A method for determining carbon footprint of a cable throughout its life cycle

Abstract

This invention relates to the field of cable carbon footprint technology and discloses a method for determining the carbon footprint of a cable throughout its entire life cycle. The method includes: acquiring carbon emission data for each stage of the cable's life cycle—raw material acquisition, manufacturing, transportation, installation, use, and end-of-life recycling—through a pre-set data acquisition system; establishing a multi-dimensional characterization model of the carbon footprint features using the acquired data; performing a life cycle stage adaptability assessment on the model to generate a carbon footprint contribution distribution map; parsing the semantic content of cable product specification parameters, deriving the corresponding environmental impact weights, and forming carbon footprint calculation constraints based on the carbon footprint contribution distribution map; generating a carbon footprint determination scheme using an optimization algorithm based on the constraints, and implementing the carbon footprint output according to the scheme. This method can systematically integrate relevant information on carbon emissions throughout the cable's entire life cycle, providing a clear path for carbon footprint determination.

Description

Technical Field

[0001] This invention relates to the field of cable carbon footprint technology, specifically a method for determining the carbon footprint of a cable throughout its entire life cycle. Background Technology

[0002] As a key basic product in fields such as power transmission and communication, the carbon emissions of cables throughout their entire life cycle have attracted much attention, making the accurate determination of the carbon footprint of cables throughout their life cycle an important need in the industry. Currently, the methods for calculating the carbon footprint of cables in the industry are mostly fragmented, failing to achieve effective coverage and integration of all stages of the cable's life cycle.

[0003] In the raw material acquisition stage, current methods often only focus on the carbon emission data of some major raw materials, ignoring the carbon emission information of auxiliary materials, consumables, and other related processes, resulting in incomplete data collection at this stage. The manufacturing stage involves multiple production processes, each with different carbon emission sources. Current data collection largely relies on manual recording or single-device monitoring, making it difficult to achieve real-time and comprehensive acquisition of carbon emission data for each process. Furthermore, data accuracy is easily affected by human factors or equipment precision. In the transportation and distribution stage, factors such as transportation mode, transportation distance, and loading rate all affect carbon emissions. However, current calculations often simplify these influencing factors, using only fixed coefficients for estimation, failing to combine actual cable transportation scenarios for accurate data collection and calculation.

[0004] Carbon emissions during the installation and use phase are often overlooked by existing methods. This phase involves carbon emissions from energy consumption and equipment use during installation, as well as energy-related carbon emissions during cable use. Current methods often fail to include these aspects in carbon footprint calculations, resulting in incomplete coverage. In the end-of-life recycling phase, different recycling processes and efficiencies have varying impacts on carbon emissions. Existing calculations often lack data analysis of specific stages of the recycling process, relying solely on industry averages for estimation, which fails to reflect the actual carbon emissions from the recycling of different cable products at the end of their lifespan.

[0005] Existing carbon footprint calculation models are mostly single-dimensional, capable of representing carbon footprint characteristics from only one perspective and failing to present the changing patterns and influencing factors of carbon footprint from multiple dimensions. In the assessment phase, there is a lack of adaptive assessments for each stage of the life cycle, making it difficult to clarify the contribution of each stage to the overall carbon footprint and pinpoint key areas for carbon emission reduction. Furthermore, existing methods do not consider the impact of differences in cable product specifications on the carbon footprint. Different cable specifications differ in raw material usage, production processes, and energy consumption; ignoring these differences leads to a lack of specificity in carbon footprint calculations, failing to accurately reflect the actual carbon emissions of different cable specifications. Moreover, existing calculation processes lack clearly defined constraints, relying heavily on empirical formulas or simplified algorithms to generate carbon footprint calculation results. The lack of algorithm optimization makes it difficult to guarantee the rationality and accuracy of the carbon footprint determination scheme, ultimately resulting in biased results in the determination of the cable's full life cycle carbon footprint, failing to provide a reliable basis for relevant decision-making in the cable industry. Summary of the Invention

[0006] The purpose of this invention is to provide a method for determining the carbon footprint of a cable throughout its entire life cycle, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for determining the carbon footprint of a cable throughout its entire life cycle, the method comprising:

[0008] Carbon emission data for each stage of the cable's life cycle—from raw material acquisition, production and manufacturing, transportation and distribution, installation and use, to scrapping and recycling—is obtained through a pre-set data acquisition system.

[0009] A multi-dimensional characterization model of carbon footprint features was established using the acquired carbon emission data.

[0010] A life cycle stage adaptive assessment is performed on the multi-dimensional characterization model to generate a carbon footprint contribution distribution map;

[0011] The semantic content contained in the cable product specification parameters is analyzed, the environmental impact weights corresponding to the semantic content are derived, and the carbon footprint contribution distribution map is combined to form carbon footprint calculation constraints.

[0012] Based on the aforementioned carbon footprint calculation constraints, an optimization algorithm is used to generate a carbon footprint determination scheme;

[0013] The carbon footprint output process is implemented according to the carbon footprint determination scheme.

[0014] Preferably, the step of acquiring carbon emission data for each stage of the cable's entire lifecycle—including raw material acquisition, manufacturing, transportation and distribution, installation and use, and end-of-life recycling—through a preset data acquisition system includes the following steps:

[0015] Build a distributed edge computing node network and deploy dedicated data acquisition agents at each stage of the lifecycle;

[0016] The dedicated data acquisition agent captures energy consumption monitoring data during the raw material mining stage, process parameter flow data during the production and manufacturing stage, and path trajectory data during the transportation stage in real time.

[0017] A dynamic time warping algorithm is used to align the time axis of time-series data acquired asynchronously from multiple sources.

[0018] An attention-based data quality assessment model was applied to perform anomaly detection and cleaning on the collected data.

[0019] The cleaned data is categorized and stored in a time-series database according to life cycle stages, forming a structured carbon emission dataset.

[0020] Preferably, the step of establishing a multi-dimensional characterization model of carbon footprint features using the acquired carbon emission data includes the following steps:

[0021] The structured carbon emission dataset is input into a spatiotemporal graph convolutional network to construct a feature graph with life cycle stages as nodes and material and energy flows as edges.

[0022] Learn the implicit carbon association features between nodes through a multi-layer graph attention mechanism;

[0023] A feature pyramid network is used to extract multi-scale carbon footprint features, including macro-level industrial chain features and micro-level process features;

[0024] Using bidirectional long short-term memory networks to capture the dynamic evolution of carbon footprint over time;

[0025] The feature maps, implicit carbon correlation features, macro-level industrial chain features, micro-level process features, and dynamic evolution laws are fused using tensors to generate a multi-dimensional representation model with spatiotemporal correlation.

[0026] Preferably, the lifecycle stage adaptive assessment of the multi-dimensional representation model includes the following steps:

[0027] We construct an adaptive evaluation framework based on deep reinforcement learning and design a reward function with the goal of minimizing carbon footprint.

