Supply chain elastic optimization decision-making system based on big data

Abstract

The invention relates to a supply chain elastic optimization decision-making system based on big data. The system comprises a data acquisition module used for performing semantic alignment processing on multi-source heterogeneous environmental protection data; the ESG risk assessment module is used for constructing a cross-enterprise ESG risk assessment model and outputting a supplier environment toughness score; the optimization decision generation module is used for analyzing an environmental protection policy text, generating a compliance verification rule, and generating a supply chain path scheme set by constructing and solving a dual-objective optimization function; and the dynamic tuning module is used for monitoring supply chain execution data in real time, dynamically triggering a dynamic rerouting decision and updating the cross-enterprise ESG risk assessment model. The system is based on the technical means of data semantic alignment, rule embedding optimization, real-time dynamic tuning and the like, not only improves the decision robustness of the supply chain in a dynamic environment, but also improves the response speed and recovery capability of the supply chain in response to environmental changes and emergencies; and a powerful technical scheme is provided for an enterprise to construct a sustainable and high-elasticity supply chain system.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a supply chain elasticity optimization decision system based on big data. Background Art

[0002] With the intensification of global climate change and the advancement of the dual carbon goals, the environmental risks and compliance pressures facing supply chains have increased significantly. Therefore, building a resilient supply chain system that adapts to the environment has become key for companies to break through development bottlenecks and achieve sustainable competitiveness. However, traditional supply chain management models have many shortcomings in areas such as risk assessment and decision optimization. For example, in terms of risk assessment, traditional ESG assessment models constructed independently by individual companies lack cross-enterprise data collaboration mechanisms and are unable to incorporate dynamic indicators such as suppliers' carbon emission sensitivity and environmental risk responsiveness into the assessment system, making it difficult to accurately quantify the impact of environmental fluctuations on various nodes in the supply chain. Furthermore, in the decision optimization phase, existing route planning and transportation mode selection models often focus on a single cost or efficiency objective, ignoring the dynamic changes in carbon emission constraints. This makes it difficult to quickly adjust to real-time environmental policies and carbon emission requirements, resulting in a lack of flexibility and adaptability in the supply chain when responding to changes in environmental policies. Summary of the Invention

[0003] Based on this, it is necessary to provide a supply chain resilience optimization decision-making system based on big data to address the above technical issues, so as to enhance the dynamic adaptability of the supply chain to environmental risks and policy changes, and improve the robustness of decision-making in a dynamic environment.

[0004] First, this application provides a supply chain resilience optimization decision-making system based on big data, including:

[0005] The data acquisition module is used to collect multi-source heterogeneous environmental data, perform semantic alignment on the multi-source heterogeneous environmental data, and generate a standardized environmental data matrix. The multi-source heterogeneous environmental data includes logistics vehicle energy consumption data, material carbon footprint data, and resource recycling rate data;

[0006] The ESG risk assessment module is used to build a cross-enterprise ESG risk assessment model based on standardized environmental data based on a federated learning framework, and output a supplier environmental resilience score that includes carbon emission sensitivity, environmental risk responsiveness, and dynamic compliance adaptation weights;

[0007] The optimization decision generation module is used to obtain and parse environmental policy texts, obtain parsed results, and compile the parsed results into compliance verification rules. The parsed results include geographical constraints, a list of prohibited materials, and effective time thresholds. Based on standardized environmental data, supplier environmental resilience scores, and compliance verification rules, a dual-objective optimization function is constructed and solved to generate a set of supply chain path solutions. The supply chain path solution set includes path selection options, transportation mode combinations, and node redundancy configurations.

[0008] The dynamic tuning module is used to monitor supply chain execution data in real time, trigger dynamic rerouting decisions based on supply chain execution data, and update cross-enterprise ESG risk assessment models. Supply chain execution data includes the real-time location of transportation vehicles, energy consumption monitoring values, and environmental parameter fluctuation data.

[0009] In one embodiment, the data acquisition module includes:

[0010] The raw data acquisition subunit is used to:

[0011] Obtain real-time energy consumption data of logistics vehicles;

[0012] Obtain carbon footprint data for bills of materials based on the enterprise ERP system. This data is calculated based on a life cycle assessment model. Also, obtain resource recycling rate data based on the enterprise production system. This resource recycling rate data includes material recovery rate, waste reuse rate, and the proportion of recycled materials.

[0013] Combining logistics vehicle energy consumption data, material carbon footprint data, and resource recycling rate data to obtain multi-source heterogeneous environmental protection data;

[0014] Quantity processing subunit, used for:

[0015] Build an environmental knowledge graph and define core ontologies for carbon emissions, resource recycling rates, and geographic fencing. The core ontology uses the OWL language to describe the hierarchical relationships between entities.

[0016] Based on the core ontology, a semantic mapping relationship across data sources is established. The semantic conflicts of multi-source heterogeneous environmental data are detected through the semantic mapping relationship. Semantic conflicts include measurement unit conflicts, time granularity deviations, and geocoding format differences, and conflict data is obtained.

[0017] Process the conflicting data according to the semantic mapping relationship to obtain pre-processed multi-source data;

[0018] The preprocessed multi-source data are temporally and spatially aligned to generate a standardized environmental data matrix. The dimensions of the standardized environmental data matrix include timestamp, geographic location and environmental indicators.

[0019] In one embodiment, the ESG risk assessment module includes:

[0020] The feature extraction subunit is used to deploy a convolutional sparse autoencoder with a mask mechanism locally in each enterprise. The standardized environmental data is input into the convolutional sparse autoencoder for feature extraction and generates an initial feature vector. A Gaussian noise mask is added to the input layer of the convolutional sparse autoencoder, and the activation of the hidden layer of the convolutional sparse autoencoder is constrained through L1 regularization.

[0021] The model building subunit is used to perform Paillier homomorphic encryption on the initial feature vector to generate an encrypted feature vector, upload the encrypted feature vector to the central server, aggregate the encrypted feature vectors of each enterprise using the adaptive weighted federated averaging algorithm, and build a cross-enterprise ESG risk assessment model;

[0022] The environmental assessment sub-unit is used to calculate the supplier's environmental resilience score based on the cross-enterprise ESG risk assessment model.

[0023] In one embodiment, the optimization decision generation module includes an environmental policy analysis subunit, which is used to:

[0024] Obtain the environmental protection policy text, which includes text clauses, appendix forms and revision history;

[0025] The BERT-GNN hybrid model is used to parse environmental policy texts and obtain parsed results. The geographical constraints include the administrative divisions, ecological protection zone coordinate ranges, and transport restricted area boundaries defined in the environmental policy text. The banned materials list includes the CAS numbers of banned raw material chemicals and the types of non-degradable packaging materials. The effective time threshold includes the effective date, expiration date, and temporary ban time window of the policy clauses.

