Improvement potential quantification method for carbon emission in mining life cycle process

Abstract

The invention discloses an improvement potential quantification method for carbon emission in a mining life cycle process, and relates to the technical field of mine carbon emission management, and the method comprises the steps: collecting real-time sensor data, equipment operation logs, process parameters and unstructured documents of each stage in the full life cycle of a gold mine, and constructing a dynamically updated multi-source data pool; based on the multi-source data pool, carbon emission entities and a carbon emission entity semantic relationship are extracted, and a carbon footprint knowledge graph is constructed; performing multi-objective optimization on a collaborative decision result, solving a non-dominated solution set, screening an optimal equilibrium solution, and generating an optimization strategy sequence and corresponding expected carbon emission; and inputting the optimization strategy sequence into a mine production scheduling process for execution, monitoring actual carbon emission data in real time, comparing the actual carbon emission data with expected carbon emission, and correcting the dynamic carbon emission reference model through a comparison result. According to the method, the carbon footprint knowledge graph is constructed, so that the semantic understanding and correlation analysis of the whole-process carbon emission of the mine from exploration to pit closing are realized.

Description

Technical Field

[0001] This invention relates to the field of carbon emission management technology in mining, and in particular to a method for quantifying the potential for improving carbon emissions throughout the mining lifecycle. Background Technology

[0002] With the advancement of the "dual carbon" goals, the mining sector is increasingly emphasizing the refined management of carbon emissions. Currently, carbon footprint accounting technology based on the product lifecycle management framework is relatively mature, enabling post-event statistical analysis of carbon emissions at each stage of mine exploration, construction, production, and closure. Simultaneously, the application of technologies such as the Internet of Things and big data has made real-time data collection and monitoring of mine production processes possible, providing a data foundation for dynamic carbon management; some advanced methods are also beginning to explore the use of technologies such as knowledge graphs to statically model carbon flow relationships.

[0003] However, most existing technical solutions treat carbon emissions at each stage of the life cycle as isolated accounting, lacking in-depth exploration of the dynamic correlation and collaborative optimization mechanism of carbon flow throughout the entire life cycle. The core shortcoming is that it fails to build a closed-loop optimization that can map the semantic relationship of carbon footprint in real time and support distributed decision agents for collaborative decision-making and dynamic optimization. This leads to carbon management strategies often lagging behind actual production dynamics, making it difficult to achieve the globally optimal emission reduction effect. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for quantifying the potential for improving carbon emissions throughout the mining lifecycle to address the problem of low efficiency in carbon emission optimization caused by isolated lifecycle stages and lack of dynamic collaborative decision-making in existing mining carbon management methods.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for quantifying the potential for improving carbon emissions throughout the lifecycle of a mining operation. The method includes: collecting real-time sensor data, equipment operation logs, process parameters, and unstructured documents at each stage of the gold mine's lifecycle to construct a dynamically updated multi-source data pool; extracting carbon emission entities and their semantic relationships based on the multi-source data pool to construct a carbon footprint knowledge graph; constructing a dynamic carbon emission benchmark model, mapping the carbon footprint knowledge graph to the dynamic carbon emission benchmark model to generate a carbon flow network, and identifying key impact nodes and associated process paths in the carbon flow network that affect global carbon emissions; assigning decision agents to each key impact node and associated process path, with all decision agents forming a distributed decision network; enabling collaborative decision-making among the decision agents in the distributed decision network through a virtual carbon quota trading mechanism to generate collaborative decision results; performing multi-objective optimization on the collaborative decision results, solving for the non-dominated solution set, and selecting the optimal equilibrium solution to generate an optimization strategy sequence and corresponding expected carbon emissions; inputting the optimization strategy sequence into the mine's production scheduling process for execution, monitoring actual carbon emission data in real time and comparing it with expected carbon emissions, and correcting the dynamic carbon emission benchmark model based on the comparison results.

[0007] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in this invention, the steps for constructing a dynamically updated multi-source data pool are as follows: The real-time sensor data, equipment operation logs, process parameters and unstructured documents of each stage of the gold mine's entire life cycle are converted into standardized time-series data through IoT protocols, data interfaces and document parsers and stored in the established time-series database. Value quantification analysis is performed on time-series data standardized by information entropy theory to generate value-labeled data streams. The value-labeled data streams are prioritized and sorted according to a predefined data value weight table to form hierarchical data streams. The hierarchical data stream is mapped to a unified spatiotemporal coordinate system, a dynamic calibration algorithm is used for quality verification, and the time series database is updated through a dynamic update mechanism to build a dynamically updated multi-source data pool.