[0028] The agent is trained and evaluated using a policy gradient algorithm, enabling it to learn to identify key influencing factors at each stage of its life cycle.

[0029] The Monte Carlo tree search algorithm was used to explore potential optimization paths in a multi-dimensional representation model.

[0030] Generate an adaptive assessment report that includes stage contribution, sensitivity indicators, and optimization priorities;

[0031] Based on the priority indicators in the assessment report, a carbon footprint contribution distribution map was constructed.

[0032] Preferably, the semantic content included in the analytical cable product specification parameters includes the following steps:

[0033] Establish an ontological model of the technical parameters of cable products and define the semantic relationships between material composition, structural dimensions, and performance indicators;

[0034] The specification parameters of text format are transformed into vector representations by applying knowledge graph embedding technology;

[0035] The importance weights of each parameter node are calculated by traversing the semantic path in the ontology model using a graph neural network.

[0036] By combining environmental impact factors in the life cycle assessment database, the environmental impact weights corresponding to semantic content are derived.

[0037] Construct a semantic parsing knowledge base that includes weight coefficients.

[0038] Preferably, the process of combining the carbon footprint contribution distribution map to form carbon footprint calculation constraints includes the following steps:

[0039] The carbon footprint contribution distribution map is transformed into a multi-dimensional constrained space, with each dimension representing a life cycle stage.

[0040] An algorithm for solving constraint satisfaction problems is applied to search for a feasible solution domain in the constraint space.

[0041] A trade-off model between economic efficiency, feasibility, and environmental impact is established using multi-objective optimization theory;

[0042] Inconsistent regions in the constraint space are eliminated through constraint propagation algorithms.

[0043] Generate a set of constraints for carbon footprint calculation that includes boundary conditions, trade-offs, and feasibility rules.

[0044] Preferably, the step of generating a carbon footprint determination scheme using an optimization algorithm based on the carbon footprint calculation constraints includes the following steps:

[0045] Design a mixed integer programming model to transform the carbon footprint calculation constraints into a mathematical programming problem;

[0046] The branch and bound algorithm is applied to solve for discrete decision variables in the model;

[0047] Interior point method is used to handle continuous variable optimization problems;

[0048] The robustness of the solution was verified through sensitivity analysis;

[0049] Generate a carbon footprint determination scheme that includes the optimal solution set and corresponding decision variables.

[0050] Preferably, the carbon footprint output process based on the carbon footprint determination scheme includes the following steps:

[0051] Encode the carbon footprint determination scheme into a standardized data exchange format;

[0052] Digital signature technology is used to ensure data integrity and immutability;

[0053] Construct a carbon footprint evidence storage system based on distributed ledger technology;

[0054] Carbon footprint data is pushed to relevant regulatory platforms via application programming interfaces (APIs).

[0055] Enables real-time verification and traceability of carbon footprint data.

[0056] Preferably, the construction of the distributed edge computing node network includes the following steps:

[0057] Deploy IoT gateways with edge computing capabilities in all aspects of the cable industry chain;

[0058] Configure a lightweight data acquisition protocol to enable communication between devices;

[0059] Differential privacy technology is used to de-identify the collected data;

[0060] Establish a secure transmission channel between edge nodes and the cloud platform;

[0061] Implement an automatic scaling up / down and failover mechanism for the data acquisition network.

[0062] Preferably, the construction of the adaptive evaluation framework based on deep reinforcement learning includes the following steps:

[0063] Design a Markov decision process that includes a state space, an action space, and a reward function;

[0064] A deep reinforcement learning network is constructed using an actor-critic architecture;

[0065] Improve training efficiency through experience playback mechanisms;

[0066] Gradually increase the complexity of assessment by applying curriculum learning strategies;

[0067] Establish a closed loop for online evaluation and feedback optimization of model performance.

[0068] Compared with the prior art, the beneficial effects of the present invention are:

[0069] Carbon emission data for the entire lifecycle of cables is acquired through a pre-designed data acquisition system. This data acquisition process covers all key stages, including raw material acquisition, manufacturing, transportation and distribution, installation and use, and end-of-life recycling, overcoming the fragmentation and limitations of existing methods in data collection. In the raw material acquisition stage, comprehensive carbon emission-related data, including major raw materials, auxiliary materials, and consumables, can be collected, avoiding calculation errors caused by missing data. In the manufacturing stage, systematic collection of carbon emission data for each production process can be achieved, reducing errors from manual recording or single-equipment monitoring and improving data accuracy. In the transportation and distribution stage, specific data such as transportation methods, distances, and loading rates can be collected based on actual transportation scenarios, replacing fixed-coefficient estimations and making the carbon emission data at this stage more realistic. In the installation and use stage, installation energy consumption, equipment usage carbon emissions, and energy consumption-related carbon emissions during use are included in the collection scope, improving the stage coverage of carbon footprint data. In the end-of-life recycling stage, data is collected for different recycling processes and efficiencies, abandoning industry average data estimations and reflecting the true carbon emissions from recycling.

[0070] By utilizing complete acquired data to establish a multi-dimensional characterization model of carbon footprint features, compared to existing single-dimensional models, this model can present the characteristics and changing patterns of the carbon footprint from multiple perspectives, more comprehensively reflecting the influencing factors of the cable's entire lifecycle carbon footprint. This helps users to more clearly understand the composition and changing trends of the carbon footprint. Performing a lifecycle-stage adaptive assessment on the multi-dimensional characterization model can clarify the contribution of each stage to the overall carbon footprint. The generated carbon footprint contribution distribution map can intuitively display the carbon emission proportion of each stage, facilitating users to quickly identify key carbon reduction links and providing a clear direction for the formulation of subsequent carbon reduction measures. This avoids the problem of existing methods failing to identify emission reduction priorities due to the lack of stage assessment.

[0071] This method analyzes the semantic content of cable product specifications and derives the corresponding environmental impact weights, fully considering the differences in raw material usage, production processes, and energy consumption among different cable specifications. Combining these weights with a carbon footprint contribution distribution map forms the constraints for carbon footprint calculation. This makes the constraints more closely reflect the actual situation of cable products, resulting in more targeted carbon footprint calculations and avoiding the inaccuracies caused by existing methods ignoring product specification differences. Based on these constraints, an optimization algorithm is used to generate a carbon footprint determination scheme. By combining actual data from the entire cable lifecycle with product characteristics, a more reasonable calculation path and parameters can be selected, ensuring the scientific validity and feasibility of the scheme. Compared to existing methods relying on empirical formulas or simplified algorithms, this significantly improves the reliability of carbon footprint determination results. Attached Figure Description

[0072] Figure 1 This is a schematic diagram illustrating the working principle of the cable lifecycle carbon footprint determination method described in this invention.