[0026] Based on the entities and relationships in the parsing results, dynamic knowledge graph completion technology is used to convert the parsing results into triple rules. Then, through the hierarchical rule induction algorithm, a multi-level decision tree structure is constructed according to the triple rules to obtain the decision tree logical expression;

[0027] Through the automated compiler, the decision tree logical expressions are converted into Drools rule engine scripts to generate compliance verification rules.

[0028] In one embodiment, the optimization decision generation module includes a multi-objective optimization subunit, which is configured to:

[0029] A carbon emission function is constructed based on the energy consumption of logistics vehicles and the carbon footprint of materials in standardized environmental data, combined with the carbon emission sensitivity in the supplier environmental resilience score;

[0030] Timeliness scores are calculated based on logistics vehicle energy consumption data and the historical average transportation times of different transportation modes. An efficiency function is constructed based on resource circulation rate data, combined with the deployment cost and activation threshold of standby nodes. Standby nodes are pre-set redundant nodes in the supply chain, used to replace failed nodes in the event of sudden disruptions. The deployment cost is calculated based on resource circulation rate data, and the activation threshold is dynamically adjusted based on the responsiveness to environmental risks.

[0031] Combining the efficiency function and the carbon emission function, a dual-objective optimization function is obtained;

[0032] Based on a dual-objective optimization function, the compliance verification rules and preset resource constraints are used as constraints, and an improved NSGA-III algorithm is used to generate a Pareto frontier solution set. The improved NSGA-III algorithm includes calculating the weight ratio of the efficiency function and the carbon emission function based on the dynamic compliance adaptation weight, and then allocating the reference point density in the objective space of the solution process according to the weight ratio. The next generation population is selected using the niching strategy based on the reference point density.

[0033] Extracting the decision variable vector of each solution from the Pareto front solution set, the decision variable vector includes path selection variables, transportation mode combination variables and node redundancy configuration variables;

[0034] Based on the coordinate range of the ecological protection zone based on the compliance verification rules, the path selection variables are geographically constrained and labeled to obtain path labels. The carbon emission function values ​​corresponding to the decision variable vectors are combined with the preset carbon emission threshold intervals to classify the paths and obtain carbon emission grade labels. Based on historical transportation data, the interruption probability of the corresponding paths in the decision variable vectors is statistically analyzed to generate stability labels.

[0035] Combining carbon emission level labels, path labels, and stability labels, a path selection scheme is obtained;

[0036] The transport mode combination variables are verified for vehicle compliance using compliance verification rules to obtain packaging compliance labels. The timeliness score is associated with the corresponding transport mode, and the energy consumption level of the transport mode is annotated in combination with the logistics vehicle energy consumption data to obtain the transport mode combination.

[0037] According to the standby node deployment cost and activation threshold in the node redundancy configuration variable, the redundancy level is divided to obtain the node redundancy configuration;

[0038] Combining the path selection scheme, transportation mode combination and node redundancy configuration, a supply chain path solution set is obtained.

[0039] In one embodiment, the dynamic tuning module includes:

[0040] Anomaly detection subunit, which is used to obtain supply chain execution data and set dynamic thresholds based on compliance verification rules to detect whether there are abnormal data in the supply chain execution data according to the dynamic thresholds, where the dynamic thresholds include vehicle energy consumption thresholds and transportation path deviation thresholds;

[0041] Exception handling subunit, used to:

[0042] When abnormal data is detected, the preset conditions are selected from the supply chain path solution set to generate a global rerouting candidate set;

[0043] Based on the environmental parameter fluctuation data, the path selection scheme in the global rerouting candidate set is optimized to obtain the optimized path selection scheme and the decision result;

[0044] The decision results are associated with the corresponding supply chain execution data to generate a training set. Based on the training set, the cross-enterprise ESG risk assessment model is updated through incremental learning to obtain an updated cross-enterprise ESG risk assessment model.

[0045] In one embodiment, the dynamic compliance adaptation weight is calculated using the following formula:

[0046]

[0047] Among them, W t is the dynamic compliance adaptation weight at time t, Environmental risk responsiveness is the matching degree of the policy effective time threshold, which is calculated based on the matching degree between the effective time threshold and the current time t. is the deviation distance from the current transport path to the coordinate range of the ecological protection zone at time t, and α, β, and γ are all adaptive coefficients.

[0048] Secondly, this application also provides a supply chain resilience optimization decision-making method based on big data, including:

[0049] Collect multi-source heterogeneous environmental data, perform semantic alignment on the data, and generate a standardized environmental data matrix. The multi-source heterogeneous environmental data includes logistics vehicle energy consumption data, material carbon footprint data, and resource recycling rate data.

[0050] Based on the federated learning framework, a cross-enterprise ESG risk assessment model is constructed based on standardized environmental data, outputting a supplier environmental resilience score that includes carbon emission sensitivity, environmental risk responsiveness, and dynamic compliance adaptation weights.

[0051] Obtain and parse environmental policy text, obtain parsed results, and compile these results into compliance verification rules. The parsed results include geographic constraints, a list of prohibited materials, and effective time thresholds. Based on standardized environmental data, supplier environmental resilience scores, and compliance verification rules, a dual-objective optimization function is constructed and solved to generate a set of supply chain routing solutions. The supply chain routing solution set includes routing options, transportation mode combinations, and node redundancy configurations.

[0052] Monitor supply chain execution data in real time, trigger dynamic rerouting decisions based on supply chain execution data, and update cross-enterprise ESG risk assessment models. Supply chain execution data includes the real-time location of transportation vehicles, energy consumption monitoring values, and environmental parameter fluctuation data.

[0053] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program and the processor implements the steps in the first aspect when executing the program.

[0054] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps in the first aspect when executed by a processor.

[0055] The aforementioned big data-based supply chain resilience optimization decision-making system, through the coordinated operation of various modules, achieves intelligent management of the entire supply chain process, from data collection, risk assessment, decision optimization, and dynamic adjustment. The data acquisition module semantically aligns multi-source data, providing an accurate and unified data foundation for subsequent modules and avoiding decision biases caused by data conflicts. The ESG risk assessment module builds a cross-enterprise ESG risk assessment model based on a federated learning framework. This not only enables cross-enterprise data collaboration but also outputs supplier environmental resilience scores, comprehensively and dynamically quantifying the impact of environmental fluctuations on supply chain nodes. The optimization decision generation module performs multi-objective optimization based on environmental policy texts, balancing carbon emissions and efficiency. The dynamic tuning module monitors supply chain execution data such as transport vehicle real-time location and energy consumption in real time. When anomalies occur, dynamic rerouting decisions are triggered and the cross-enterprise ESG risk assessment model is simultaneously updated. This enhances the supply chain's ability to respond quickly to emergencies such as extreme weather and policy changes, and improves the timeliness and accuracy of ESG risk assessments.