[0008] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in this invention, the steps for constructing the carbon footprint knowledge graph are as follows: Based on text data from a multi-source data pool, a multi-task learning model is used to simultaneously identify four types of entities and their semantic relationships: equipment, process, energy consumption, and carbon emissions, and generate a set of entity semantic relationship triples. Align the set of entity semantic relation triples with the spatiotemporal coordinate system, and use graph attention network to calculate the spatiotemporal association strength between entities to form a knowledge graph pattern layer; Based on the knowledge graph pattern layer, a carbon footprint knowledge graph is constructed by using entities as nodes, entity semantic relationships as edges, and spatiotemporal correlation strength as the weight attribute of the edges.

[0009] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in this invention, the steps for constructing a dynamic carbon emission benchmark model are as follows: A data fusion layer is built based on a multi-source data pool, a graph feature layer is built based on a carbon footprint knowledge graph, and a network topology layer is built based on the LCA life cycle assessment method. By connecting the data fusion layer, graph feature layer, and network topology layer in real time through a data flow pipeline, a dynamic carbon emission benchmark model is constructed.

[0010] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in this invention, the steps for generating the carbon flow network are as follows: The data fusion layer based on the dynamic carbon emission benchmark model provides real-time monitoring data and extracts the spatiotemporal correlation strength weight matrix by combining the entity semantic relationships in the carbon footprint knowledge graph. Based on the spatiotemporal correlation strength weight matrix, carbon flow features are extracted through the graph feature layer, and the carbon flow weight of each edge in the carbon flow network is calculated. Based on carbon flow weights, the entity nodes and semantic relationship edges in the carbon footprint knowledge graph are mapped to carbon flow nodes and carbon flow weighted edges in the carbon flow network through the network topology layer, thus generating the carbon flow network.

[0011] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in this invention, the steps for identifying key impact nodes and associated process paths in the carbon flow network that affect global carbon emissions are as follows: Based on the carbon flow network, the carbon load attributes of carbon flow nodes and the weight attributes of carbon flow weighted edges are extracted, the influence value of each carbon flow node is calculated, and carbon flow nodes whose influence values ​​exceed the preset influence value threshold are identified as key influence nodes. The process paths associated with key influencing nodes are extracted in the carbon flow network using a path traversal algorithm.

[0012] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions during the mining life cycle as described in this invention, a decision agent is initialized for each key impact node based on key impact nodes and associated process paths. By connecting all decision agents through a carbon flow network topology, a distributed decision network with a hierarchical communication structure is constructed.

[0013] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in this invention, the steps for generating collaborative decision-making results are as follows: Each decision-making agent is assigned an initial virtual carbon allowance and trading strategy via smart contracts; Based on the allocated virtual carbon allowances and trading strategies, multiple rounds of allowance trading are conducted through a two-way auction mechanism, and the trading strategies are dynamically updated during the trading process; When quota trading reaches equilibrium, collaborative decision-making results are generated based on the current trading strategy.

[0014] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in this invention, the steps for generating the optimization strategy sequence and the corresponding expected carbon emissions are as follows: The collaborative decision-making results are mapped to a multi-dimensional objective space to construct a multi-objective optimization framework. Based on the multi-objective optimization framework, the Pareto front contraction algorithm is used to solve the non-dominated solution set. The optimal equilibrium solution is obtained by screening the non-dominated solution set through the fuzzy preference ensemble method. The optimal equilibrium solution is then converted into an optimization strategy sequence, and the corresponding expected carbon emissions are generated through simulation using a dynamic carbon emission benchmark model.

[0015] As a preferred embodiment of the method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in this invention, the steps for correcting the dynamic carbon emission benchmark model through comparison results are as follows: The optimization strategy sequence is decomposed into control commands input into the mine production scheduling process, and actual carbon emission data is collected in real time through sensors. The actual carbon emission data is compared with the expected carbon emission to generate comparison results. A hybrid filtering algorithm is used to perform data fusion and error analysis on the comparison results to generate correction parameters. The dynamic carbon emission baseline model is modified using correction parameters to form a closed-loop iterative optimization.