[0073] Figure 2 A flowchart for carbon emission data acquisition and processing;

[0074] Figure 3 This is a flowchart for the lifecycle stage adaptive assessment. Detailed Implementation

[0075] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0076] Please see Figure 1 This invention provides a method for determining the carbon footprint of a cable throughout its entire lifecycle. The method includes: calculating the carbon footprint by integrating data from each stage of the cable's lifecycle; relying on a pre-set data acquisition system to obtain carbon emission data for each stage of the cable's lifecycle, including raw material acquisition, manufacturing, transportation and distribution, installation and use, and end-of-life recycling. The acquired carbon emission data is used to establish a multi-dimensional characterization model of the carbon footprint, which can depict the characteristics of carbon emissions from multiple perspectives. A lifecycle stage adaptability assessment is performed on the multi-dimensional characterization model, generating a carbon footprint contribution distribution map that clearly shows the degree of influence of different stages on the total carbon footprint. The method further analyzes the semantic content contained in the cable product specification parameters, deriving the environmental impact weights corresponding to the semantic content, and combining these weights with the carbon footprint contribution distribution map to form a set of carbon footprint calculation constraints. Based on the carbon footprint calculation constraints, an optimization algorithm is used to generate the final carbon footprint determination scheme.

[0077] Example 1: See Figure 2A distributed edge computing node network forms the basic architecture of the entire data acquisition system. IoT gateways with edge computing capabilities are deployed at various physical locations throughout the cable industry chain, including raw material acquisition, manufacturing, transportation, installation, and end-of-life recycling. Each IoT gateway operates as an independent edge computing node. These distributed IoT gateways are connected via enterprise private networks or virtual private networks to form a unified edge computing resource pool. The network topology adopts a mesh network design to enhance robustness. A dedicated data acquisition agent is deployed on each IoT gateway in the distributed edge computing node network. This dedicated data acquisition agent is a custom-developed software module responsible for communication and interaction with field devices. A lightweight data acquisition protocol is configured to achieve communication between devices. The protocol stack uses a simplified version of the MQTT protocol, and the message header structure has been optimized to reduce network transmission overhead. The lightweight data acquisition protocol supports publish and subscribe modes, decoupling data producers and consumers. The protocol has a built-in heartbeat mechanism to maintain connection status. Energy consumption monitoring data from the raw material mining stage is captured in real time through the dedicated data acquisition agent. This energy consumption monitoring data comes from field instruments such as smart meters and flow meters, and the data acquisition frequency is configurable according to process characteristics. Real-time capture of process parameter stream data during the manufacturing stage, including continuously changing process variables such as extruder temperature, traction speed, and cooling water flow rate. Real-time capture of route trajectory data during transportation, acquired via vehicle-mounted GPS terminals, simultaneously recording timestamps, latitude and longitude coordinates, and driving status information. Differential privacy technology is employed to anonymize the collected data. This technology achieves privacy protection by adding precisely calculated noise to the original data; the distribution and intensity of the noise are strictly controlled using privacy budget parameters. Differential privacy processing is performed locally on edge computing nodes, preventing the original sensitive data from leaving the collection point.

[0078] A dynamic time warping algorithm is employed to align the timeline of asynchronously acquired time-series data from multiple sources. This algorithm can flexibly stretch or compress time series to find the optimal alignment path. The asynchronously acquired time-series data originates from field devices of different manufacturers and models, each with slight clock deviations. The data processed by the dynamic time warping algorithm has a unified time reference system, establishing a consistent time benchmark for subsequent analysis. An attention-based data quality assessment model is applied to detect and clean the acquired data for anomalies. This model employs an encoder-decoder architecture, with the attention weight matrix automatically learning the importance distribution of data points. The anomaly detection module identifies data points deviating from normal patterns, the data cleaning module uses interpolation or removal strategies to handle outliers, and the data quality assessment report generation module outputs a quality score for each data stream. A secure transmission channel is established between edge nodes and the cloud platform. This secure channel is encrypted using the TLS protocol with national cryptographic algorithms, and a two-way authentication mechanism ensures the legitimacy of both parties. The transmission channel has automatic reconnection and session recovery functions to handle abnormal situations such as network interruptions. After the secure transmission channel is established, the edge nodes upload the processed data to the cloud platform in batches. The cleaned data is categorized and stored in a time-series database according to its lifecycle stages. The chosen time-series database is the open-source version InfluxDB, whose storage engine is optimized for efficient writing and querying of time-series data. Independent tables are created for each of the five lifecycle stages: raw material acquisition, production and manufacturing, transportation and distribution, installation and use, and disposal and recycling. Each table includes fields such as timestamps, measurement point identifiers, numerical values, and quality identifiers. This results in a structured carbon emission dataset with a unified schema definition, supporting SQL-like query languages ​​for data access.

[0079] An automatic scaling and failover mechanism for the data acquisition network is implemented. The automatic scaling function is achieved through Kubernetes container orchestration technology, dynamically adjusting the number of edge computing node instances based on the data acquisition load. The failover mechanism is based on heartbeat detection and master-slave switching principles; when the master node fails, the standby node automatically takes over the service. A monitoring system tracks the health status of the distributed edge computing node network in real time, and the operation and maintenance management platform provides a visualized network topology monitoring interface. The data acquisition stage of the cable lifecycle carbon footprint determination method achieves distributed processing, proximity computing, and privacy protection through the distributed edge computing node network. A dedicated data acquisition agent encapsulates the communication details with specific device models, providing a standardized data acquisition interface. A lightweight data acquisition protocol reduces network bandwidth consumption, adapting to the relatively limited network conditions in industrial environments. The application of differential privacy technology balances the needs of data utility and privacy protection. A dynamic time warping algorithm solves the time synchronization problem of data from multi-source heterogeneous devices. An attention-based data quality assessment model improves the intelligence level of data cleaning. A secure transmission channel ensures the confidentiality and integrity of data during transmission. The selection of a time-series database optimizes the storage efficiency of time-series data. Automatic scaling up / down and failover mechanisms enhance the resilience and reliability of the data acquisition system.

[0080] Example 2: A structured carbon emission dataset is input into a spatiotemporal graph convolutional network. The structured carbon emission dataset originates from data collected and preprocessed by the distributed edge computing node network described in Example 1. The dataset contains carbon emission information in time-series format. A spatiotemporal graph convolutional network is a deep learning architecture specifically designed for processing graph-structured time-series data. Its design integrates the ability of graph convolutional networks to capture spatial relationships with the ability of temporal models to model dynamic changes. A feature graph is constructed with lifecycle stages as nodes and material and energy flows as edges. The node set covers five core stages: raw material acquisition, manufacturing, transportation and distribution, installation and use, and end-of-life recycling. Each node embeds attributes including the average carbon emission intensity, cumulative carbon emissions, and carbon emission trend slope for that stage. Edges connect lifecycle stages with material or energy transfer relationships. The weight of the edges is determined by the amount of material or energy flowing between stages. The topology of the feature graph reflects the networked characteristics of carbon flow throughout the cable's lifecycle. This study employs a multi-layered graph attention mechanism to learn implicit carbon correlation features between nodes. The graph attention mechanism dynamically adjusts the weights of information transmission by calculating attention coefficients between nodes, avoiding the fixed neighborhood aggregation method in traditional graph convolution. The first layer focuses on the influence between directly adjacent nodes, calculating the attention weights of raw material acquisition nodes on production nodes to capture the direct impact of raw material selection on carbon emissions in the production process. The second layer extends to the second-order neighborhood, analyzing the indirect carbon impact of raw material acquisition nodes on transportation and distribution nodes through production nodes, revealing cross-stage carbon correlation paths in the life cycle. This multi-layered graph attention mechanism forms a hierarchical feature extraction process, with each layer outputting a weighted node feature representation. The implicit carbon correlation features encode the complex carbon dependencies between nodes in the form of high-dimensional vectors.

