network; convert each node and corresponding relationship in the trusted supply chain network into a graph structure, and construct a pollution contribution assessment model based on a graph neural network; input the processed pollution data into the pollution contribution assessment model for calculation to obtain the pollution contribution; predict the pollution contribution through a preset time series prediction model, calculate the dynamic adaptive threshold and pollution residual; and perform corresponding warning operations according to the comparison result of the pollution residual and the dynamic adaptive threshold.

[0022] Based on the above, the present application embodiment also provides a supply chain environmental risk early warning method, referring to Figure 1 , Figure 1This is a flow chart of the first embodiment of the supply chain environmental risk early warning method of this application. The method is applied to a blockchain network with multiple functional modules, such as Figure 2 The blockchain network architecture diagram shown in the figure includes a node enterprise unit, an environmental data collection and chain writing unit, a smart contract unit, a state storage and audit unit, a decentralized Oracle unit, and a management and response unit. The node enterprise unit is responsible for mapping each enterprise in the supply chain to a blockchain node. The environmental data collection and chain writing unit is responsible for obtaining the pollution data of the node enterprises in the three links of production, transportation and warehousing through the multi-source data network. The smart contract unit is used to define the supply chain graph structure, graph neural network training, pollution contribution assessment model, dynamic threshold calculation and the execution logic of the multi-level early warning and response mechanism. The state storage and audit unit is used to store all environmental data and model results. The decentralized Oracle unit realizes the trusted bridge from the chain to the chain. The management and response unit is responsible for executing the early warning operation when the pollution forecast exceeds the threshold.

[0023] In this embodiment, the supply chain environmental risk early warning method includes steps S10 to S70: Step S10: Obtain pollution data.

[0024] It's important to note that the environmental data collection and write-chain unit uses sensors and IoT devices deployed throughout the supply chain (production, transportation, and warehousing) to collect real-time pollution data from node companies. This data covers a wide range of pollutants, such as sulfur dioxide, nitrogen oxides, particulate matter, chemical oxygen demand, and volatile organic compounds, comprehensively reflecting the environmental impact of companies at all stages. This process provides a solid data foundation for subsequent risk assessments, dynamic early warnings, and response mechanisms.

[0025] Furthermore, step S10 also includes: collecting pollutant concentration data by deploying sensors, obtaining transportation route and emission data using the positioning devices on transport vehicles, and collecting pollution indicators in the warehousing process using warehousing environmental monitoring equipment. Specifically, first, real-time pollutant concentration data is collected by deploying a sensor network. These sensors can be installed in production facilities, transport vehicles, and warehousing environments to monitor emissions of key pollutants such as sulfur dioxide, nitrogen oxides, and particulate matter. This distributed monitoring approach not only provides highly accurate data but also covers every key link in the supply chain. Second, detailed transportation route and emission data is obtained using positioning devices (such as GPS) on transport vehicles. This includes not only vehicle trajectories but also emission intensity and fuel consumption along different road sections. Integrating with a geographic information system (GIS) allows for further analysis of pollution hotspots within specific areas and their impact areas, providing a scientific basis for optimizing logistics routes. Furthermore, environmental monitoring equipment is deployed in the warehousing process to collect pollution indicators such as solid waste emissions, volatile organic compound emissions, particulate matter emissions, and wastewater discharge. This data reflects the environmental impact of the warehousing process and helps identify potential pollution sources. Next, pollutant concentration data, transportation routes, emissions data, and pollution indicators are spatiotemporally aligned with regional environmental indicators obtained from satellite remote sensing to generate multi-source data. Finally, federated learning technology is used to aggregate this multi-source data to obtain pollution data. Specifically, to more comprehensively understand environmental conditions, it is necessary to spatiotemporally align the aforementioned pollutant concentration data, transportation routes, emissions data, and pollution indicators from storage processes with regional environmental indicators obtained from satellite remote sensing to obtain a multi-source dataset. Satellite remote sensing technology can provide a wide range of environmental background information, such as air quality index and changes in vegetation cover, to supplement the limitations of ground-based monitoring. Next, federated learning technology is used to aggregate this multi-source data to obtain the final combined pollution data. Federated learning is a distributed machine learning method that combines data from multiple participants for model training without sharing the original data. This method not only protects the data privacy of each node but also effectively integrates data from different sources to improve the generalization and accuracy of the prediction model.

[0026] Step S20: performing standardization processing on the pollution data to obtain processed pollution data.

[0027] It should be noted that the original pollution data obtained needs to be standardized due to the problems of dimensional differences and inconsistent data ranges. To this end, the dynamic industry benchmark standardization method is adopted in this embodiment. Specifically, this method calculates the standardization coefficients by analyzing the corresponding indicator values of the top 10% excellent companies and the bottom 10% lagging companies in the industry annual report. Then, these coefficients are used to transform the original pollution data, eliminate dimensional differences, and obtain standardized pollution data. The specific formula is: in, Indicates the number of node enterprises, the total number is , Representation link, Represents the production process, Represents the transportation link, Indicates the warehousing link, The corresponding indicators represent the top 10% of outstanding companies and the bottom 10% of lagging companies in the industry's annual reports. This process not only ensures cross-industry comparability of data, but also improves its transparency and credibility.

[0028] In step S30, the processed pollution data is hashed and stored on the chain to obtain a trusted supply chain network.

[0029] It's important to note that after standardizing the pollution data, it undergoes further processing to ensure its security and immutability. This process is achieved through hashing technology, ensuring the integrity and transparency of the data within the supply chain environmental risk assessment system. Hash encryption is a one-way encryption algorithm that converts data of any length into a fixed-length hash value.