[0056] Compared with traditional supply chain management systems, this system is based on technical means such as data semantic alignment, federated learning collaboration, rule embedding optimization and real-time dynamic tuning. It not only ensures the reliability of decision-making and optimizes resource allocation efficiency, but also further improves the supply chain's response speed and recovery capabilities to environmental changes and emergencies, effectively reduces operational risks, and provides enterprises with a powerful technical solution for building a sustainable and highly flexible supply chain system. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0058] Figure 1 A schematic diagram of the structure of a supply chain elasticity optimization decision system based on big data provided by an exemplary embodiment of the present invention;

[0059] Figure 2 A flow chart of a supply chain resilience optimization decision-making method based on big data provided as an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0061] In one embodiment, Figure 1 As shown, a supply chain elasticity optimization decision system 100 based on big data is provided. This embodiment uses the system applied to a terminal as an example. It is understandable that the system can also be applied to a server, or to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the system includes:

[0062] The data acquisition module 101 is used to collect multi-source heterogeneous environmental data, perform semantic alignment on the multi-source heterogeneous environmental data, and generate a standardized environmental data matrix. The multi-source heterogeneous environmental data includes logistics vehicle energy consumption data, material carbon footprint data, and resource recycling rate data.

[0063] Specifically, the data acquisition module 101 can acquire multi-source heterogeneous environmental data from the enterprise's internal production management system, logistics and transportation system, and other sources. Logistics vehicle energy consumption data can include data on fuel and electricity consumption during transportation by different types of transportation vehicles, such as trucks, ships, and aircraft. Material carbon footprint data can record greenhouse gas emissions generated throughout the entire life cycle, from raw material extraction and processing to product transportation. Resource recycling rate data reflects the proportion of resources reused and recycled in the supply chain, reflecting the sustainability of resource utilization.

[0064] Because data from different sources often exists in different formats, structures, and semantics, this module uses natural language processing technology and machine learning algorithms to perform semantic parsing and mapping on the raw data, converting it into a standardized environmental data matrix with unified semantics and format. This process not only ensures data consistency and accuracy but also provides a high-quality data foundation for subsequent analysis and evaluation.

[0065] The ESG risk assessment module 102 is used to build a cross-enterprise ESG risk assessment model based on a federated learning framework and standardized environmental data, and output a supplier environmental resilience score. The supplier environmental resilience score includes carbon emission sensitivity, environmental risk responsiveness and dynamic compliance adaptation weights.

[0066] Specifically, federated learning is a distributed machine learning technology that enables collaborative data modeling across multiple enterprises without sharing original data. In this module, through the federated learning framework, different enterprises in the supply chain can participate in the training of a cross-enterprise ESG risk assessment model while retaining their own data locally. For example, a cross-enterprise ESG risk assessment model can be constructed using deep learning or machine learning algorithms based on standardized environmental data. This model learns and analyzes standardized environmental data to output a supplier environmental resilience score. Carbon emission sensitivity measures the impact of a supplier's production and operation activities on carbon emissions, environmental risk responsiveness reflects a supplier's ability and speed to respond to environmental risks such as natural disasters and policy changes, and dynamic compliance adaptation weighting reflects a supplier's adaptability and compliance with evolving environmental regulations and policies. This comprehensive scoring of these three dimensions enables a comprehensive and objective assessment of a supplier's environmental resilience, providing key insights for supply chain risk management and helping enterprise managers identify potential environmental risks and formulate appropriate risk response strategies.

[0067] The optimization decision generation module 103 is used to obtain and parse the environmental protection policy text, obtain the parsing results, and compile the parsing results into compliance verification rules. The parsing results include geographical constraints, a list of banned materials, and an effective time threshold. Based on standardized environmental protection data, supplier environmental resilience scores, and compliance verification rules, a dual-objective optimization function is constructed and solved to generate a supply chain path solution set. The supply chain path solution set includes path selection solutions, transportation mode combinations, and node redundancy configurations.

[0068] Specifically, natural language processing techniques can be used to deeply analyze the acquired environmental policy text, extracting key information such as geographic constraints restricting transportation or production activities in certain areas, a banned materials list that explicitly prohibits the use of materials, and policy effective time thresholds, thereby generating analytical results. These analytical results can then be compiled into compliance verification rules, which can be applied as constraints in subsequent optimization decision-making processes. Furthermore, standardized environmental data, supplier environmental resilience scores, and compliance verification rules can be combined to construct a dual-objective optimization function. This function can be designed to minimize supply chain costs such as transportation and inventory costs, while maximizing the environmental sustainability of the supply chain by reducing carbon emissions and improving resource utilization. Based on this function, algorithms such as genetic algorithms and simulated annealing can be used to solve the problem, generating a set of feasible supply chain routing solutions. These solutions can include multiple aspects such as routing selection, transportation mode combinations, and node redundancy configuration. Among them, the path selection solution provides the optimal transportation route while meeting environmental protection and cost constraints. The transportation mode combination solution outputs the most suitable transportation mode combination according to different transportation needs and environmental requirements. The node redundancy configuration solution can improve the flexibility and reliability of the supply chain in the face of emergencies by reasonably configuring redundant nodes in the supply chain.

[0069] The dynamic tuning module 104 is used to monitor supply chain execution data in real time, trigger dynamic rerouting decisions based on the supply chain execution data, and update the cross-enterprise ESG risk assessment model, where the supply chain execution data includes the real-time location of transportation vehicles, energy consumption monitoring values, and environmental parameter fluctuation data.

[0070] Specifically, this module can monitor the operating status of the supply chain in real time by monitoring the real-time position of transportation vehicles in the supply chain, energy consumption monitoring values ​​and environmental parameter fluctuation data.

[0071] When supply chain execution data detects anomalies or meets preset trigger conditions, the module can trigger dynamic rerouting decisions. For example, if a transport vehicle is unable to travel along its planned route due to road congestion, natural disasters, or other factors, the system can reroute the shipment based on real-time traffic information and other available routes to ensure timely delivery. Furthermore, the module can update cross-enterprise ESG risk assessment models based on the latest supply chain execution data, enabling the model to dynamically reflect changes in environmental risks across the supply chain, thereby providing real-time, accurate risk assessment information for continuous supply chain optimization.