[0081] A feature pyramid network is employed to extract multi-scale carbon footprint features. This network captures both macro-level (industry chain level) and micro-level (process level) features of the carbon footprint through feature maps at different scales. Macro-level features characterize the carbon footprint pattern throughout the entire cable lifecycle from a global perspective, including the coordinated changes in carbon emissions at each stage of the industry chain over time and the overall network structure characteristics of carbon flow between stages. Micro-level features focus on the detailed carbon emission mechanisms within a single lifecycle stage, such as the carbon emission profile under different combinations of process parameters in the manufacturing stage and the potential impact of the molecular structure characteristics of raw materials on carbon emissions. The feature pyramid network combines low-resolution, strongly semantic features with high-resolution, weakly semantic features through top-down paths and lateral connections, achieving multi-scale feature fusion and enhancement. A bidirectional long short-term memory (LSTM) network is used to capture the dynamic evolution of the carbon footprint over time. This network consists of forward and backward LSM networks, simultaneously considering the past and future context of the time series. The forward long short-term memory network processes carbon emission data chronologically, learning the carbon footprint evolution pattern from early to late stages. The backward long short-term memory network processes data in reverse chronological order, capturing the feedback effect of later stages on earlier stages. The memory units of the bidirectional long short-term memory network selectively retain or forget historical information through a gating mechanism, avoiding long-term dependency problems. The hidden state vector encodes the dynamic characteristics of the carbon footprint over time. These dynamic evolutionary patterns over time include periodic fluctuations, trend changes, and abrupt change detection in carbon emissions. These patterns reveal the intrinsic dynamics of carbon footprint development over time.

[0082] This paper employs tensor fusion to integrate features from different sources and with different properties within a unified mathematical space. Feature maps provide the basic topological structure between lifecycle stages; latent carbon-related features encode nonlinear interactions between stages; macro-industry-level features describe system-level carbon footprint patterns; micro-process-level features reveal details of carbon emissions in local processes; and dynamic evolution laws capture the temporal behavior of the carbon footprint. Tensor fusion uses a tensor decomposition-based strategy to map different features to a shared latent space and calculates the interaction effects between features through tensor multiplication. The fusion process introduces learnable weight parameters, and the network automatically adjusts the contribution of various features to the final representation, generating a multi-dimensional representation model with spatiotemporal correlation. This multi-dimensional representation model is a high-dimensional, dense vector representation that simultaneously encodes the spatial distribution and temporal evolution characteristics of the carbon footprint, comprehensively representing the carbon footprint characteristics throughout the cable's entire lifecycle. Spatiotemporal graph convolutional networks process feature maps, transforming each stage of the lifecycle and its connections into computable graph-structured data. Graph attention mechanisms empower the model to focus on important connections, avoiding interference from irrelevant information. The multi-scale feature extraction architecture of the feature pyramid network adapts to the need for both macro and micro perspectives in carbon footprint analysis, while bidirectional long short-term memory networks model temporal dependencies, capturing the dynamic nature of the carbon footprint. Tensor fusion technology integrates heterogeneous features into a unified representation space, generating a multi-dimensional representation model that provides a rich feature foundation for subsequent adaptive assessment and optimization decisions. The entire modeling process reflects a multi-layered and multi-faceted understanding of the complexity of the cable lifecycle carbon footprint. Feature maps serve as the structural skeleton, implicit carbon-related features reveal deep connections, multi-scale features cover different analytical granularities, dynamic evolution laws describe temporal behavior, and tensor fusion achieves information integration.

[0083] Example 3: See Figure 3 This paper constructs an adaptive evaluation framework based on deep reinforcement learning. The framework aims to systematically evaluate the impact, sensitivity, and optimization potential of each life cycle stage in a multi-dimensional representation model on the overall carbon footprint. Deep reinforcement learning learns the optimal evaluation strategy through the interaction between the agent and the environment. A Markov decision process is designed, comprising a state space, an action space, and a reward function. The state space is defined by the high-dimensional feature vectors output by the multi-dimensional representation model. The action space encompasses the set of all feasible operations that adjust the life cycle stages. The reward function is designed to quantitatively evaluate the combined impact of actions on environmental performance and economic efficiency. The mathematical expression of the reward function is as follows:

[0084]

[0085] in: Indicates the moment of decision-making The reward value obtained, Indicates the moment of decision-making The change in total life-cycle carbon emissions resulting from taking action. It is the carbon price conversion factor. Indicates the moment of decision-making The added value generated after taking action It is a value conversion factor. Indicates the moment of decision-making The marginal cost required to implement the action. It is the cost multiplication factor.

[0086] A deep reinforcement learning network is constructed using an actor-critic architecture. The actor network is a policy network whose input is the state space vector of a Markov decision process, and whose output is a probability distribution in the action space. The policy network parameters determine the agent's tendency to choose different actions in a specific state. The critic network is a value network whose input is a state vector or state-action pair, and whose output is an estimate of the expected cumulative reward. The value network parameters are used to evaluate the quality of a state or state-action pair. The combination of the actor-critic architecture allows policy updates to depend on both actual empirical rewards and the guidance of the value function. The actor network is responsible for generating behavior, and the critic network is responsible for evaluating behavior. The evaluation agent is trained using a policy gradient algorithm, enabling it to learn to identify key influencing factors at each lifecycle stage. The policy gradient algorithm directly optimizes the parameters of the policy function, adjusting the policy in the direction of increasing expected reward. During training, the evaluation agent interacts with a cable lifecycle simulation environment, exploring the state space defined by the multi-dimensional representation model, performing actions, and observing the rewards and state transitions returned by the environment.

[0087] This paper employs the Monte Carlo Tree Search (MCS) algorithm to explore potential optimization paths in a multi-dimensional representation model. MCS is a tree-based search algorithm that selects the most promising action sequence through iterative simulation. The algorithm starts from the root node representing the current state and iteratively executes four steps: selection, expansion, simulation, and backtracking. In the selection step, the existing tree structure is traversed, balancing the exploration of new nodes with the utilization of known high-reward nodes based on an upper confidence bound algorithm. In the expansion step, the tree structure is expanded when insufficiently explored nodes are encountered. In the simulation step, a randomized strategy or a fast evaluation strategy is used to simulate from new nodes until the end of the round. In the backtracking step, the simulation results are backpropagated to update the visit count and average reward value of nodes on the path. The MCS algorithm outputs a suggested action sequence starting from the current state and its value assessment.