[0030] Furthermore, step S30 also includes: obtaining a blockchain network, distributing and calculating the processed contaminated data through IPFS to obtain a hash digest. Specifically, the standardized contaminated data is converted into a unique hash digest by using a secure hash algorithm (such as SHA-256). The hash digest, as the digital fingerprint of the data, is unique and irreversible, and can effectively prevent data tampering during transmission and storage. The specific formula is: in Using the SHA-256 algorithm, Represents the hash summary, which is a unique content identifier. This process ensures the integrity and consistency of data from acquisition to storage. Secondly, based on the hash summary and combined with the blockchain network, the node identity public key is obtained. Specifically, when each node joins the supply chain network, it generates a pair of keys, including a private key and a public key, through an asymmetric encryption algorithm (such as RSA or elliptic curve encryption algorithm). The private key is securely stored by the node for signing and decryption operations; the public key is publicly disclosed in the blockchain network for identity authentication and data encryption. Through the combination of hash summary and blockchain network, a unique and trusted digital identity public key can be generated for each node. Next, the node identity is obtained by calculation based on the node identity public key. Specifically, based on the node's public key, a unique node identity is generated through a hash function (such as SHA-256). The hash function converts the public key into a fixed-length hash value. This hash value serves as the node identity, ensuring the uniqueness and non-tamperability of the node identity. The specific formula is: in, Representation node The public key of Denotes a hash function. In the process of building a trusted supply chain network, ensuring secure data access and the correct execution of contract terms is crucial. To this end, smart contracts assign access rights to nodes based on their identities, thereby controlling which nodes have the right to operate specific data or execute certain contract terms. This mechanism is implemented through a simple discriminant formula: If the node's digital identity exists in the authorized node set , True is returned, indicating that the node has the corresponding operation permission; conversely, if the node's digital identity is not in the set, False is returned, indicating that the node does not have permission. Finally, the node identity and hash summary are aggregated in combination with the timestamp and stored on the chain to the blockchain network to obtain a trusted supply chain network. Specifically, the timestamp records the exact time of data generation or chain operation, providing a time dimension guarantee for data traceability. The aggregated node identity, hash summary and timestamp are further processed through an encryption algorithm to generate an unalterable blockchain record, which is then uploaded to the blockchain network to achieve a trusted bridge from the off-chain to the on-chain, and a trusted supply chain network is constructed. The specific formula is: in, Indicates the data record on the chain, A timestamp is used to record the precise time when data is generated or uploaded to the blockchain. This process not only ensures the authenticity and immutability of the data, but also enables transparent storage and traceability of data through blockchain's distributed ledger technology.

[0031] Step S40: Convert each node and corresponding relationship in the trusted supply chain network into a graph structure, and construct a pollution contribution assessment model based on a graph neural network.

[0032] It's important to note that by converting a trusted supply chain network with encrypted information into a graph structure and building a pollution contribution assessment model based on a graph neural network, we can not only clearly visualize the logistics relationships between nodes (enterprises) within the supply chain, but also more accurately quantify the pollution contribution of each node within the entire network.

[0033] Specifically, this process first maps the node companies in the supply chain into nodes in a graph structure, and defines logistics relationships as edges in the graph, thus constructing a directed attribute graph. Based on this, node and edge attribute matrices are defined, respectively recording multidimensional characteristic information of the node companies (such as pollution data, emission frequency, operation level, scale, industry category, and historical compliance rate), as well as characteristic information such as the type of logistics relationship, cooperation intensity, supply frequency, and physical distance of the edges.

[0034] Subsequently, a graph neural network (GNN) model was used to model the graph structure. During the initialization phase, each node's attribute vector was converted into an initial state vector through a linear transformation and embedded as input. During the model's iterations, attention weights for neighboring edges were calculated based on the feature representations of the node and its neighbors, and the node's feature representation was automatically updated. This process not only considers the node's own characteristics but also aggregates information from neighboring nodes, enabling the model to capture the propagation path and intensity of pollution within the supply chain network.

[0035] Finally, based on the updated node feature representation, the direct pollution contribution, conducted pollution contribution, and historical violation correction items of each node enterprise are calculated. These contributions are weighted and summed using preset weight coefficients to obtain the total pollution contribution of the node. On this basis, a pollution contribution assessment model is constructed, which comprehensively considers direct pollution contribution, conducted contribution, and violation correction items. Direct pollution contribution refers to the amount of pollution generated by the node enterprise itself during production, transportation, or warehousing; conducted contribution involves how pollution from upstream enterprises is transmitted downstream along the supply chain; and violation correction items are used to adjust the additional penalties brought about by historical violations to prevent accidental events from excessively affecting the overall score. The entire assessment process is automatically executed through smart contracts, and the results are written to the blockchain's state tree, ensuring transparency and immutability.

[0036] Step S50: Input the processed pollution data into a pollution contribution evaluation model for calculation to obtain the pollution contribution.

[0037] It's important to note that the pollution contribution assessment model is based on a graph neural network architecture. By constructing a supply chain graph, it maps node companies and their logistics relationships into nodes and edges within the graph, and associates pollution data with these nodes and edges. The graph neural network model utilizes a multi-layered information transfer mechanism to aggregate feature information from nodes and their neighbors, updating the node's feature representation.

[0038] Specifically, standardized pollution data is input into a pollution contribution assessment model built using a graph neural network (GNN). First, a smart contract module transforms the trusted supply chain network with encrypted information into a graph structure, where nodes represent enterprises and edges represent logistics relationships. Each node contains not only its own pollution data but also multi-dimensional feature information such as emission frequency, operation level, and scale. This rich attribute information provides a solid foundation for the subsequent calculation of each node's impact coefficient. Based on this graph structure, node and edge attribute matrices are defined, and node features and corresponding neighbor node features are initialized. Then, a multi-channel graph attention network (GAT) mechanism is used to calculate attention weights between nodes. This mechanism automatically learns and updates the feature representations of nodes and their neighbors. Through multiple iterative updates, a feature representation for each node is ultimately obtained. Next, the node's pollution contribution is calculated based on these feature representations. This calculation includes direct pollution contribution, conducted pollution contribution, and a correction for historical violations. Direct pollution contribution refers to the amount of pollution generated by a node enterprise during production, transportation, or warehousing. Conducted pollution contribution considers how pollution from upstream enterprises is transmitted down the supply chain. The historical violation modifier adjusts the additional penalties imposed by past violations to prevent isolated incidents from excessively impacting the overall score. The entire assessment process is automated through smart contracts, with the results written to the blockchain's state tree, ensuring transparency and immutability. This approach not only improves the understanding and management of environmental risks in the supply chain but also promotes a greener and more sustainable development model.

[0039] Step S60 : Calculate the pollution contribution through a preset time series prediction model, and output a dynamic adaptive threshold and pollution residual.

[0040] It's important to note that after standardizing the pollution data and converting it into a graph structure, the next step is to conduct in-depth analysis and prediction of this data using a pre-defined time series prediction model. This pre-defined time series prediction model is a LSTM-Transformer hybrid time series prediction model, which includes a long short-term memory network and a Transformer encoder. The LSTM (Long Short-Term Memory) network excels at learning short-term dependencies in pollution data and can effectively capture changing trends. The Transformer encoder, through its attention mechanism, enhances its understanding of long-term dependencies, which is particularly important when pollution events span long time periods. This hybrid model combines the advantages of both, resulting in more accurate predictions.