[0016] The beneficial effects of this invention are as follows: By constructing a carbon footprint knowledge graph, semantic understanding and correlation analysis of carbon emissions throughout the entire process of mine exploration to closure are achieved; within the framework of product lifecycle management, the carbon footprint knowledge graph intuitively reveals the key sources and transmission paths of carbon flows across lifecycles, providing a unified and semantically rich knowledge foundation for carbon flow network generation, key node identification, and distributed collaborative decision-making based on the product lifecycle management concept; by associating and integrating unstructured document data from each stage with real-time sensor data, the dynamic carbon emission benchmark model is ensured to possess both profound full lifecycle process background knowledge and the ability to dynamically respond to changes in actual production status, thereby laying a precise and reasonable data foundation for the entire carbon improvement potential quantification method. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a method for quantifying the potential for improving carbon emissions throughout the lifecycle of mining operations.

[0019] Figure 2 A flowchart for constructing a dynamically updated multi-source data pool.

[0020] Figure 3 A flowchart for constructing a carbon footprint knowledge graph.

[0021] Figure 4 A flowchart for constructing a dynamic carbon emission benchmark model and generating a carbon flow network. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for quantifying the potential for improvement in carbon emissions throughout the life cycle of mining operations, comprising the following steps: S1. Collect real-time sensor data, equipment operation logs, process parameters and unstructured documents at each stage of the gold mine's entire life cycle, and build a dynamically updated multi-source data pool; based on the multi-source data pool, extract carbon emission entities and semantic relationships between carbon emission entities, and build a carbon footprint knowledge graph. S1.1: Through IoT protocols, data interfaces, and document parsers, real-time sensor data, equipment operation logs, process parameters, and unstructured documents at each stage of the gold mine's entire lifecycle are converted into standardized time-series data and stored in the established time-series database. Specifically, a time-series data storage engine supporting high-concurrency writes and time-range queries is selected. A unified data table structure is defined, including timestamp fields, data source type fields, source identifier fields, and numerical or text content fields, to establish a time-series database. Based on the data characteristics of each stage in the gold mine's entire lifecycle, corresponding data retention strategies and compression rules are preset to ensure storage efficiency and query performance under long-term operation. Real-time sensor data from each stage in the gold mine's entire lifecycle is collected through IoT protocols, and sensors from different manufacturers, types, and communication methods are uniformly connected using standardized protocols. Equipment operation logs and process parameters are obtained through data interfaces, and the raw records from programmable logic controllers, distributed controllers, and enterprise resource planning are aligned and time-series marked. Unstructured documents are read through a document parser, and carbon emission-related events, operation descriptions, and time information are extracted according to preset text extraction rules and assigned unified timestamps. The four types of data are aligned according to a unified time benchmark and converted into standardized time-series data with time tags, data source identifiers, and semantic types. The standardized time-series data is written into the established time-series database to achieve chronological organization, efficient storage, and fast retrieval.

[0026] It should be noted that the preset data retention strategy and compression rules refer to determining the retention duration, archiving method and time granularity compression method for different types of data in advance, based on the usage frequency, importance and storage cost of data at each stage of the gold mine's entire life cycle, before establishing the time-series database. Pre-defined text extraction rules refer to the matching logic and structured mapping criteria that are predefined to identify and extract specific semantic content based on carbon emission-related keywords, event patterns, and time expressions before the document parser processes unstructured documents.

[0027] It should be noted that each stage refers to different links in the entire life cycle of gold mine mining, including exploration, construction, mining, mineral processing, smelting, mine closure and ecological restoration. Real-time sensor data refers to the instantaneous values ​​that reflect the status of equipment or the environmental conditions, continuously collected by various sensors (such as energy consumption, gas, displacement, and temperature sensors) deployed at the gold mine site. Equipment operation logs refer to records automatically generated during the operation of gold mine production equipment, including operation and maintenance information such as start-up and shutdown times, running time, fault alarms, and load status; Process parameters refer to the key technical indicators that control and affect the production process in a specific mining or processing process, such as crushing particle size, flotation reagent dosage, grinding concentration, and recovery rate. Unstructured documents refer to text-based materials that cannot be directly parsed using fixed fields, such as design specifications, environmental impact assessment reports, operating procedures, inspection records, and meeting minutes—documents that exist in natural language.