[0088] The experience replay mechanism improves training efficiency by storing experience data samples generated from the agent's interaction with the environment in a fixed-capacity experience replay buffer. Each experience data sample includes the current state, the action performed, the immediate reward obtained, the next state, and a flag indicating whether the round has ended. During training, a small batch of experience data is randomly sampled from the experience replay buffer. This random sampling breaks the temporal correlation between data and reduces variance during training. The experience replay mechanism allows each experience sample to be used multiple times, improving data utilization efficiency and contributing to the stability of the deep neural network training process. A curriculum learning strategy is applied to gradually increase evaluation complexity. This strategy decomposes the complex lifecycle stage adaptive evaluation task into a sequence of tasks from easy to difficult. Initially, the evaluation agent is trained in a simplified environment, for example, considering only the interactions of the raw material acquisition and manufacturing stages, with both the state space and action space reduced. As the training rounds increase, more lifecycle stages are gradually introduced, increasing the dimensionality of the state vector and the complexity of the action space, ultimately training in a complete model environment encompassing all stages of raw material acquisition, manufacturing, transportation, installation, use, and disposal / recycling. The course learning strategy mimics the human learning process, avoiding the learning difficulties caused by excessive complexity in the initial stage, and helps the policy network converge to a better solution more smoothly.

[0089] An adaptive assessment report is generated, including stage contribution, sensitivity indicators, and optimization priorities. Stage contribution calculates the proportion of carbon emissions from each life cycle stage in the total life cycle carbon emissions. Sensitivity indicators are measured by perturbing key input parameters and observing the rate of change in the output carbon footprint; for example, local sensitivity analysis is used to calculate the elasticity of the impact of process parameter changes on stage carbon emissions. Optimization priority is a comprehensive indicator that ranks stages based on both stage contribution and sensitivity indicators; stages with high contribution and high sensitivity are typically assigned higher optimization priorities. The adaptive assessment report outputs quantitative assessment results in a structured document format. Based on the priority indicators in the assessment report, a carbon footprint contribution distribution map is constructed. This map is a visualization tool, typically presented as a Sankey diagram or radar chart. A Sankey diagram visually displays the carbon flow and its direction between different life cycle stages, with the arrow width proportional to the carbon flow. A radar chart can simultaneously display the contribution and sensitivity of multiple stages; each stage corresponds to an axis on the radar chart, with the contribution value determining the point's position on the axis, and the sensitivity value represented by color intensity or marker size. The carbon footprint contribution distribution map makes complex assessment results easier to understand and analyze.

[0090] A closed-loop system for online model performance evaluation and feedback optimization is established. The online evaluation module continuously monitors the performance of the deep reinforcement learning agent in the test environment, recording performance metrics such as average round reward, value function estimation error, and policy entropy. The feedback optimization module dynamically adjusts training hyperparameters, such as learning rate, exploration rate, and discount factor, or triggers additional training tasks based on performance metrics. This closed-loop system enables the adaptive evaluation framework to learn and continuously improve, adapting to changes in the distribution of cable lifecycle data. The deep reinforcement learning-based adaptive evaluation framework transforms the evaluation process into a sequential decision problem, with Markov decision processes providing the foundation for formal modeling. The actor-critic architecture utilizes value function estimation to guide policy search, improving learning efficiency. Monte Carlo tree search intelligently focuses on promising regions for in-depth exploration within the vast state-action space. The experience replay mechanism improves sample efficiency and stabilizes training through data reuse. The curriculum learning strategy guides the agent to learn complex evaluation strategies by progressively increasing task complexity. The adaptive evaluation report provides a structured and quantitative summary of the evaluation results. The carbon footprint contribution distribution map provides a visual insight into the evaluation results. The closed-loop system of online model performance evaluation and feedback optimization ensures the long-term effectiveness and adaptability of the evaluation framework.

[0091] Example 4: An ontological model of cable product technical parameters is established. The model is formally expressed using OWL language, defining the types, attributes, and interrelationships of core concepts within the cable product domain. Core concept categories include material entities, structural entities, and performance entities. Material entities encompass specific material categories such as conductor materials, insulation materials, and sheath materials, each possessing physical properties such as density, conductivity, and coefficient of thermal expansion. Structural entities describe the cable's geometric characteristics, such as conductor cross-sectional area, insulation layer thickness, and shielding layer braiding density. Performance entities characterize the cable's functional indicators, such as rated voltage, current carrying capacity, and service life. Knowledge graph embedding technology is applied to transform the text-formatted specifications into vector representations. This technology maps entities and relations in the ontological model to a low-dimensional continuous vector space, preserving the original graph structure information. The TransE algorithm is used for embedding learning, treating relations as translation operations in the entity vector space, such that the sum of the head entity vector and the relation vector approximately equals the tail entity vector. The specifications for the text format are derived from product design documents, technical standard manuals, and enterprise databases. This unstructured text data is processed by named entity recognition and relation extraction modules to extract entity instances and relation instances that conform to the ontology model definition. The generated vector representation is a fixed-dimensional floating-point array, with each entity and relation corresponding to a unique vector. The distance and direction in the vector space reflect the semantic similarity and relation strength between entities.

[0092] By traversing the semantic paths in the ontology model using a graph neural network (Graph Neural Network), the importance weights of each parameter node are calculated. The Graph Neural Network employs a graph attention network architecture, and the message passing mechanism allows nodes to aggregate information from their neighbors. The traversed semantic paths include synthetic paths from material composition to structural dimensions, mapping paths from structural dimensions to performance indicators, and causal paths from performance indicators to environmental impacts. The node state update formula in each layer of the Graph Neural Network includes the calculation of attention weights for neighboring nodes; neighboring nodes with larger attention coefficients contribute more to the state update of the central node. After multiple rounds of message passing, each node obtains a hidden state vector that incorporates global graph structure information. The importance weight of the node is decoded from the hidden state vector through a fully connected layer, and the importance weight reflects the centrality of the parameter node in the overall semantic network. Combining environmental impact factors from the life cycle assessment database, the environmental impact weights corresponding to the semantic content are derived. The life cycle assessment database contains environmental impact indicators such as global warming potential, acidification potential, and eutrophication potential for various materials and production processes. The matching process of environmental impact factors is based on the semantic associations defined in the ontology model. For example, the energy consumption attribute of a conductor material entity during its use stage is linked to the carbon emission impact factor during its usage stage. The calculation of environmental impact weights is a multi-step derivation process. First, the significance of basic environmental impact factors is adjusted according to the importance weights of parameter nodes. Then, the impact is propagated along the semantic path, ultimately generating a comprehensive environmental impact weight score. The environmental impact weight score is a normalized value used to quantify the relative contribution of each product specification parameter to the environmental impact throughout its entire life cycle.