[0041] Next, we use an LSTM-Transformer hybrid time series forecasting model to predict pollution contributions. This process first involves constructing a historical pollution contribution series for each node, based on all historical pollution data stored in the state storage and audit unit. To ensure effective model training, we employ a sliding time window approach, constructing a data sample set and a true value set for each node from the past N time steps. This not only helps capture short-term pollution trends but also provides a foundation for long-term pattern recognition. After the forecast is complete, the pollution residual is calculated—the difference between the actual pollution contribution and the predicted value. Then, the threshold is dynamically updated using an exponentially weighted moving average. This approach accounts for the temporal fluctuations of pollution data and allows the threshold to adaptively adjust based on the latest pollution trends. Furthermore, to more comprehensively reflect the mutual influence between nodes in the supply chain network, we introduce a collaborative weight matrix that incorporates information on pollution fluctuations at other related nodes. This allows for alerts to be triggered even if the pollution level at a node does not directly exceed the threshold, if it is affected by increased pollution at neighboring nodes.

[0042] Step S70: executing a corresponding warning operation according to the comparison result between the pollution residual and the dynamic adaptive threshold.

[0043] It should be noted that the smart contract unit monitors the pollution residual of each node enterprise in real time and compares it with a dynamic adaptive threshold. The pollution residual refers to the difference between the actual pollution contribution and the predicted value. The dynamic adaptive threshold automatically adjusts based on fluctuations in pollution data to reflect the true level of current environmental risk.

[0044] Based on the comparison of the pollution residual and the dynamic adaptive threshold, the spatiotemporal weight of each node enterprise will be automatically calculated and stored in the blockchain state storage module. These weights will be used to assess the overall risk level of the supply chain network and trigger the corresponding network-level early warning and response mechanism. The early warning level is divided into multiple levels based on the size of the pollution residual and the degree of risk. In this embodiment, it is divided into blue, yellow, orange, and red warnings. Specifically, the judgment is: in , both represent weights, represents the dynamic adaptive threshold, Each warning level corresponds to different response measures, ensuring that each node in the supply chain can take appropriate action based on the degree of risk. Specifically, when the warning level is blue, a risk warning will be issued and the company will be required to review its data to confirm whether there are any data anomalies or potential risks. When the warning level is yellow, the company will be required to implement environmental protection training, quarterly spot checks, and other measures to reduce pollution risks. When the warning level is orange, the company will be required to make rectifications within a specified period and undergo monthly audits to ensure that the company is taking effective measures to control pollution. When the warning level is red, more stringent measures will be taken, such as freezing transactions and imposing penalties, to prevent further escalation of pollution risks.

[0045] This implementation ensures data credibility by collecting pollution data from production, transportation, and warehousing, standardizing it, and storing it on-chain with hash encryption. A graph neural network model is used to evaluate pollution data, and combined with a time series prediction model, thresholds are dynamically adjusted and pollution residuals are calculated. Comparisons between residuals and thresholds trigger multi-level early warning responses, improving the accuracy and automation of early warnings for supply chain environmental risk management.

[0046] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above embodiment 1 can be referred to the above introduction and will not be described in detail later. Figure 3 The supply chain environmental risk early warning method step S40 further includes steps S201 to S207: Step S201 : transforming and modeling the supply chain network into a directed attribute graph, where nodes represent enterprises and edges represent logistics relationships.

[0047] It should be noted that the supply chain is abstracted into a graph structure through the smart contract module. , where the node set Represents each enterprise in the supply chain, and the edge set Characterize the logistics relationship between enterprises and define the adjacency matrix . Will standardize the database The associations are mapped to the corresponding graph elements, where the production link data and the warehousing link data are bound to the nodes, and the transportation link data are bound to the edges. This graph structure can clearly reflect the interactions and dependencies between the nodes in the supply chain, providing a basic framework for subsequent pollution contribution assessment.

[0048] Step S202: defining a node attribute matrix and an edge attribute matrix based on the directed attribute graph.

[0049] It should be noted that the node attribute matrix includes pollution data (such as the emission volume of various pollutants), emission frequency (the frequency of pollutant emissions), operation scale (the scale of the enterprise's production activities), business scale (the economic scale of the enterprise), industry category (the type of industry to which the enterprise belongs) and historical compliance rate (the company's past compliance with regulations). The specific formula is: Among them, each dimension Represents nodes respectively In the link Pollution data, emission frequency, operation level, business scale, industry category and historical compliance rate indicators in the node pollution data From the standardized dataset , and through the graph structure Mapping and binding with supply chain topology: ,The remaining indicators are obtained from actual business links.

[0050] The edge attribute matrix includes the type of logistics relationship (such as raw material supply, product transportation, etc.), supply frequency (frequency of logistics activities), cooperation intensity (the degree of cooperation between enterprises) and physical distance (the actual distance between node enterprises). Specifically, it is: Each dimension Represents edges The corresponding logistics relationship type, supply frequency, cooperation intensity, and physical distance indicator information are obtained from the actual business links.

[0051] By defining the node attribute matrix and the edge attribute matrix, the graph structure of the supply chain can be updated to a directed attribute graph of the supply chain. This data not only provides rich input information for the graph neural network model, but also enables the model to more accurately capture the propagation paths and intensity of pollution within the supply chain network. For example, by analyzing the pollution data and emission frequency of nodes, the direct pollution contribution of individual enterprises can be assessed; and by analyzing the logistics relationship types and supply frequency of edges, the risk of conducted pollution between enterprises can be assessed.

[0052] Ultimately, these node and edge attribute information provide a solid data foundation for building a pollution contribution assessment model, enabling the model to accurately identify the pollution responsibility of each node.

[0053] Step S203 , initializing according to the node attribute matrix and the edge attribute matrix to obtain initial node features and corresponding neighbor node features.

[0054] It should be noted that in the initialization phase, the attribute vector of each node in the node attribute matrix is first processed through a linear transformation to generate the initial feature representation of each node. This process can be expressed as: in, represents the node feature map weight matrix, is the bias term. In order to capture the relationship between nodes, it is necessary to extract the neighbor node features corresponding to each node. This step is completed based on the edge attribute matrix. Through the adjacency matrix, the neighbor nodes of each node can be determined and the initial feature representation of these neighbor nodes can be extracted. According to the formula, the features of the neighbor nodes can be obtained as : in, Represents the adjacency matrix.

[0055] In step S204, attention weights between nodes are obtained by calculating the multi-channel graph attention mechanism based on the initial node features and the neighbor node features.

[0056] It should be noted that the core of the multi-channel graph attention mechanism is to learn different relationships between nodes simultaneously through multiple attention heads, thereby capturing richer and more complex network structure information. Each attention head independently calculates the attention weights between nodes, which reflect the importance of nodes in a specific channel. In this way, the model can better capture the heterogeneous relationships between nodes, such as different types of logistics relationships, cooperative relationships of different strengths, etc. Based on the initialization node features Features of its neighboring nodes , calculate the first layer neighbor edge attention weight . And in the subsequent In the layer, based on the node features of the previous layer and the features of the neighboring nodes, the node of Attention weights of order neighbors , the specific formula is: in, Indicates the number of layers of message passing, For all neighbor nodes The normalization function of is a nonlinear activation function, represents the weight vector of the attention mechanism, For nodes of The set of neighbor nodes of order , Representation node exist In the link order features, Represents neighbor nodes exist In the link order features, Represents the edge attribute The mapping transformation, Represents a vector concatenation operation.