[0072] The above system mainly includes four modules. The data acquisition module 101 can effectively integrate data from different sources by semantically aligning multi-source heterogeneous environmental data, providing a unified and accurate data foundation for subsequent modules. The ESG risk assessment module 102 builds a cross-enterprise ESG risk assessment model based on the federated learning framework, which can achieve cross-enterprise data collaboration and output supplier environmental resilience scores that include carbon emission sensitivity, environmental risk responsiveness, and dynamic compliance adaptation weights to dynamically evaluate the environmental risks of supply chain nodes. The optimization decision generation module 103 generates compliance verification rules by parsing environmental policy texts, taking into account multiple factors such as geographical constraints, banned material lists, and effective time thresholds, and combines standardized environmental data and supplier environmental resilience scores for multi-objective optimization to generate a supply chain path solution set, which can better balance economic and environmental benefits and improve the scientificity and sustainability of supply chain decision optimization. The dynamic tuning module 104 monitors supply chain execution data such as the real-time location of transport vehicles and energy consumption monitoring values ​​in real time. When an anomaly occurs, it triggers dynamic rerouting decisions in a timely manner and simultaneously updates the cross-enterprise ESG risk assessment model, enabling the supply chain to adapt to environmental changes and policy adjustments in a timely manner, effectively improving the elasticity and adaptability of the supply chain and ensuring its stable operation in a dynamic environment.

[0073] In one embodiment, the data acquisition module 101 includes:

[0074] The raw data acquisition sub-unit is used to: obtain real-time energy consumption data of logistics vehicles; obtain carbon footprint data of the bill of materials based on the enterprise ERP system, which is calculated and generated based on the life cycle assessment model; and obtain resource recycling rate data based on the enterprise production system. Resource recycling rate data includes material recovery rate, waste reuse rate, and recycled material ratio; combine logistics vehicle energy consumption data, material carbon footprint data, and resource recycling rate data to obtain multi-source heterogeneous environmental protection data;

[0075] The quantity processing subunit is used to: construct an environmental protection knowledge graph, define the core ontology of carbon emissions, resource circulation rate and geographic fence in the environmental protection knowledge graph, and describe the hierarchical relationship between entities in the core ontology using the OWL language; establish a semantic mapping relationship across data sources based on the core ontology, and detect semantic conflicts in multi-source heterogeneous environmental protection data through the semantic mapping relationship. Semantic conflicts include measurement unit conflicts, time granularity deviations and geographic coding format differences to obtain conflicting data; process the conflicting data according to the semantic mapping relationship to obtain preprocessed multi-source data; perform spatiotemporal alignment on the preprocessed multi-source data to generate a standardized environmental protection data matrix. The dimensions of the standardized environmental protection data matrix include timestamp, geographic location and environmental protection indicators.

[0076] Specifically, IoT sensors can be deployed in the engine compartment of transport vehicles, or fuel consumption data such as vehicle electronic control unit records, gas station refueling records, and driver manual records can be read, and data processing and integration can be performed to obtain logistics vehicle energy consumption data. Bill of materials carbon footprint data can be extracted from the enterprise resource planning system, and based on the life cycle assessment model, the carbon emissions of the material's entire life cycle, from raw material mining, production and processing, transportation and distribution to waste disposal, can be quantified and calculated. Resource recycling rate data can be obtained through the enterprise production management system. Among them, the material recovery rate can be calculated by counting the recycling ratio of recyclable materials such as metal scrap and packaging materials in the production process. The waste reuse rate can be calculated by calculating the proportion of waste that is re-entered into production after treatment. The proportion of recycled materials can be obtained by quantifying the proportion of recycled materials, such as recycled plastics, used in production raw materials.

[0077] Furthermore, an environmental knowledge graph can be constructed using OWL to define core ontologies, primarily carbon emissions, resource recycling rates, and geofencing, and their hierarchical relationships. For example, a logistics vehicle can be defined as a subclass of the carbon emissions ontology, with attributes linking data such as energy consumption or carbon footprint. Spatial relationships between the geofencing ontology, transportation routes, and ecological protection zones can also be established. Furthermore, this knowledge graph can store entity relationships using a subject-verb-object triple structure. Subsequently, semantic mapping can be performed between the multi-source, heterogeneous environmental data and the knowledge graph to automatically detect semantic conflicts. For example, predefined unit conversion operators within the knowledge graph can be used to automatically convert unit differences for the same indicator across different data sources to a unified standard. Data with inconsistent temporal granularity can be aligned using linear interpolation algorithms, and coordinate conversion algorithms can be used to unify different geocoding formats to ensure spatial data consistency. The preprocessed multi-source data can be structured and reorganized according to the three dimensions of timestamp, geolocation, and environmental indicators to generate a standardized environmental data matrix. Environmental indicators can include core parameters such as energy consumption, carbon footprint, and resource recycling rates.

[0078] In one embodiment, the ESG risk assessment module 102 includes:

[0079] The feature extraction subunit is used to deploy a convolutional sparse autoencoder with a mask mechanism locally in each enterprise. The standardized environmental data is input into the convolutional sparse autoencoder for feature extraction and generates an initial feature vector. A Gaussian noise mask is added to the input layer of the convolutional sparse autoencoder, and the activation of the hidden layer of the convolutional sparse autoencoder is constrained through L1 regularization.

[0080] The model building subunit is used to perform Paillier homomorphic encryption on the initial feature vector to generate an encrypted feature vector, upload the encrypted feature vector to the central server, aggregate the encrypted feature vectors of each enterprise using the adaptive weighted federated averaging algorithm, and build a cross-enterprise ESG risk assessment model;

[0081] The environmental assessment sub-unit is used to calculate the supplier's environmental resilience score based on the cross-enterprise ESG risk assessment model.