[0028] S1.2: Value quantification analysis is performed on time-series data standardized by information entropy theory to generate value-labeled data streams; Specifically, standardized time-series data are grouped according to data source type and source identifier, with each group corresponding to an independent data item; the values ​​of continuous data items are discretized and divided into a finite number of mutually exclusive intervals; within a preset time window, the frequency of occurrence of each discrete value of the data item is counted, and its occurrence frequency is calculated; based on the information entropy theory, the frequency distribution of the values ​​is transformed into information entropy values; the contribution of data items to the overall carbon emission characterization is distinguished according to the level of information entropy values, with higher information entropy values ​​indicating richer information and greater value for the data item; a corresponding information entropy value is attached to each standardized time-series data as a value marker, forming a data stream with value markers, i.e., a value-marked data stream.

[0029] Based on information entropy theory, the frequency distribution of the values ​​is transformed into information entropy values, expressed as: ; ; In the formula, This represents the information entropy value of a data item within a time window; This represents the total number of different discrete values ​​that a data item takes within a time window; Indicates the index of different discrete values; Indicates the first The frequency of each discrete value occurring within a time window; Indicates a base-2 pair Take the logarithm. Indicates the first The number of times a discrete value appears within a time window; This represents the total number of observations recorded for a data item within the time window.

[0030] It should be noted that the preset time window refers to the length of a continuous time period or time interval that is determined in advance for statistically analyzing the value distribution of a certain data item before the information entropy calculation is performed on the standardized time series data.

[0031] S1.3: Prioritize the value-marked data streams according to a predefined data value weight table to form hierarchical data streams; Specifically, based on a predefined data value weight table, the information entropy value of each record in the value-marked data stream is compared, and the information entropy value is mapped to the corresponding data value weight level. According to the data value weight level, all records in the value-marked data stream are sorted, with records of higher weight levels ranked first. The sorted records are grouped according to weight level to form subsequences of different priorities, which are then combined into a hierarchical data stream.

[0032] It should be noted that the predefined data value weight table is set based on the degree of influence of different data source types on carbon emission characterization at each stage of the gold mine's entire life cycle. The specific setting steps are as follows: Based on the gold mine's process flow and carbon emission source distribution, key carbon emission links are identified, including mining, crushing, grinding, flotation, and tailings disposal. For each key carbon emission link, the data source types directly associated with it are listed, including real-time sensor data, equipment operation logs, process parameters, and unstructured documents. Based on historical operating experience and carbon accounting standards, the information representativeness, timeliness, and irreplaceability of each data source type in the corresponding link are comprehensively evaluated. The evaluation results are converted into high, medium, and low data value weights and bound to specific data source types and source identifiers to form a predefined data value weight table.

[0033] S1.4: Map the hierarchical data stream to a unified spatiotemporal coordinate system, use a dynamic calibration algorithm for quality verification, and update the time series database through a dynamic update mechanism to build a dynamically updated multi-source data pool; Specifically, the timestamps and spatial identifiers (such as equipment installation locations) contained in each record of the hierarchical data stream are used as a benchmark and mapped to a unified spatiotemporal coordinate system, aligning data from different data sources at the same time and spatial location. A dynamic calibration algorithm is used to correct outliers, missing segments, or time offsets in the hierarchical data stream based on the changing trends of similar data items at adjacent time points, the physical reasonable range, and multi-source consistency rules. Through a dynamic update mechanism, the quality-verified hierarchical data stream is continuously written into the time-series database, overwriting or appending existing records at the corresponding time and spatial location. Ultimately, a dynamically updated multi-source data pool is formed, covering all stages of the gold mine's entire life cycle and integrating real-time sensor data, equipment operation logs, process parameters, and unstructured documents.

[0034] It should be noted that the dynamic update mechanism refers to the method of continuously adding or overwriting data records at corresponding spatiotemporal locations in the established time-series database based on newly arrived, quality-verified graded data streams during the entire life cycle of a gold mine, in order to keep the content of the multi-source data pool synchronized with the actual on-site status.