[0057] Step S205: Iteratively update the initial node features and the corresponding neighbor node features based on the attention weight to obtain the feature representation of each node.

[0058] It should be noted that based on and , automatically updating nodes through alternating iterations The characteristics of the node are finally obtained exist Link Layer features, the specific formula is: in, for In the link layer weight matrices, Indicates the total number of layers of message passing, Representation node The degree, Representation node The degree, Representation node of The set of neighbor nodes of order , is the adjacency matrix, Represents neighbor nodes exist In the link Layer features.

[0059] Step S206: Calculate the total pollution contribution of the node through feature representation.

[0060] It's important to note that this method accurately quantifies the environmental risk responsibility of each node enterprise by comprehensively considering factors such as a node's direct pollution contribution, conducted pollution contribution, and historical violations. This process is based on the node's feature representation, which is obtained by aggregating and updating the feature information of the node and its neighbors using a graph neural network (GNN). This representation comprehensively reflects the pollution characteristics and relationships of nodes in the supply chain network.

[0061] Furthermore, step S206 also includes: calculating the direct pollution contribution, conducted pollution contribution and historical violation correction items of the node based on the characteristic representation. Specifically, the direct pollution contribution, conducted pollution contribution and historical violation correction items can be obtained based on the characteristics of the node. Direct pollution contribution refers to the pollutant emissions directly generated by the node enterprise in the production, transportation and warehousing links. This part of the calculation is based on the characteristic representation of the node, including multi-dimensional features such as pollution data, emission frequency, and operation level. Before calculating the direct pollution contribution, calculate the node In the link The specific formula for the pollution link weight in is: in, represents the Sigmoid activation function, represents the linearly changing weight matrix, is the bias term. Based on the above pollution link weights, the direct pollution contribution is obtained. The specific formula is: .

[0062] The conducted pollution contribution is obtained. The conducted pollution contribution refers to the contribution of node enterprises in transferring pollution to other nodes through the logistics relationship of the supply chain network. The specific formula is: in, The hierarchical attenuation coefficient indicates the degree to which pollution decays with the length of the propagation path. The historical violation correction item refers to the number of environmental violations committed by the node enterprise in the past three years and is used to correct the pollution contribution. This part is calculated based on the number of historical violations publicly disclosed, and the total pollution contribution is adjusted using the adjustment coefficient. The specific formula is: in, is the adjustment coefficient, which is used to control the impact of historical violations on the total contribution. is the number of historical violations that have been made public. Then, the direct pollution contribution, conducted pollution contribution and historical violation correction item are weighted and summed using the preset weight coefficient to obtain the total pollution contribution of the node. Specifically, the preset weight coefficient is set according to the importance of each pollution contribution in the environmental risk assessment. The direct pollution contribution reflects the pollution emissions of the node enterprise itself and usually has a higher weight; the conducted pollution contribution takes into account the impact of the node enterprise on other nodes through the supply chain network, and its weight reflects the complexity of pollution propagation in the supply chain; the historical violation correction item is based on the company's past environmental violation record, and the penalty for violations is increased by adjusting the coefficient. These weight coefficients can be adjusted according to actual business needs and historical data to ensure the accuracy and rationality of the calculation results. The specific weight coefficient formula is: in is the direct pollution contribution weight coefficient, is the weight coefficient of the contribution to the conducted pollution, is the weight coefficient of the historical violation correction item, is the weight mapping matrix, is the normalization function, Based on the above results, the total pollution contribution is obtained, and the specific formula is: , The above total pollution contribution needs to take into account the impact of various factors on environmental risks and adjust the weight coefficient accordingly. For example, if the direct pollution in a supply chain has a more significant impact on the environment, the weight coefficient can be appropriately increased. If the long-term impact of historical violations on environmental risks is greater, the The weight of .

[0063] Step S207: constructing a pollution contribution evaluation model based on the total pollution contribution.

[0064] This implementation transforms the supply chain network into a directed attribute graph, defines node and edge attribute matrices, initializes node features, and uses a multi-channel graph attention mechanism to calculate attention weights between nodes. Node features are iteratively updated based on the attention weights, and the node's total pollution contribution is calculated. Ultimately, a pollution contribution assessment model is constructed to improve the efficiency and transparency of supply chain environmental risk management.

[0065] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the first embodiment can be referred to the above introduction and will not be described in detail later. Figure 4 The supply chain environmental risk warning step S50 further includes steps S301 to S305: Step S301: Obtain historical pollution contribution data and construct a time series sample set.

[0066] It should be noted that the historical pollution contribution data stored in the state storage and audit unit of the blockchain network is relied upon. By extracting the historical pollution data of each node in the production, transportation and warehousing links from these records, a comprehensive historical pollution contribution time series can be established. Specifically, for each node enterprise in the supply chain, , we arrange the pollution contributions in the past period in chronological order to form a time series sample set, the formula is: , represents the total number of nodes in the supply chain network, Represents the number of historical moments.

[0067] Next, the sliding window technique is used to construct the past The data sample set and the true value set at each moment provide the basis for subsequent time series analysis. The specific formula is: in is the data sample set, is the true value set, and to speed up the convergence Min-mix normalization is used to keep the range in [0,1]. The formula is: in represents the minimum value in the pollution contribution sequence, In this way, we can capture the changing trend of pollution contribution of each node over time and use it to train the prediction model.

[0068] Step S302 : Based on the time series sample set, a long short-term memory network of a preset time series prediction model is used to capture short-term dependency features and generate a hidden state sequence.

[0069] It should be noted that the hidden state not only contains the input information at the current moment, but also integrates the information of all previous time steps, so that the model can "remember" the past state and affect the current output. The core of the LSTM model lies in its unique gating mechanism, including the forget gate, input gate, and output gate. The forget gate determines which past information needs to be discarded; the input gate is responsible for updating the new information to the cell state; and the output gate controls the impact of the current cell state on the output. In this way, LSTM can effectively avoid the gradient vanishing or exploding problems encountered by traditional RNNs when processing long time series, thereby more accurately capturing the subtle differences in the pollution contribution over time, and ultimately generating a hidden state sequence. , the specific formula is: in Forget gate, which controls the degree of retention of historical information. represents the forget gate weight, is the forget gate bias, represents the input gate, represents the input gate weight, represents the input gate bias, Represents the unit status value, Indicates the updated value of the unit door status, represents the unit state weight, represents the cell state bias, represents the Sigmoid activation function, represents element-wise product, represents the output gate weight, represents the output gate bias, Indicates the final output.