[0082] Specifically, a convolutional sparse autoencoder model can be deployed on the local servers or data processing nodes of each supply chain participant. This model consists of two parts: an encoder and a decoder. The encoder compresses the input standardized environmental data matrix into a low-dimensional feature vector, while the decoder reconstructs the original data from the feature vector, learning data features by minimizing reconstruction error. When the standardized environmental data matrix is ​​input into the model, Gaussian noise with a mean of 0 and a standard deviation of 0.1 can be superimposed to simulate the interference that may occur during data acquisition and transmission, forcing the model to learn more stable and reliable feature representations. Furthermore, during model training, the weights of hidden layer neurons based on an L1 regularization term can be introduced into the loss function, enabling the model to learn a sparse feature representation, activating only a small number of key neurons. This effectively extracts core features related to supplier ESG risks and avoids model overfitting. Finally, the output is a low-dimensional initial feature vector containing abstract features related to ESG indicators such as carbon emissions and resource recycling in the standardized environmental data. The vector can then be encrypted using the Paillier homomorphic encryption algorithm through addition and multiplication operations on the ciphertext, so that the encrypted feature vector can be directly operated without decryption when performing aggregation calculations on the central server, thereby ensuring the privacy and security of corporate data. Furthermore, the central server can use an adaptive weighted federated averaging algorithm to aggregate the encrypted feature vectors of each enterprise. The aggregation weight can be dynamically adjusted based on the environmental risk responsiveness calculated based on indicators such as the enterprise's historical rerouting response speed and redundant node activation efficiency. Through multiple iterative aggregation operations, the central server can gradually update and construct a cross-enterprise ESG risk assessment model. This model integrates the environmental data characteristics of each enterprise in the supply chain and can assess the ESG risk level of suppliers from a global perspective.

[0083] Illustratively, the model outputs intermediate results related to supplier ESG risk through forward propagation calculations. Based on these intermediate results, combined with pre-set scoring rules and algorithms, a supplier environmental resilience score can be calculated. Carbon emission sensitivity quantifies a supplier's responsiveness to carbon emission fluctuations, while environmental risk responsiveness assesses a supplier's risk response capabilities based on its historical performance in responding to environmental emergencies, including metrics such as rerouting decision speed and backup node activation efficiency. Dynamic compliance adaptation weighting dynamically adjusts a supplier's weighting between environmental and economic objectives based on the latest environmental policy requirements and market changes, reflecting its ability to adapt to policy changes.

[0084] In one embodiment, the optimization decision generation module 103 includes an environmental policy analysis subunit, which is used to:

[0085] Obtain the environmental protection policy text, which includes text clauses, appendix forms and revision history;

[0086] The BERT-GNN hybrid model is used to parse environmental policy texts and obtain parsed results. The geographical constraints include the administrative divisions, ecological protection zone coordinate ranges, and transport restricted area boundaries defined in the environmental policy text. The banned materials list includes the CAS numbers of banned raw material chemicals and the types of non-degradable packaging materials. The effective time threshold includes the effective date, expiration date, and temporary ban time window of the policy clauses.

[0087] Based on the entities and relationships in the parsing results, dynamic knowledge graph completion technology is used to convert the parsing results into triple rules. Then, through the hierarchical rule induction algorithm, a multi-level decision tree structure is constructed according to the triple rules to obtain the decision tree logical expression;

[0088] Through the automated compiler, the decision tree logical expressions are converted into Drools rule engine scripts to generate compliance verification rules.

[0089] Specifically, the text of environmental protection policies can be obtained from official websites of government departments, policy release platforms, and policy information services subscribed to internally by enterprises. Among them, the text clauses contain the specific provisions and requirements of the policy, the attached tables list the specific substance lists, emission standard values ​​and other detailed information, and the revision history records the evolution of the policy, which helps to understand the latest dynamics and future trends of the policy. The BERT-GNN hybrid model can then be used to parse the environmental protection policy text. Among them, the BERT model can deeply understand the semantic information in the text clauses and extract the relationship between key entities such as specific chemical substances, geographical areas, etc. and entities. GNN (Graph Neural Network) further enhances the ability to process structured information of policy texts. Through this hybrid model and using the conditional random field algorithm for sequence labeling, key information such as geographical constraints, banned material lists, and effective time thresholds can be accurately extracted.

[0090] Dynamic knowledge graph completion technology can then be used to convert the entities and relationships in the key information into triple rules, and combined with the decision tree induction algorithm to build a multi-level decision tree structure. This algorithm can organize the rules hierarchically, dividing the levels according to factors such as the importance and scope of application of the rules, and forming a decision tree logical expression. Each node in the decision tree represents a decision point, the branches represent different decision conditions and paths, and the final leaf node is the specific decision result. The Drools rule engine is a business rule management tool that can efficiently execute complex business logic rules. The compiler converts the logical structure of the decision tree into elements such as conditions, actions, and priorities in the Drools rule script. The generated compliance verification rules can be directly applied to other modules of the system to perform real-time compliance verification on various activities in the supply chain.

[0091] In one embodiment, the optimization decision generation module 103 includes a multi-objective optimization subunit, which is used to:

[0092] A carbon emission function is constructed based on the energy consumption of logistics vehicles and the carbon footprint of materials in standardized environmental data, combined with the carbon emission sensitivity in the supplier environmental resilience score;

[0093] Timeliness scores are calculated based on logistics vehicle energy consumption data and the historical average transportation times of different transportation modes. An efficiency function is constructed based on resource circulation rate data, combined with the deployment cost and activation threshold of backup nodes. Backup nodes are pre-set redundant nodes in the supply chain, used to replace failed nodes in the event of sudden disruptions. The deployment cost is calculated based on resource circulation rate data, and the activation threshold is dynamically adjusted based on environmental risk responsiveness. Combining the efficiency function with the carbon emission function yields a dual-objective optimization function.

[0094] Based on a dual-objective optimization function, the compliance verification rules and preset resource constraints are used as constraints, and an improved NSGA-III algorithm is used to generate a Pareto frontier solution set. The improved NSGA-III algorithm includes calculating the weight ratio of the efficiency function and the carbon emission function based on the dynamic compliance adaptation weight, and then allocating the reference point density in the objective space of the solution process according to the weight ratio. The next generation population is selected using the niching strategy based on the reference point density.

[0095] Extracting the decision variable vector of each solution from the Pareto front solution set, the decision variable vector includes path selection variables, transportation mode combination variables and node redundancy configuration variables;

[0096] Based on the coordinate range of the ecological protection zone based on the compliance verification rules, the path selection variables are geographically constrained and labeled to obtain path labels. The carbon emission function values ​​corresponding to the decision variable vectors are combined with the preset carbon emission threshold intervals to classify the levels and obtain carbon emission grade labels. Based on historical transportation data, the interruption probability of the corresponding path in the decision variable vector is calculated to generate a stability label. The carbon emission grade label, path label, and stability label are combined to obtain the path selection plan.

[0097] The transport mode combination variables are verified for vehicle compliance using compliance verification rules to obtain packaging compliance labels. The timeliness score is associated with the corresponding transport mode, and the energy consumption level of the transport mode is annotated in combination with the logistics vehicle energy consumption data to obtain the transport mode combination.