[0093] A semantic parsing knowledge base with weighted coefficients is constructed and stored using a graph database. Nodes represent entity instances with weighted coefficients, and edges represent relation instances with weights. The weighted coefficients include the importance weight and environmental impact weight of the node. The schema layer of the knowledge base is consistent with the ontology model, while the instance layer contains actual data extracted from specific cable product specifications. The semantic parsing knowledge base supports complex graph query operations, such as finding the key parameters with the greatest impact on carbon emissions and analyzing the cascading environmental impacts of parameter changes. The knowledge base's update mechanism allows for the incorporation of new product specification data or updated life cycle assessment data, maintaining the timeliness of the semantic parsing results. The carbon footprint contribution distribution map is transformed into a multi-dimensional constrained space, with each dimension representing a life cycle stage. The number of dimensions in the multi-dimensional constrained space is consistent with the number of cable life cycle stages, typically including five orthogonal dimensions: raw material acquisition, manufacturing, transportation and distribution, installation and use, and end-of-life recycling. The coordinate value on each dimension represents the carbon footprint contribution of that stage, and the coordinate values ​​are normalized and constrained to be between zero and one. The quantitative data in the carbon footprint contribution distribution map are mapped to various dimensions of a multidimensional constraint space through linear transformation. The inter-stage flow relationships in the map are transformed into inter-dimensional constraints in the multidimensional constraint space. The geometry of the multidimensional constraint space is determined by the characteristics of the carbon footprint contribution distribution map and may exhibit convex polyhedra, star-shaped regions, or other complex shapes.

[0094] An algorithm for solving constraint satisfaction problems is applied to search for a feasible solution domain in the constraint space. This algorithm treats each lifecycle stage as a variable, with the domain of the variable being all feasible carbon footprint levels for that stage. Constraints include intra-stage and inter-stage constraints. Intra-stage constraints limit the range of values ​​for individual variables, while inter-stage constraints describe the relationships between variables, such as the correlation between raw material selection and production energy consumption. A backtracking search algorithm is used to solve the problem, recursively attempting to assign values ​​to variables and backtracking to the previous decision point when a constraint is violated. Arc consistency is used to simplify the problem before the search, reducing the size of the variable's domain and improving search efficiency by propagating constraints. The solution process outputs a set of variable assignment combinations that satisfy all constraints, with each combination corresponding to a feasible carbon footprint distribution pattern. A multi-objective optimization theory is used to establish a trade-off model between economics, feasibility, and environmental impact. This model includes three conflicting objective functions: the environmental impact objective function minimizes the total lifecycle carbon footprint; the economic objective function minimizes the total cost of the cable product; and the feasibility objective function maximizes the technical implementation difficulty score. The objective function is mathematically expressed using either the weighted sum method or the ε-constraint method. The weighted sum method transforms multiple objectives into a weighted sum of single objectives, while the ε-constraint method treats one objective as the primary objective and the others as constraints. The solution to the trade-off model is a Pareto optimal set, where each solution represents an optimal balance between environmental impact, economic efficiency, and feasibility under given conditions. No single solution is superior to another in all objectives.

[0095] Constraint propagation algorithms eliminate inconsistencies in the constraint space by leveraging local consistency of constraints to deduce global consistency of variables. Typical constraint propagation algorithms include AC-3 and PC-2. The algorithm maintains a constraint queue, continuously retrieving constraints from the queue for revision. Revision operations check if variable values ​​satisfy the constraints and remove values ​​from the variable's domain that do not. The constraint propagation process gradually narrows the variable's domain, sometimes even directly deriving unique values ​​for certain variables. When the variable's domain becomes empty, it indicates the existence of inconsistencies and no solution in the constraint space. Constraint propagation algorithms are used in conjunction with constraint satisfaction problem-solving algorithms, dynamically propagating constraints during the search process to identify and prune invalid search branches in advance. A set of carbon footprint calculation constraints is generated, including boundary conditions, trade-offs, and feasibility rules. Boundary conditions define the upper and lower limits of carbon footprint values ​​at each life cycle stage, typically derived from technological limits, regulatory standards, or historical data statistics. Trade-offs are described in the form of mathematical inequalities, such as the marginal relationship curve between unit carbon emission reduction cost and emission reduction. Feasibility rules are a set of logical judgment conditions used to exclude carbon footprint distribution schemes that are technically infeasible or economically unreasonable, such as rules that use recycled materials may increase production costs but reduce energy consumption during use. The set of carbon footprint calculation constraints is encoded in a standardized XML or JSON format for easy reading and parsing by subsequent optimization algorithms. Refer to Table 1, which shows the mapping relationship between the importance weights and environmental impact weights obtained after semantic parsing of the main cable specifications.

[0096] Table 1: Weight Mapping Table for Cable Specification Parameters

[0097] Parameter name Parameter type Importance weight Environmental impact weight conductor cross-sectional area Structural dimensions 0.85 0.76 Insulation layer thickness Structural dimensions 0.78 0.69 Sheath material density Material properties 0.72 0.81 Rated voltage Performance indicators 0.91 0.88 Operating temperature range Performance indicators 0.67 0.73 Conductivity of conductor materials Material properties 0.89 0.79

[0098] The ontological model of cable product technical parameters provides a structured knowledge framework for semantic parsing, and knowledge graph embedding technology enables the effective conversion of symbolic knowledge into numerical vectors. Graph neural networks traversing semantic paths can capture deep relationships between parameters, and the introduction of a life cycle assessment database establishes a quantitative link between technical parameters and environmental impact. The semantic parsing knowledge base integrates various weight coefficients generated during the parsing process, and the conversion from carbon footprint contribution distribution maps to a multi-dimensional constraint space transforms the evaluation results into a computable mathematical form. Constraint satisfaction problem-solving algorithms systematically explore feasible solutions in complex constraint spaces, and multi-objective optimization theory addresses the inherent conflicts between different optimization objectives. Constraint propagation algorithms improve the efficiency of constraint processing, and the carbon footprint calculation constraint set provides explicit input for the final optimization decision.

[0099] Example 5: A mixed-integer programming model is designed to transform the carbon footprint calculation constraints into a mathematical programming problem. The variables in the mixed-integer programming model include continuous and integer variables. Continuous variables represent the continuous values ​​of carbon emissions at each stage of the cable's lifecycle; for example, the energy consumption value during the manufacturing stage is a continuous variable. Integer variables are used to represent discrete decision choices; for example, the choice of raw material type is an integer variable, where a value of one represents choosing Class A conductor material, and a value of zero represents not choosing it. The set of carbon footprint calculation constraints is transformed into a set of constraints for the mixed-integer programming model. Each constraint corresponds to a mathematical inequality or equality. Boundary conditions are transformed into upper and lower bound constraints for variables, trade-offs are transformed into trade-off constraints between multiple objectives, and feasibility rules are transformed into logical constraints. The objective function is set to minimize the total carbon footprint over the entire lifecycle. The expression of the objective function is a weighted sum of carbon emissions at each stage, and the weight coefficients are derived from the environmental impact weights obtained through semantic parsing.