[0070] In step S303, the hidden state sequence is input into the Transformer encoder of the preset time series prediction model, and long-term trend features are extracted through the multi-head attention mechanism.

[0071] It should be noted that the hidden state sequence is input into the Transformer encoder of the pollution contribution assessment model. The Transformer is known for its powerful parallel processing capabilities and efficient attention mechanism, and is particularly suitable for extracting complex patterns and long-term trends in long time series.

[0072] Specifically, the hidden state sequence generated from the LSTM is passed as input to the Transformer encoder. At this stage, the Transformer processes these hidden states through its multi-head self-attention mechanism. The multi-head attention mechanism allows the model to focus on different parts of the sequence at the same time and learn complex interaction patterns in the input data. Each "head" can independently focus on different aspects of the sequence, thereby capturing multi-level information in the changes in pollution contribution. Specifically, for the hidden state of each time step, the Transformer calculates the query, key, and value matrices, and then calculates the attention score through the multi-head attention mechanism. The specific formula is: in, represents the query matrix, represents the bond matrix, represents the value matrix, represents the dimension of the key, Indicates the An attention head, represents vector concatenation, Represents the output transformation matrix.

[0073] Step S304: perform prediction based on the long-term trend characteristics to obtain the predicted pollution contribution.

[0074] It should be noted that based on the feature representation output by the Transformer encoder, the pollution contribution prediction value of each node is obtained through full connection layer mapping. The specific formula is: in is the output of the Transformer encoder, representing the node At the moment The pollution trend embedding vector is expressed as follows: in, represents a feedforward neural network, Representation layer normalization.

[0075] Next, set the loss function to optimize the model. The specific loss function formula is: in is the total number of samples, is the regularization coefficient, .

[0076] Step S305 , performing calculation based on the predicted pollution contribution and the pollution contribution to obtain the pollution residual.

[0077] It should be noted that the difference between the actual pollution contribution and the predicted value is calculated, that is, the pollution residual. The pollution residual reflects the deviation between the model prediction and the actual situation. The specific formula is: in, is the actual pollution contribution, This adaptive mechanism enables intelligent adjustment of warning limits based on the latest pollution data fluctuations, improving the sensitivity and accuracy of warnings.

[0078] Step S306 , dynamically adjusting the warning threshold multiple based on the pollution residual combined with the node collaborative weight matrix to generate a dynamic adaptive threshold.

[0079] It should be noted that in supply chain environmental risk assessment, dynamically adjusting warning thresholds is the key to ensuring accurate warnings and timely responses.

[0080] Furthermore, step S306 includes obtaining the logistics relationships between nodes. Specifically, these relationships not only reflect the topological structure of the supply chain network but also reveal the paths of pollutant transmission along the supply chain. By abstracting the supply chain into a graph structure through the smart contract module and mapping each node and the logistics relationships between them into this graph, the connections between upstream and downstream companies can be clearly identified. The pollution residuals are then subjected to an exponentially weighted moving average to obtain the mean and variance. To further optimize prediction accuracy, we use the exponentially weighted moving average (EWMA) method to smooth these residuals, obtaining the mean and variance of the residuals. The formula is: in, is the mean of the residuals, is the variance of the residual, is a smoothing coefficient with a range of (0, 1). A collaborative weight matrix is constructed based on the logistics relationships between nodes. This matrix quantifies the correlation between the pollution contributions of each node. For example, if an increase in pollution at a node is likely to trigger an increase in pollution levels at its upstream and downstream nodes, the corresponding weight will be higher. This matrix not only considers direct upstream and downstream relationships but can also be extended to indirect influences throughout the entire supply chain network, thus fully reflecting the interactions between nodes.

[0081] The basic threshold multiple is modified based on the collaborative weight matrix to generate the warning threshold. Specifically, the basic threshold multiple is a preset constant value used to adjust the size of the warning threshold. The threshold multiple is modified by the collaborative weight matrix to respond to the impact of pollution changes of other nodes in the network. The specific formula is: in, As the base multiple, is the collaborative weight matrix between nodes, satisfying , is the residual mapping matrix. Finally, the dynamic adaptive threshold is generated based on the warning threshold combined with the mean and variance. The specific formula is: in, is the weighted average, is the threshold multiple, is the weighted mean standard deviation.

[0082] This implementation constructs a historical pollution contribution sequence, uses LSTM to capture short-term dependencies, and Transformer to extract long-term trends, to predict future pollution contributions and calculate residuals. Based on the residuals and the node collaborative weight matrix, it dynamically adjusts warning thresholds and generates dynamic adaptive thresholds, improving the accuracy, real-time nature, and intelligence of environmental risk management.

[0083] Based on the first embodiment of the present application, in the fourth embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction and will not be described in detail later. Figure 5 The supply chain environmental risk warning step S60 further includes steps S401 to S403: Step S401 : Compare the node pollution residual with the dynamic adaptive threshold to obtain the warning level.

[0084] It should be noted that the warning level is divided into multiple levels according to the size of the pollution residual and the degree of risk, and the corresponding warning actions include freezing transactions, rectification within a specified period, or environmental protection training. In this embodiment, the warning levels are divided into blue, yellow, orange, and red. Specifically, the judgment is: in , both represent weights. Each warning level corresponds to different response measures, ensuring that each node in the supply chain can take appropriate actions based on the degree of risk. Specifically, when the warning level reaches blue, a risk warning will be issued and the company will be required to review its data to confirm whether there are any data anomalies or potential risks. When the warning level reaches yellow, the company will be required to implement measures such as environmental protection training and quarterly spot checks to reduce pollution risks. When the warning level reaches orange, the company will be required to make rectifications within a specified period and undergo monthly audits to ensure that the company is taking effective measures to control pollution. When the warning level reaches red, more stringent measures will be taken, such as freezing transactions and imposing penalties, to prevent further escalation of pollution risks.

[0085] Step S402: Calculate the normalized residual and spatiotemporal weight of each node according to the specific parameters corresponding to the warning level.

[0086] It should be noted that, based on the specific parameters corresponding to the warning level, the residual normalization quantity and spatiotemporal weight of each node are further calculated to comprehensively assess the overall risk level of the supply chain. First, the pollution residual of each node is normalized to eliminate the dimensional differences between different links and different enterprises, and obtain a unified and comparable residual normalization quantity. This indicator reflects the relative degree to which the node enterprise deviates from its predicted pollution contribution at the current moment. Then, the node is assigned a corresponding spatiotemporal weight based on its supply chain position and its historical impact. The time dimension reflects the persistence of pollution fluctuations, while the space dimension reflects the criticality of the node in the graph structure, such as the number of upstream and downstream connections, transmission influence, etc. The specific formula is: in, is the contamination residual, Indicates a period of time, is the weighted average, represents the weighted mean standard deviation, For node enterprises The degree in the network, is the time decay coefficient.