[0035] S1.5: Based on text data in a multi-source data pool, a multi-task learning model is used to simultaneously identify four types of entities and their semantic relationships: equipment, process, energy consumption, and carbon emissions, and generate a set of entity semantic relationship triples. Specifically, unstructured documents are extracted from a dynamically updated multi-source data pool, and records containing text data are selected as the raw corpus. The raw corpus is then fed into a multi-task learning model. The multi-task learning model simultaneously performs four types of entity boundary and type labeling on each piece of text data, and determines whether there are semantic relationship types between the identified entities, including "the equipment belongs to the process", "the process consumes energy", and "energy consumption generates carbon emissions". The identified entities and their semantic relationships are organized in the format of "head entity-relationship-tail entity" to form a set of entity semantic relationship triples.

[0036] It should be noted that the pre-training process of the multi-task learning model is based on a large-scale historical text corpus from gold mines. This historical text corpus consists of unstructured documents, technical manuals, environmental impact assessment reports, and operating procedures. During the pre-training phase, a masked language modeling task and an entity boundary prediction task are jointly used to drive model parameter updates, enabling the model to learn the contextual representation of words in the professional context of gold mines. Supervised training of four named entity recognition sub-tasks and entity semantic relationship classification sub-tasks is carried out simultaneously on a labeled seed dataset. The four named entity categories include equipment, process, energy consumption, and carbon emissions, and the entity semantic relationship types include semantic relationships such as "equipment belongs to process," "process consumes energy," and "energy consumption generates carbon emissions." Through an architecture that shares a bottom-level encoding layer and an independent top-level classification layer, the multi-task learning model simultaneously optimizes entity recognition accuracy and relationship classification accuracy during training, ultimately forming a pre-trained multi-task learning model with synchronous recognition capabilities.

[0037] S1.6: Align the set of entity semantic relation triples with the spatiotemporal coordinate system, and use a graph attention network to calculate the spatiotemporal association strength between entities to form a knowledge graph pattern layer; Specifically, each entity in the entity semantic relation triple set is associated with its corresponding timestamp and spatial location throughout the entire life cycle of the gold mine, thereby aligning the entity semantic relation triple set with a unified spatiotemporal coordinate system. Based on the aligned triples and their spatiotemporal attributes, a graph attention network is used to calculate the spatiotemporal association strength between entities. Finally, with entities as nodes and semantic relations with spatiotemporal association strength as edges, a knowledge graph pattern layer is formed.

[0038] The spatiotemporal association strength between entities is calculated using a graph attention network, expressed as follows: ; In the formula, Representing entities With entity The spatiotemporal correlation strength between them; Indicates the time decay coefficient; Representing entities With entity The absolute value of the time difference between them in a unified spacetime coordinate system; Indicates the spatial attenuation coefficient; Representing entities With entity Euclidean distance between corresponding spatial locations.

[0039] It should be noted that the time decay coefficient It is set based on the typical time intervals between different activities in the gold mining process and the rate at which they affect carbon emissions, aiming to reflect the degree to which the strength of the correlation between entities decreases rapidly as the time difference increases; the example value is 0.1. Spatial attenuation coefficient It is set based on the average distance between different facilities or activity sites within a gold mine and its correlation with their contribution to carbon emissions, and is used to adjust for the rate at which the inter-entity correlation weakens due to increased spatial distance; the example value is 0.005.

[0040] S1.7: Based on the knowledge graph pattern layer, entities are used as nodes, semantic relationships of entities are used as edges, and spatiotemporal correlation strength is used as the weight attribute of the edges to construct a carbon footprint knowledge graph; Specifically, based on the knowledge graph pattern layer, four types of entities—equipment, process, energy consumption, and carbon emissions—are used as nodes in the graph. The semantic relationships in the entity semantic relationship triples are used as edges connecting the corresponding nodes, and the spatiotemporal correlation strength is used as the weight attribute of the edges, forming a structured graph data composed of nodes, edges, and their weight attributes. The structured graph data fully preserves the entity type, semantic relationship type, and spatiotemporal correlation strength of the edges, thus constructing a carbon footprint knowledge graph.