[0098] According to the standby node deployment cost and activation threshold in the node redundancy configuration variables, the redundancy level is divided to obtain the node redundancy configuration; the supply chain path solution set is obtained by combining the path selection scheme, transportation mode combination and node redundancy configuration.

[0099] Specifically, a backup node is a pre-configured redundant node in the supply chain that can replace a failed node in the event of a sudden disruption, ensuring supply chain continuity. The deployment cost of this node can be calculated based on resource circulation rate data, while the activation threshold can be dynamically adjusted based on environmental risk responsiveness to adapt to different risk conditions. An efficiency function is constructed based on the timeliness score, resource circulation rate data, backup node deployment cost, and activation threshold, balancing transportation timeliness with redundant configuration costs. Combining the efficiency function with the carbon emission function, a dual-objective optimization function is constructed, achieving coordinated optimization of minimizing carbon emissions and maximizing operational efficiency. This function can then be solved using the improved NSGA-III algorithm. The density of reference points in the target space is dynamically adjusted based on the dynamic compliance adaptation weights in the supplier's environmental resilience score. When policies prioritize environmental protection, the spacing between reference points in the carbon emission target dimension can be reduced, and the algorithm prioritizes low-emission solutions.

[0100] Schematically, the dynamic compliance adaptation weight can be calculated using the following formula:

[0101]

[0102] Among them, W t is the dynamic compliance adaptation weight at time t, Environmental risk responsiveness is the matching degree of the policy effective time threshold, which is calculated based on the matching degree between the effective time threshold and the current time t. is the deviation distance from the current transport path to the coordinate range of the ecological protection zone at time t, and α, β, and γ are all adaptive coefficients.

[0103] Furthermore, during the selection process, based on the density of reference points, a Niching strategy can be used to calculate the Euclidean distance between individuals and reference points, prioritizing evenly distributed individuals to avoid concentrating the solution set in a specific target area, thus ensuring the uniformity and integrity of the Pareto front. Compliance verification rules and preset resource constraints, such as maximum vehicle load and the upper limit on the number of backup nodes, can be used as algorithmic constraints. Individuals violating the rules can be directly assigned high penalty values ​​to ensure that the generated Pareto front solution set meets compliance requirements.

[0104] Specifically, the decision variable vector for each solution can be extracted from the Pareto front solution set. Based on this vector, the corresponding path selection plan can be obtained through geographical constraint annotation, carbon emission level classification, and stability assessment. Furthermore, the transport mode combination variable can be verified for vehicle compliance according to compliance verification rules, and the timeliness score can be associated with the corresponding transport mode. The energy consumption level of the transport mode can be annotated in combination with logistics vehicle energy consumption data to obtain a complete transport mode combination. In addition, the node redundancy configuration variable can be used to classify the redundancy level based on the standby node deployment cost and activation threshold, thereby obtaining a node redundancy configuration. Combining the above solutions, a set of supply chain path solutions can be obtained, providing scientific and optimized decision support for supply chain managers.

[0105] In one embodiment, the dynamic tuning module 104 includes:

[0106] Anomaly detection subunit, which is used to obtain supply chain execution data and set dynamic thresholds based on compliance verification rules to detect whether there are abnormal data in the supply chain execution data according to the dynamic thresholds, where the dynamic thresholds include vehicle energy consumption thresholds and transportation path deviation thresholds;

[0107] The exception handling subunit is used to: when abnormal data is detected, screen out solutions with preset conditions from the supply chain path solution set to generate a global rerouting candidate set; based on environmental parameter fluctuation data, optimize the path selection solutions in the global rerouting candidate set to obtain an optimized path selection solution and a decision result; associate the decision result with the corresponding supply chain execution data to generate a training set, and based on the training set, update the cross-enterprise ESG risk assessment model through incremental learning to obtain an updated cross-enterprise ESG risk assessment model.

[0108] Specifically, the anomaly detection subunit can acquire various data from the supply chain execution process in real time, including real-time energy consumption data, location information, and route deviations of transport vehicles. This data can reflect the real-time operational status of the supply chain. Dynamic thresholds can be set based on compliance verification rules, combined with historical data and current operational status. A theoretical energy consumption baseline can be calculated using a regression model based on factors such as vehicle type, load weight, and road conditions. This threshold can be used to determine whether a transport vehicle's energy consumption is within a reasonable range. Based on the geographic coordinate sequence of the compliant route plan, the Euclidean distance between the real-time location and the planned route can be calculated to set a route deviation threshold. This threshold can be used to determine whether a transport route has experienced an abnormal deviation. A sliding window algorithm can then be used to compare the supply chain execution data against the dynamic threshold in real time. If the data exceeds the threshold, an anomaly is identified. For example, if a transport vehicle's energy consumption exceeds the vehicle energy consumption threshold, or if the route deviation exceeds the route deviation threshold, the anomaly detection mechanism is triggered.

[0109] When the anomaly detection subunit detects anomalous data, it filters out options from the supply chain routing set that meet pre-set criteria to generate a global rerouting candidate set. These pre-set criteria can include route feasibility, cost-effectiveness, and environmental impact, ensuring that the candidate options can effectively address anomalies. Environmental parameter fluctuation data can include factors influencing supply chain operations, such as weather changes, traffic conditions (such as road congestion index), and environmental events. Based on this data, the routing options in the global rerouting candidate set can be optimized to improve supply chain efficiency and adaptability. For example, meteorological models can be used to predict 24-hour weather conditions along the route to avoid areas prone to severe weather such as heavy rain and strong winds. Costs can be calculated based on real-time updated transportation costs and carbon emissions estimates. Finally, the TOPSIS method is used to comprehensively evaluate each option's environmental performance, timeliness, and cost to determine the optimal routing option. The optimized routing option is used as the decision outcome and correlated with the corresponding supply chain execution data to generate a training set. Based on this training set, the cross-enterprise ESG risk assessment model is updated through incremental learning. This allows the model to adapt to new data and situations without retraining the entire model, further improving its accuracy and timeliness.

[0110] Based on the same inventive concept, Figure 2 As shown, the embodiments of the present application also provide a supply chain elasticity optimization decision-making method based on big data. The implementation solution provided by this method is similar to the implementation solution described in the above system. Therefore, the specific limitations of one or more embodiments of the supply chain elasticity optimization decision-making method based on big data provided below can be found in the above limitations of the supply chain elasticity optimization decision-making system based on big data, and will not be repeated here. The method includes:

[0111] S201: Collect multi-source heterogeneous environmental data, perform semantic alignment on the multi-source heterogeneous environmental data, and generate a standardized environmental data matrix. The multi-source heterogeneous environmental data includes logistics vehicle energy consumption data, material carbon footprint data, and resource recycling rate data.