[0087] Step S403: Calculate based on the normalized residual and the spatiotemporal weight to generate the overall risk value of the supply chain.

[0088] It should be noted that the normalized residual and the time and space weights are weighted and summed to generate the overall risk value of the supply chain. The specific formula is: It realizes the quantitative assessment of risks from individual nodes to the global network, supports dynamic risk monitoring and graded response decision-making, and improves the scientific nature and collaborative governance capabilities of supply chain environmental risk management.

[0089] Step S404: performing corresponding warning operations according to the multi-level warning and response mechanism matching the overall risk value.

[0090] It should be noted that when the overall risk value does not exceed the preset first-level risk threshold, a first-level warning is triggered and an internal risk warning is issued; when the overall risk value exceeds the preset first-level risk threshold and the overall risk value does not exceed the preset second-level risk threshold, a second-level warning is triggered, cross-node linkage is initiated, and existing production, transportation, and warehousing processes are optimized; when the overall risk value exceeds the preset second-level risk threshold and the overall risk value does not exceed the preset third-level risk threshold, a third-level warning is triggered, supply chain resource allocation is adjusted, and business volume with high-risk nodes is reduced; when the overall risk value exceeds the preset third-level risk threshold, a fourth-level warning is triggered, supply chain reconstruction procedures are executed, and business dealings with high-risk nodes are suspended. Specifically, based on the mapping of the overall risk value to the four-level warning mechanism, the specific formula is: in, This mechanism automatically triggers cross-organizational collaborative governance processes through smart contracts to ensure the immediacy and consistency of response measures. For example, when the pollution residual exceeds the first-level warning threshold, an internal risk warning is issued and the data review and model reassessment process begins; if the second-level warning standard is reached, a cross-node linkage meeting is initiated to promote the green optimization of production, transportation and warehousing processes; the third-level warning triggers the automatic reduction of business quotas at high-risk nodes and optimizes resource allocation; the fourth-level warning suspends related business transactions and initiates the supply chain reconstruction process.

[0091] This embodiment determines the warning level by comparing the node pollution residuals with the dynamic adaptive threshold, calculates the normalized residual amount and the spatiotemporal weight to generate the overall risk value, and matches the multi-level warning mechanism to execute the response operation, thereby achieving accurate warning and dynamic response, and improving the efficiency and transparency of supply chain environmental risk management.

[0092] Based on the first embodiment of this application, this application also provides a supply chain environmental risk early warning device, please refer to Figure 6 , the device comprises: The acquisition module 10 is used to acquire pollution data, which includes pollution data in production, transportation and warehousing.

[0093] The processing module 20 is used to perform standardization processing on the pollution data to obtain processed pollution data.

[0094] The encryption module 30 is used to hash and encrypt the processed contamination data and store it on the chain to obtain a trusted supply chain network chain.

[0095] The construction module 40 is used to convert the nodes and corresponding relationships in the trusted supply chain network into a graph structure and construct a pollution contribution assessment model based on a graph neural network.

[0096] The calculation module 50 is used to input the processed pollution data into the pollution contribution evaluation model for calculation to obtain the pollution contribution.

[0097] The prediction module 60 is used to predict the pollution contribution through a preset time series prediction model, calculate the dynamic adaptive threshold and the pollution residual, and the preset time series prediction model includes a long short-term memory network and a Transformer encoder.

[0098] Execution module 70 is used to perform corresponding warning operations based on the comparison results of the pollution residual and the dynamic adaptive threshold; it is also used to compare the node pollution residual with the dynamic adaptive threshold to obtain the warning level; calculate the residual normalization amount and spatiotemporal weight of each node according to the specific parameters corresponding to the warning level; calculate based on the residual normalization amount and spatiotemporal weight to generate the overall risk value of the supply chain; match the multi-level warning and response mechanism according to the overall risk value to execute corresponding warning operations.

[0099] The supply chain environmental risk warning device provided in this application, employing the supply chain environmental risk warning method described in the aforementioned embodiments, can address the technical problem of improving the efficiency of supply chain risk warnings. Compared to the prior art, the supply chain environmental risk warning device provided in this application achieves the same beneficial effects as the supply chain environmental risk warning method described in the aforementioned embodiments. Other technical features of the supply chain environmental risk warning device are the same as those disclosed in the aforementioned embodiments and are not further elaborated here.

[0100] In one embodiment, the encryption module 30 is also used to obtain a blockchain network; the processed contaminated data is distributedly stored and calculated through IPFS to obtain a hash summary; the node identity public key is obtained based on the hash summary combined with the blockchain network; the node identity is obtained by calculation based on the node identity public key; the node identity and the hash summary are aggregated in combination with the timestamp and stored on the chain in the blockchain network to obtain a trusted supply chain network.

[0101] In one embodiment, the construction module 40 is further configured to convert and model the supply chain network into a directed attribute graph, wherein nodes represent enterprises and edges represent logistics relationships; define a node attribute matrix and an edge attribute matrix based on the directed attribute graph, wherein the node attribute matrix includes pollution data, emission frequency, operation level, business scale, industry category, and historical compliance rate, and the edge attribute matrix includes logistics relationship type, supply frequency, cooperation intensity, and physical distance; initialize according to the node attribute matrix and the edge attribute matrix to obtain initial node features and corresponding neighbor node features; calculate the attention weights between nodes based on the initial node features and the neighbor node features through a multi-channel graph attention mechanism; iteratively update the initial node features and the corresponding neighbor node features based on the attention weights to obtain feature representations of each node; calculate according to the feature representation to obtain the total pollution contribution of the node; and construct a pollution contribution assessment model based on the total pollution contribution; the step of calculating according to the feature representation to obtain the total pollution contribution of the node includes: calculating the direct pollution contribution, conducted pollution contribution, and historical violation correction term of the node according to the feature representation; and performing weighted summation of the direct pollution contribution, conducted pollution contribution, and historical violation correction term using a preset weight coefficient to obtain the total pollution contribution of the node.

[0102] In one embodiment, the prediction module 60 is also used to obtain historical pollution contribution data; based on the historical pollution contribution data, a long short-term memory network of a preset time series prediction model is used to capture short-term dependency features and generate a hidden state sequence; the hidden state sequence is input into the Transformer encoder of the preset time series prediction model, and long-term trend features are extracted through a multi-head attention mechanism; prediction is performed based on the long-term trend features to obtain a predicted pollution contribution; calculation is performed based on the predicted pollution contribution and the pollution contribution to obtain a pollution residual; based on the pollution residual and the node collaborative weight matrix, the warning threshold multiple is dynamically adjusted to generate a dynamic adaptive threshold.