[0100] This paper applies the branch and bound algorithm to solve the discrete decision variables in a mixed integer programming model. The algorithm searches the solution space of the model through systematic enumeration and intelligent pruning. It begins by solving a relaxation problem, which ignores the integer requirement of variables, treating integer variables as continuous variables, thus obtaining a relaxation solution. The value of the relaxation solution provides a lower bound for the optimal value. The algorithm then performs a branching operation based on this relaxation solution. Each branching operation selects an integer variable whose value in the relaxation solution is not an integer. The algorithm creates two subproblems: one that forces the variable to zero, and the other that forces it to one. The branching process constructs a tree structure, where each node represents a subproblem. The bounding operation calculates the upper and lower bounds for each node. The upper bound is derived from the best feasible solution found so far, and the lower bound is derived from the value of the node's relaxation solution. When the lower bound of a node exceeds the current upper bound, the node is pruned, eliminating the need for further branching. This pruning operation significantly reduces the number of nodes to be explored. The branch and bound algorithm continues until the entire tree has been searched or a preset termination condition is met.

[0101] Interior-point method is employed to handle continuous variable optimization problems. Starting from an initial point within the feasible region, the interior-point method approaches the optimal solution along the central path. It constructs a barrier function, transforming the constrained optimization problem into a series of unconstrained optimization problems. The barrier function tends to infinity at the feasible region boundary, forcing the search path to remain within the feasible region. In each iteration, the interior-point method solves a system of Newton's equations, calculates the search direction and step size, and the iteration points gradually approach the optimal solution. The interior-point method is computationally efficient in handling large-scale continuous optimization problems, with a fast convergence speed, making it suitable for solving cable carbon footprint models with a large number of continuous variables. The interior-point method works in conjunction with the branch and bound algorithm; the branch and bound algorithm handles discrete decisions, while the interior-point method handles continuous optimization under given discrete decisions. Sensitivity analysis is used to verify the robustness of the solution. Sensitivity analysis studies how the optimal solution of the mixed-integer programming model changes with the input parameters. The input parameters include coefficients in the carbon footprint calculation constraints, such as carbon emission factors, cost coefficients, and technical parameters. Sensitivity analysis calculates the allowable range of variation for these parameters; within this allowable range, the fundamental solution of the optimal solution remains unchanged. Shadow price analysis reveals the impact of each unit increase in the right-hand side of the constraints on the objective function value. ReducedCost analysis shows how much a non-basic variable needs to change to be included in the basic variable set. Sensitivity analysis reports provide quantitative indicators of optimal solution stability, helping decision-makers understand the reliability of the model solution and identify the key parameters that have the greatest impact on the outcome.

[0102] A carbon footprint determination scheme is generated, containing the optimal solution set and corresponding decision variables. This scheme is a structured output document with two parts: the optimal solution set (multiple non-dominated solutions on the Pareto optimal frontier, each corresponding to a carbon footprint level, cost level, and feasibility score) and the corresponding decision variable values. These decision variables detail the specific measures required to achieve each optimal solution, such as conductor material selection, insulation thickness setting, and production process parameter adjustments. The solution set is arranged in ascending order of environmental impact, facilitating selection by decision-makers based on actual needs. The scheme is encoded into a standardized data exchange format, JSON-LD, which adds semantic annotations to JSON and supports associated data standards. The encoding process maps various data elements in the scheme to a predefined JSON schema. The optimal solution set is encoded as an array of objects, each containing three attributes: carbon footprint value, cost value, and feasibility score. Decision variables are encoded as nested object structures, with top-level objects representing lifecycle stages and lower-level objects representing specific parameter settings for each stage. Standardized data exchange formats ensure the interoperability of carbon footprint determination schemes across different software platforms, facilitating subsequent system reading and parsing.

[0103] Digital signature technology ensures data integrity and immutability. Based on asymmetric encryption algorithms, it uses RSA to generate key pairs: a private key for signing and a public key for verification. The signing process first hashes the JSON-LD data of the carbon footprint determination scheme to generate a fixed-length message digest. Then, the message digest is encrypted using the private key to generate the digital signature. The digital signature is stored or transmitted along with the original data. Verification uses the corresponding public key to decrypt the digital signature to obtain the message digest, while simultaneously recalculating the hash value of the original data and comparing the two hash values. Digital signature technology provides identity authentication and data integrity verification; any tampering with the data will result in verification failure. A carbon footprint notarization system based on distributed ledger technology is constructed. This system uses a consortium blockchain architecture, with network nodes maintained by authorized members such as cable manufacturers, regulatory agencies, and certification centers. The carbon footprint notarization system defines a smart contract that specifies the rules and procedures for uploading carbon footprint data to the blockchain. When a new carbon footprint determination scheme is generated, the notarization method of the smart contract is invoked. The evidence storage method packages the JSON-LD data of the carbon footprint determination scheme and the corresponding digital signature into a transaction. After the transaction is verified by the node consensus mechanism, it is packaged into a block, and the blocks are linked into a chain through hash pointers. The decentralized, immutable, and traceable characteristics of distributed ledger technology provide a reliable evidence storage environment for carbon footprint data.

[0104] Carbon footprint data is pushed to relevant regulatory platforms via an application programming interface (API). The API adopts a RESTful architecture, accepting HTTP POST requests with the request body containing the JSON-LD encoded carbon footprint determination scheme. The API implements authentication and access control, allowing only authorized clients to call it. It returns standardized status codes and response messages, such as 200 for success and 400 for an error. The regulatory platform sends a confirmation receipt upon receiving the data. The API supports asynchronous communication to avoid network latency blocking the main business process. Real-time verification and traceability of carbon footprint data are implemented. Real-time verification allows any stakeholder to verify the integrity and authenticity of the data through a query interface. The verifier submits a unique identifier for the carbon footprint data; the system retrieves the corresponding record from the distributed ledger, parses the original data and digital signature, and automatically executes the verification process. The traceability function, based on the immutability of the distributed ledger, allows tracing historical version changes of the carbon footprint determination scheme. Each update generates a new record, recording the timestamp and operator information. Real-time verification and traceability enhance the transparency and credibility of carbon footprint data. Mixed-integer programming models formalize the complex carbon footprint optimization problem into a computable mathematical problem. Branch-and-bound algorithms systematically handle discrete decisions within the model, while interior-point methods efficiently solve continuous subproblems. Sensitivity analysis provides decision-makers with crucial information about solution stability, and carbon footprint determination schemes transform mathematical solutions into concrete action guidelines. JSON-LD encoding enhances machine readability and semantic interoperability, while digital signature technology is the cornerstone of ensuring data authenticity. Distributed ledger technology constructs a trusted data storage infrastructure, and application programming interfaces enable seamless data flow between systems. Real-time verification and traceability meet regulatory and auditing requirements.