[0112] S202: Based on the federated learning framework, a cross-enterprise ESG risk assessment model is constructed based on standardized environmental data, outputting a supplier environmental resilience score that includes carbon emission sensitivity, environmental risk responsiveness, and dynamic compliance adaptation weights.

[0113] S203: Obtain and parse the environmental policy text to obtain the parsed results, and compile the parsed results into compliance verification rules. The parsed results include geographical constraints, a list of prohibited materials, and an effective time threshold. Based on standardized environmental data, supplier environmental resilience scores, and compliance verification rules, a dual-objective optimization function is constructed and solved to generate a supply chain path solution set. The supply chain path solution set includes path selection options, transportation mode combinations, and node redundancy configurations.

[0114] S204: Monitor supply chain execution data in real time, trigger dynamic rerouting decisions based on the supply chain execution data, and update the cross-enterprise ESG risk assessment model. The supply chain execution data includes the real-time location of transportation vehicles, energy consumption monitoring values, and environmental parameter fluctuation data.

[0115] In the above method, by semantically aligning multi-source heterogeneous environmental data, not only is effective integration of data from different sources, such as logistics vehicle energy consumption and material carbon footprint, achieved, but also by generating a standardized environmental data matrix, a unified and accurate data foundation is provided for subsequent analysis, further enhancing the scientific nature of decision-making. Secondly, a cross-enterprise ESG risk assessment model is constructed based on a federated learning framework, avoiding the limitations of traditional single-enterprise independent assessments in terms of data sharing and privacy protection. Moreover, by outputting supplier environmental resilience scores, the model not only improves the comprehensiveness of risk assessments but also more accurately reflects the supplier's resilience level in the face of environmental risks. Furthermore, the compliance verification rules obtained from the environmental policy text parsing are embedded in a dual-objective optimization function for solution. This not only ensures that the decision-making plan strictly adheres to the environmental policy, but also achieves a balance between carbon emissions and efficiency through the coordinated optimization of path selection, transportation mode combination, and node redundancy configuration, thereby improving the practical feasibility and overall benefits of the decision-making plan. Finally, by real-time monitoring of supply chain execution data to trigger dynamic rerouting decisions and synchronously updating cross-enterprise ESG risk assessment models, it is not only possible to respond to emergencies during transportation in a timely manner, but also to enhance the model's adaptability to dynamic environments and the timeliness of risk assessment, further improving the resilience and stability of the supply chain.

[0116] In an exemplary embodiment, the present invention further provides a computer device comprising a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the big data-based supply chain resilience optimization decision system of the present application. A multi-core processor is preferred to improve the system's parallel processing capabilities. The memory provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate large amounts of supply information and computing tasks.

[0117] In an exemplary embodiment, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the supply chain resilience optimization decision system based on big data of the present application. The computer-readable storage medium may include: a read-only memory (ROM), a random access memory (RAM), a solid-state drive (SSD), or an optical disk, etc. Among them, the random access memory may include a resistance random access memory (ReRAM) and a dynamic random access memory (DRAM).

[0118] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A supply chain elasticity optimization decision system based on big data, characterized by: The system comprises: A data acquisition module is used to collect multi-source heterogeneous environmental data, perform semantic alignment processing on the multi-source heterogeneous environmental data, and generate a standardized environmental data matrix. The multi-source heterogeneous environmental data includes logistics vehicle energy consumption data, material carbon footprint data, and resource recycling rate data; An ESG risk assessment module, which is used to build a cross-enterprise ESG risk assessment model based on the standardized environmental data based on a federated learning framework, and output a supplier environmental resilience score, which includes carbon emission sensitivity, environmental risk responsiveness, and dynamic compliance adaptation weights; An optimization decision generation module is used to obtain and parse environmental policy texts, obtain parsed results, and compile the parsed results into compliance verification rules. The parsed results include geographic constraints, a list of prohibited materials, and an effective time threshold. Based on the standardized environmental data, the supplier environmental resilience score, and the compliance verification rules, a dual-objective optimization function is constructed and solved to generate a set of supply chain path solutions. The supply chain path solution set includes path selection options, transportation mode combinations, and node redundancy configurations. A dynamic tuning module is used to monitor supply chain execution data in real time, trigger dynamic rerouting decisions based on the supply chain execution data, and update the cross-enterprise ESG risk assessment model, wherein the supply chain execution data includes the real-time location of transportation vehicles, energy consumption monitoring values, and environmental parameter fluctuation data.

2. The system according to claim 1, wherein: The data acquisition module includes: The raw data acquisition subunit is used to: Obtaining real-time energy consumption data of the logistics vehicle; Obtaining the carbon footprint data of the bill of materials based on the enterprise ERP system, the carbon footprint data of the bill of materials being calculated and generated based on a life cycle assessment model, and obtaining the resource recycling rate data based on the enterprise production system, the resource recycling rate data including the material recovery rate, the waste reuse rate, and the proportion of recycled materials; Combining the logistics vehicle energy consumption data, the material carbon footprint data, and the resource recycling rate data to obtain the multi-source heterogeneous environmental protection data; Quantity processing subunit, used for: Construct an environmental knowledge graph, define core ontologies of carbon emissions, resource recycling rates, and geo-fencing in the environmental knowledge graph, and describe hierarchical relationships between entities using the OWL language; Establishing a semantic mapping relationship across data sources based on the core ontology, and detecting semantic conflicts among the multi-source heterogeneous environmental protection data through the semantic mapping relationship, wherein the semantic conflicts include measurement unit conflicts, time granularity deviations, and geocoding format differences, and obtaining conflict data; Performing data processing on the conflicting data according to the semantic mapping relationship to obtain pre-processed multi-source data; The pre-processed multi-source data is subjected to spatiotemporal alignment processing to generate the standardized environmental protection data matrix, where the dimensions of the standardized environmental protection data matrix include timestamp, geographic location and environmental protection indicators.