[0103] In one embodiment, the prediction module 60 is also used to obtain the logistics relationship between nodes; perform exponentially weighted moving average on the pollution residual to obtain the mean and variance; construct a collaborative weight matrix based on the logistics relationship between nodes; correct the basic threshold multiple based on the collaborative weight matrix to generate an early warning threshold; and generate a dynamic adaptive threshold based on the early warning threshold combined with the mean and variance.

[0104] In one embodiment, the execution module 70 is also used to trigger a first-level warning and issue an internal risk warning when the overall risk value does not exceed the preset first-level risk threshold; trigger a second-level warning when the overall risk value exceeds the preset first-level risk threshold and the overall risk value does not exceed the preset second-level risk threshold, start cross-node linkage and optimize existing production, transportation and warehousing processes; trigger a third-level warning when the overall risk value exceeds the preset second-level risk threshold and the overall risk value does not exceed the preset third-level risk threshold, adjust the supply chain resource allocation and reduce the business volume with high-risk nodes; when the overall risk value exceeds the preset third-level risk threshold, trigger a fourth-level warning, execute the supply chain reconstruction procedure and suspend business dealings with high-risk nodes.

[0105] In one embodiment, the acquisition module 10 is also used to collect pollutant concentration data by deploying sensors; obtain transportation path and emission data through the positioning device of the transportation vehicle; collect pollution indicators of the warehousing link through warehousing environment monitoring equipment; align the pollutant concentration data, transportation path and emission data and pollution indicators with the regional environmental indicators obtained by satellite remote sensing in time and space to obtain multi-source data; and aggregate the multi-source data through federated learning technology to obtain pollution data.

[0106] The present application provides a supply chain environmental risk warning device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the supply chain environmental risk warning method in the above-mentioned embodiment one.

[0107] Reference below Figure 7, which shows a schematic diagram of the structure of a supply chain environmental risk warning device suitable for implementing embodiments of the present application. The supply chain environmental risk warning device in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 7 The supply chain environmental risk warning device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0108] like Figure 7 As shown, the supply chain environmental risk early warning device may include a processing device 1001 (e.g., a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 1002 or programs loaded from storage device 1003 into random access memory (RAM) 1004. RAM 1004 also stores various programs and data required for the operation of the supply chain environmental risk early warning device. Processing device 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following can be connected to I / O interface 1006: input devices 1007, such as a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008, such as a liquid crystal display (LCD), speaker, vibrator, etc.; storage device 1003, such as a magnetic tape or hard disk; and communication device 1009. Communication device 1009 allows the supply chain environmental risk warning device to communicate wirelessly or wired with other devices to exchange data. Although various supply chain environmental risk warning devices are shown in the figure, it should be understood that not all of the shown devices are required to be implemented or present. More or fewer devices may be implemented or present instead.

[0109] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0110] The supply chain environmental risk warning device provided in this application, employing the supply chain environmental risk warning method described in the aforementioned embodiment, can address the technical problem of improving the efficiency of supply chain risk warnings. Compared to the prior art, the beneficial effects of the supply chain environmental risk warning device provided in this application are the same as those of the supply chain environmental risk warning method described in the aforementioned embodiment. Other technical features of this supply chain environmental risk warning device are the same as those disclosed in the aforementioned embodiment and are not further elaborated here.

[0111] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0112] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0113] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, and the computer-readable program instructions are used to execute the supply chain environmental risk early warning method in the above-mentioned embodiment.

[0114] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible storage medium that contains or stores a program that can be executed by instructions, used by a device, or used in conjunction with it. The program code contained on the computer-readable storage medium may be transmitted using any suitable storage medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0115] The above-mentioned computer-readable storage medium may be included in the supply chain environmental risk early warning device; or it may exist independently without being assembled into the supply chain environmental risk early warning device.

[0116] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the supply chain environmental risk early warning device, the supply chain environmental risk early warning device can write computer program code for performing the operations of the present application in one or more programming languages or a combination thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, and C++, as well as conventional procedural programming languages such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, via the Internet using an Internet service provider).

[0117] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based implementation that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0118] The modules described in the embodiments of the present application may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.

[0119] The computer-readable storage medium provided in this application stores computer-readable program instructions (i.e., a computer program) for executing the aforementioned supply chain environmental risk early warning method. This computer-readable storage medium addresses the technical problem of improving the efficiency of supply chain risk early warning. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are similar to those of the supply chain environmental risk early warning method provided in the aforementioned embodiments, and are not further elaborated here.

[0120] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned supply chain environmental risk early warning method when executed by a processor.

[0121] The computer program product provided in this application can solve the technical problem of how to improve the efficiency of supply chain risk early warning. Compared with the existing technology, the beneficial effects of the computer program product provided in this application are the same as those of the supply chain environmental risk early warning method provided in the above embodiment, and will not be elaborated here.

[0122] The above description is only part of the embodiments of the present application and does not limit the patent scope of the present application. All equivalent structural transformations made by using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A supply chain environmental risk early warning method, characterized in that: The method comprises: Obtaining pollution data, including pollution data from production, transportation, and storage; performing standardization processing on the pollution data to obtain processed pollution data; The processed pollution data is hashed and stored on the chain to obtain a trusted supply chain network; Convert each node and corresponding relationship in the trusted supply chain network into a graph structure, and construct a pollution contribution assessment model based on a graph neural network; Inputting the processed pollution data into the pollution contribution assessment model for calculation to obtain the pollution contribution; Calculating the pollution contribution through a preset time series prediction model, and outputting a dynamic adaptive threshold and a pollution residual, wherein the preset time series prediction model includes a long short-term memory network and a Transformer encoder; Executing a corresponding warning operation according to a comparison result of the pollution residual and the dynamic adaptive threshold; The step of performing a corresponding warning operation according to a comparison result between the pollution residual and the dynamic adaptive threshold comprises: Comparing the node pollution residual with the dynamic adaptive threshold to obtain the warning level; Calculate the normalized residual and spatiotemporal weight of each node according to the specific parameters of the warning level; Calculating based on the normalized residual amount and the spatiotemporal weight to generate an overall supply chain risk value; According to the overall risk value, a multi-level warning and response mechanism is matched to execute corresponding warning operations.

2. The method according to claim 1, wherein The step of hashing and encrypting the processed contamination data and storing it on the chain to obtain a trusted supply chain network includes: Get the blockchain network; The processed contaminated data is distributed and calculated through IPFS to obtain a hash summary. The specific formula is: in Using the SHA-256 algorithm, Represents a hash digest, which is a unique content identifier; Obtaining a node identity public key based on the hash summary and the blockchain network; The node identity is obtained by calculation based on the node identity public key. The specific formula is: in, Representation node The public key of represents a hash function; The node identity and the hash summary are aggregated in combination with the timestamp and stored on the blockchain network to obtain a trusted supply chain network.