[0105] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for determining the carbon footprint of a cable throughout its entire life cycle, characterized in that, The method calculates the carbon footprint using data from each stage of the integrated cable's lifecycle, specifically including the following operations: Carbon emission data for each stage of the cable's life cycle—from raw material acquisition, production and manufacturing, transportation and distribution, installation and use, to scrapping and recycling—is obtained through a pre-set data acquisition system. A multi-dimensional characterization model of carbon footprint features was established using the acquired carbon emission data. A life cycle stage adaptive assessment is performed on the multi-dimensional characterization model to generate a carbon footprint contribution distribution map; The semantic content contained in the cable product specification parameters is analyzed, the environmental impact weights corresponding to the semantic content are derived, and the carbon footprint contribution distribution map is combined to form carbon footprint calculation constraints. Based on the aforementioned carbon footprint calculation constraints, an optimization algorithm is used to generate a carbon footprint determination scheme; Implement the carbon footprint output process according to the carbon footprint determination scheme; The process of generating a carbon footprint determination scheme using an optimization algorithm based on the carbon footprint calculation constraints includes the following steps: A mixed-integer programming model is designed to transform the carbon footprint calculation constraints into a mathematical programming problem; the branch and bound algorithm is applied to solve the discrete decision variables in the model; the interior point method is used to handle the continuous variable optimization problem; the robustness of the solution is verified through sensitivity analysis; and a carbon footprint determination scheme containing the optimal solution set and corresponding decision variables is generated. The process of establishing a multi-dimensional characterization model of carbon footprint features using the acquired carbon emission data includes the following steps: The structured carbon emission dataset is input into a spatiotemporal graph convolutional network to construct a feature graph with life cycle stages as nodes and material and energy flows as edges. Learn the implicit carbon association features between nodes through a multi-layer graph attention mechanism; A feature pyramid network is used to extract multi-scale carbon footprint features, including macro-level industrial chain features and micro-level process features; Using bidirectional long short-term memory networks to capture the dynamic evolution of carbon footprint over time; The feature maps, implicit carbon correlation features, macro-industrial chain level features, micro-process level features, and dynamic evolution laws are fused using tensors to generate a multi-dimensional representation model with spatiotemporal correlation. The lifecycle-stage adaptive assessment of the multidimensional representation model includes the following steps: We construct an adaptive evaluation framework based on deep reinforcement learning and design a reward function with the goal of minimizing carbon footprint. The agent is trained and evaluated using a policy gradient algorithm, enabling it to learn to identify key influencing factors at each stage of its life cycle. The Monte Carlo tree search algorithm was used to explore potential optimization paths in a multi-dimensional representation model. Generate an adaptive assessment report that includes stage contribution, sensitivity indicators, and optimization priorities; Based on the priority indicators in the assessment report, a carbon footprint contribution distribution map was constructed.

2. The method for determining the carbon footprint of a cable throughout its entire life cycle according to claim 1, characterized in that, The process of acquiring carbon emission data for each stage of the cable's entire lifecycle—including raw material acquisition, manufacturing, transportation and distribution, installation and use, and end-of-life recycling—through a pre-set data acquisition system includes the following steps: Build a distributed edge computing node network and deploy dedicated data acquisition agents at each stage of the lifecycle; The dedicated data acquisition agent captures energy consumption monitoring data during the raw material mining stage, process parameter flow data during the production and manufacturing stage, and path trajectory data during the transportation stage in real time. A dynamic time warping algorithm is used to align the time axis of time-series data acquired asynchronously from multiple sources. An attention-based data quality assessment model was applied to perform anomaly detection and cleaning on the collected data. The cleaned data is categorized and stored in a time-series database according to life cycle stages, forming a structured carbon emission dataset.

3. The method for determining the carbon footprint of a cable throughout its entire life cycle according to claim 1, characterized in that, The semantic content contained in the analytical cable product specification parameters includes the following steps: Establish an ontological model of the technical parameters of cable products and define the semantic relationships between material composition, structural dimensions, and performance indicators; The specification parameters of text format are transformed into vector representations by applying knowledge graph embedding technology; The importance weights of each parameter node are calculated by traversing the semantic path in the ontology model using a graph neural network. By combining environmental impact factors in the life cycle assessment database, the environmental impact weights corresponding to semantic content are derived. Construct a semantic parsing knowledge base that includes weight coefficients.

4. The method for determining the carbon footprint of a cable throughout its entire life cycle according to claim 1, characterized in that, The steps involved in forming the carbon footprint calculation constraints by combining the carbon footprint contribution distribution map are as follows: The carbon footprint contribution distribution map is transformed into a multi-dimensional constrained space, with each dimension representing a life cycle stage. An algorithm for solving constraint satisfaction problems is applied to search for a feasible solution domain in the constraint space. A trade-off model between economic efficiency, feasibility, and environmental impact is established using multi-objective optimization theory; Inconsistent regions in the constraint space are eliminated through constraint propagation algorithms. Generate a set of constraints for carbon footprint calculation that includes boundary conditions, trade-offs, and feasibility rules.

5. The method for determining the carbon footprint of a cable throughout its entire life cycle according to claim 1, characterized in that, The process of implementing carbon footprint output based on the carbon footprint determination scheme includes the following steps: Encode the carbon footprint determination scheme into a standardized data exchange format; Digital signature technology is used to ensure data integrity and immutability; Construct a carbon footprint evidence storage system based on distributed ledger technology; Carbon footprint data is pushed to relevant regulatory platforms via application programming interfaces (APIs). Enables real-time verification and traceability of carbon footprint data.

6. The method for determining the carbon footprint of a cable throughout its entire life cycle according to claim 2, characterized in that, The construction of the distributed edge computing node network includes the following steps: Deploy IoT gateways with edge computing capabilities in all aspects of the cable industry chain; Configure a lightweight data acquisition protocol to enable communication between devices; Differential privacy technology is used to de-identify the collected data; Establish a secure transmission channel between edge nodes and the cloud platform; Implement an automatic scaling up / down and failover mechanism for the data acquisition network.

7. The method for determining the carbon footprint of a cable throughout its entire life cycle according to claim 1, characterized in that, The construction of the adaptive evaluation framework based on deep reinforcement learning includes the following steps: Design a Markov decision process that includes a state space, an action space, and a reward function; A deep reinforcement learning network is constructed using an actor-critic architecture; Improve training efficiency through experience playback mechanisms; Gradually increase the complexity of assessment by applying curriculum learning strategies; Establish a closed loop for online evaluation and feedback optimization of model performance.

Images