3. The system according to claim 1, wherein: The ESG risk assessment module includes: A feature extraction subunit is configured to locally deploy a convolutional sparse autoencoder with a masking mechanism at each enterprise, input the standardized environmental protection data into the convolutional sparse autoencoder for feature extraction, and generate an initial feature vector, wherein a Gaussian noise mask is added to the input layer of the convolutional sparse autoencoder, and the activation of the hidden layer of the convolutional sparse autoencoder is constrained by L1 regularization; a model construction subunit, configured to perform Paillier homomorphic encryption on the initial feature vector to generate an encrypted feature vector, upload the encrypted feature vector to a central server, aggregate the encrypted feature vectors of each enterprise according to an adaptive weighted federated averaging algorithm, and construct the cross-enterprise ESG risk assessment model; The environmental assessment sub-unit is used to calculate the supplier's environmental resilience score based on the cross-enterprise ESG risk assessment model.

4. The system according to claim 1, wherein: The optimization decision generation module includes the environmental policy analysis subunit, which is used to: Obtaining the environmental protection policy text, which includes text clauses, appendix forms, and revision history; The environmental protection policy text is parsed using a BERT-GNN hybrid model to obtain the parsed result, wherein the geographical constraints include the administrative divisions, the coordinate range of the ecological protection zone, and the boundary of the transport restricted zone defined in the environmental protection policy text; the banned materials list includes the CAS numbers of banned raw material chemicals and the types of non-degradable packaging materials; and the effective time threshold includes the effective date, expiration date, and temporary ban time window of the policy clause; Based on the entities and relationships in the parsing results, the parsing results are converted into triple rules using dynamic knowledge graph completion technology, and a multi-level decision tree structure is constructed according to the triple rules through a hierarchical rule induction algorithm to obtain a decision tree logical expression; The decision tree logic expression is converted into a Drools rule engine script through an automated compiler to generate the compliance verification rules.

5. The system according to claim 1, wherein: The optimization decision generation module includes a multi-objective optimization subunit, which is used to: Constructing a carbon emission function based on the logistics vehicle energy consumption and material carbon footprint in the standardized environmental data and the carbon emission sensitivity in the supplier environmental resilience score; A timeliness score is calculated based on the logistics vehicle energy consumption data and the historical average transportation time of different transportation modes. An efficiency function is constructed based on the resource circulation rate data and the deployment cost and activation threshold of the backup node. The backup node is a preset redundant node in the supply chain used to replace the failed node in the event of a sudden interruption. The deployment cost is calculated based on the resource circulation rate data, and the activation threshold is dynamically adjusted based on the environmental risk responsiveness. Combining the efficiency function and the carbon emission function to obtain the dual-objective optimization function; Based on the dual-objective optimization function, the compliance verification rules and preset resource constraints are used as constraints, and an improved NSGA-III algorithm is used to generate a Pareto front solution set. The improved NSGA-III algorithm includes calculating a weight ratio of the efficiency function and the carbon emission function according to the dynamic compliance adaptation weight, and allocating a reference point density according to the weight ratio in the target space of the solution process based on the weight ratio, and selecting the next generation population using a niching strategy based on the reference point density. Extracting a decision variable vector of each solution from the Pareto front solution set, wherein the decision variable vector includes a path selection variable, a transportation mode combination variable, and a node redundancy configuration variable; The path selection variables are geographically constrained and labeled based on the coordinate range of the ecological protection zone of the compliance verification rule to obtain the path label. The carbon emission function value corresponding to the decision variable vector is combined with the preset carbon emission threshold interval to divide the levels to obtain the carbon emission level label. Based on the historical transportation data, the interruption probability of the corresponding path in the decision variable vector is calculated to generate a stability label. Combining the carbon emission level label, the path label, and the stability label to obtain the path selection solution; Performing vehicle compliance verification on the transport mode combination variable using the compliance verification rule to obtain a packaging compliance label, associating the timeliness score with the corresponding transport mode, and annotating the energy consumption level of the transport mode in combination with the logistics vehicle energy consumption data to obtain the transport mode combination; According to the standby node deployment cost and activation threshold in the node redundancy configuration variable, redundancy levels are divided to obtain the node redundancy configuration; The supply chain path solution set is obtained by combining the path selection solution, the transportation mode combination and the node redundancy configuration.

6. The system according to claim 5, characterized in that The dynamic tuning module includes: an anomaly detection subunit, configured to obtain the supply chain execution data, set dynamic thresholds based on the compliance verification rules, and detect whether there is abnormal data in the supply chain execution data according to the dynamic thresholds, wherein the dynamic thresholds include vehicle energy consumption thresholds and transportation path deviation thresholds; Exception handling subunit, used to: When the abnormal data is detected, solutions with preset conditions are screened out from the supply chain path solution set to generate a global rerouting candidate set; Based on the environmental parameter fluctuation data, the path selection scheme in the global rerouting candidate set is optimized to obtain an optimized path selection scheme and a decision result; The decision result and the corresponding supply chain execution data are associated to generate a training set, and based on the training set, the cross-enterprise ESG risk assessment model is updated through incremental learning to obtain the updated cross-enterprise ESG risk assessment model.

7. The system according to claim 5, characterized in that The dynamic compliance adaptation weight is calculated using the following formula: Among them, W t is the dynamic compliance adaptation weight at time t, The environmental risk response is the policy effective time threshold matching degree, which is calculated based on the matching degree between the effective time threshold and the current time t. is the deviation distance from the current transport path at time t to the coordinate range of the ecological protection zone, and α, β, and γ are all adaptive coefficients.

8. A supply chain elasticity optimization decision-making method based on big data, characterized by: The method comprises: Collecting multi-source heterogeneous environmental data, performing semantic alignment processing on the multi-source heterogeneous environmental data, and generating a standardized environmental data matrix. The multi-source heterogeneous environmental data includes logistics vehicle energy consumption data, material carbon footprint data, and resource recycling rate data; Based on the federated learning framework, a cross-enterprise ESG risk assessment model is constructed based on the standardized environmental data to output a supplier environmental resilience score, which includes carbon emission sensitivity, environmental risk responsiveness, and dynamic compliance adaptation weights; Obtaining and parsing the environmental policy text to obtain parsing results, and compiling the parsing results into compliance verification rules. The parsing results include geographic constraints, a list of prohibited materials, and an effective time threshold. Based on the standardized environmental data, the supplier environmental resilience score, and the compliance verification rules, a dual-objective optimization function is constructed and solved to generate a supply chain path solution set. The supply chain path solution set includes path selection solutions, transportation mode combinations, and node redundancy configurations. Monitor supply chain execution data in real time, trigger dynamic rerouting decisions based on the supply chain execution data, and update the cross-enterprise ESG risk assessment model, wherein the supply chain execution data includes the real-time location of transportation vehicles, energy consumption monitoring values, and environmental parameter fluctuation data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the system according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the system according to any one of claims 1 to 7 are implemented.

Images