3. The method according to claim 1, wherein The steps of converting the nodes and corresponding relationships in the trusted supply chain network into a graph structure and constructing a pollution contribution assessment model based on a graph neural network include: The supply chain network is transformed and modeled as a directed attribute graph, where nodes represent enterprises and edges represent logistics relationships; Based on the directed attribute graph, a node attribute matrix and an edge attribute matrix are defined, wherein the node attribute matrix includes pollution data, emission frequency, operation level, business scale, industry category, and historical compliance rate, and the edge attribute matrix includes logistics relationship type, supply frequency, cooperation intensity, and physical distance; Initialization is performed based on the node attribute matrix and the edge attribute matrix to obtain the initial node features and the corresponding neighbor node features. The formula for the initial node features is: in, represents the node feature map weight matrix, is a bias term, and the formula for the corresponding neighbor node feature is: in, represents the adjacency matrix; The attention weights between nodes are calculated based on the initial node features and the neighbor node features through the multi-channel graph attention mechanism. The specific formula is: in, Indicates the number of layers of message passing, For all neighbor nodes The normalization function of is a nonlinear activation function, represents the weight vector of the attention mechanism, For nodes of The set of neighbor nodes of order , Representation node exist In the link order features, Represents neighbor nodes exist In the link order features, Represents the edge attribute The mapping transformation, Represents vector concatenation operation; Based on the attention weight, the initial node features and the corresponding neighbor node features are iteratively updated to obtain the feature representation of each node. The specific formula is: in, for In the link layer weight matrices, Indicates the total number of layers of message passing, Representation node The degree, Representation node The degree, Representation node of The set of neighbor nodes of order , is the adjacency matrix, Represents neighbor nodes exist In the link Layer characteristics; Calculate the total pollution contribution of the node by using the characteristic representation; Constructing a pollution contribution assessment model based on the total pollution contribution; The step of calculating the total pollution contribution of the node by using the characteristic representation includes: representing the direct pollution contribution, conducted pollution contribution, and historical violation correction term of the computing node according to the characteristics; The direct pollution contribution, the conducted pollution contribution and the historical violation correction item are weighted and summed using a preset weight coefficient to obtain the total pollution contribution of the node.

4. The method according to claim 1, wherein The step of calculating the pollution contribution through a preset time series prediction model and outputting a dynamic adaptive threshold and a pollution residual includes: Obtain historical pollution contribution data and construct a time series sample set; Based on the time series sample set, a long short-term memory network of a preset time series prediction model is used to capture short-term dependency features and generate a hidden state sequence; Input the hidden state sequence into the Transformer encoder of a preset time series prediction model, and extract long-term trend features through a multi-head attention mechanism; Performing a prediction based on the long-term trend characteristics to obtain a predicted pollution contribution; Calculating based on the predicted pollution contribution and the pollution contribution to obtain a pollution residual; Based on the pollution residual and the node collaborative weight matrix, the warning threshold multiple is dynamically adjusted to generate a dynamic adaptive threshold.

5. The method according to claim 4, wherein The step of dynamically adjusting the warning threshold multiple based on the prediction residual in combination with the node collaborative weight matrix to generate a dynamic adaptive threshold includes: Obtain the logistics relationship between nodes; Performing exponentially weighted moving average on the contamination residuals to obtain a mean and a variance; Constructing a collaborative weight matrix based on the logistics relationship between the nodes; Based on the collaborative weight matrix, the basic threshold multiple is modified to generate an early warning threshold, where the basic threshold multiple is a preset constant value used to adjust the size of the early warning threshold; A dynamic adaptive threshold is generated according to the warning threshold in combination with the mean and variance.

6. The method according to claim 1, wherein The step of performing corresponding warning operations according to the matching multi-level warning and response mechanism of the overall risk value includes: When the overall risk value exceeds the preset first-level risk threshold and the overall risk value does not exceed the preset second-level risk threshold, a first-level warning is triggered and an internal risk warning is issued; When the overall risk value exceeds the preset second-level risk threshold and the overall risk value does not exceed the preset third-level risk threshold, a second-level warning is triggered, cross-node linkage is initiated, and existing production, transportation, and warehousing processes are optimized; When the overall risk value exceeds the preset third-level risk threshold and the overall risk value does not exceed the preset fourth-level risk threshold, a third-level warning is triggered to adjust the supply chain resource allocation and reduce the business volume with high-risk nodes; When the overall risk value exceeds the preset level 4 risk threshold, a level 4 warning is triggered, the supply chain reconstruction procedure is executed and business dealings with high-risk nodes are suspended.

7. The method according to claim 1, wherein The step of obtaining pollution data includes: Obtain pollutant concentration data by deploying sensors; Obtain transport route and emission data through the positioning device of the transport vehicle; Obtain pollution indicators in the storage process through storage environment monitoring equipment; Performing spatiotemporal alignment of the pollutant concentration data, the transport route and emission data, and the pollution index with regional environmental indicators acquired through satellite remote sensing to obtain multi-source data; The multi-source data is aggregated through federated learning technology to obtain contaminated data.

8. A supply chain environmental risk early warning device, characterized in that: The device comprises: An acquisition module is used to acquire pollution data, including pollution data from production, transportation, and storage; A processing module, configured to perform standardization processing on the pollution data to obtain processed pollution data; An encryption module, used to hash and encrypt the processed contamination data and store it on the chain to obtain a trusted supply chain network; A construction module is used to convert the nodes and corresponding relationships in the trusted supply chain network into a graph structure and build a pollution contribution assessment model based on a graph neural network; a calculation module, configured to input the processed pollution data into the pollution contribution assessment model for calculation to obtain a pollution contribution; A prediction module, configured to calculate the pollution contribution through a preset time series prediction model and output a dynamic adaptive threshold and a pollution residual, wherein the preset time series prediction model includes a long short-term memory network and a Transformer encoder; An execution module is used to perform corresponding warning operations based on the comparison results of the pollution residual and the dynamic adaptive threshold; it is also used to compare the node pollution residual with the dynamic adaptive threshold to obtain the warning level; calculate the residual normalization amount and spatiotemporal weight of each node according to the specific parameters corresponding to the warning level; calculate based on the residual normalization amount and the spatiotemporal weight to generate the overall risk value of the supply chain; match the multi-level warning and response mechanism according to the overall risk value to perform corresponding warning operations.

9. A supply chain environmental risk early warning device, characterized in that: The device includes: a memory, a processor, and a supply chain environmental risk early warning program stored in the memory and running on the processor, wherein the supply chain environmental risk early warning program is configured to implement the steps of the supply chain environmental risk early warning method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores a supply chain environmental risk early warning program, which, when executed by the processor, implements the steps of the supply chain environmental risk early warning method according to any one of claims 1 to 7.

Images