[0041] S2. Construct a dynamic carbon emission benchmark model, map the carbon footprint knowledge graph to the dynamic carbon emission benchmark model, generate a carbon flow network, and identify the key impact nodes and associated process paths in the carbon flow network that affect global carbon emissions. S2.1: A data fusion layer is built based on a multi-source data pool, a graph feature layer is built based on a carbon footprint knowledge graph, and a network topology layer is built based on the LCA life cycle assessment method; Specifically, based on real-time sensor data, equipment operation logs, process parameters, and unstructured documents aligned to a unified spatiotemporal coordinate system from a multi-source data pool, the original observation records are organized according to time sequence and spatial location to form a data fusion layer covering all stages of the gold mine's entire life cycle. Based on the structured graph data in the carbon footprint knowledge graph, with equipment, processes, energy consumption, and carbon emissions as nodes, entity semantic relationships as edges, and spatiotemporal correlation strength as edge weight attributes, node type, edge type, and edge weight attributes are extracted as feature representations to form a graph feature layer. Based on the mine life cycle stage division, process boundaries, and upstream and downstream material and energy flow relationships defined in the LCA life cycle assessment method, the specific production activities in each stage are mapped as network nodes, and the material flow, energy flow, and carbon emission flow are mapped as directed connections to construct a network topology layer reflecting the material and energy transfer paths throughout the entire life cycle.

[0042] S2.2: Connect the data fusion layer, graph feature layer and network topology layer in real time through the data flow pipeline to build a dynamic carbon emission benchmark model; Specifically, through the data flow pipeline, real-time sensor data, equipment operation logs, process parameters, and unstructured documents organized according to unified spatiotemporal coordinates in the data fusion layer are aligned with the structured graph data in the graph feature layer, which uses equipment, processes, energy consumption, and carbon emissions as nodes, entity semantic relationships as edges, and spatiotemporal correlation strength as edge weight attributes, by aligning them with timestamps and spatial locations. At the same time, the directed connections in the network topology layer, which reflect the material and energy transfer paths throughout the entire life cycle and are constructed based on the LCA life cycle assessment method, are matched with the aligned data at the stage and activity granularity. Within the data flow pipeline, the content of the three layers is aggregated at the same spatiotemporal granularity, using time windows as units, to form a joint data fragment containing original observations, semantic correlation features, and life cycle process structure. The joint data fragment is continuously updated and serves as the basic component of the dynamic carbon emission benchmark model.

[0043] S2.3: The data fusion layer based on the dynamic carbon emission benchmark model provides real-time monitoring data and extracts the spatiotemporal correlation strength weight matrix by combining the entity semantic relationships in the carbon footprint knowledge graph. Specifically, based on the real-time monitoring data provided by the data fusion layer of the dynamic carbon emission benchmark model, the equipment, processes, energy consumption and carbon emission-related observations contained therein are identified, and the corresponding nodes of the relevant observations are matched in the carbon footprint knowledge graph; according to the entity semantic relationship edges connecting the corresponding nodes in the carbon footprint knowledge graph, the spatiotemporal correlation strength attached to each edge is extracted as a weight value; all involved nodes are arranged in a preset order, and according to whether there is an entity semantic relationship between each pair, the corresponding spatiotemporal correlation strength is filled into the corresponding position of the matrix, and the remaining positions are set to empty or zero, forming a spatiotemporal correlation strength weight matrix.

[0044] It should be noted that the preset order refers to the order of nodes determined in advance based on the typical occurrence logic of the four types of entities—equipment, process, energy consumption, and carbon emissions—in the entire life cycle of a gold mine or the process sequence defined by the LCA life cycle assessment method, before constructing the spatiotemporal correlation strength weight matrix.

[0045] S2.4: Based on the spatiotemporal correlation strength weight matrix, carbon flow features are extracted through the graph feature layer, and the carbon flow weight of each edge in the carbon flow network is calculated. Specifically, based on the entity pairs indicated by the non-empty positions in the spatiotemporal correlation strength weight matrix and their corresponding spatiotemporal correlation strength, the entity semantic relationship types and edge weight attributes between entity pairs are extracted from the graph feature layer; combined with the carbon emission factor and material and energy flow direction defined in the LCA life cycle assessment method, each entity semantic relationship edge is mapped to a carbon flow path; based on the strength value of the edge in the spatiotemporal correlation strength weight matrix, the corresponding carbon flow path is assigned a corresponding carbon flow weight, forming the carbon flow weight of each edge in the carbon flow network.

[0046] S2.5: Based on carbon flow weights, the entity nodes and semantic relationship edges in the carbon footprint knowledge graph are mapped to carbon flow nodes and carbon flow weighted edges in the carbon flow network through the network topology layer, thereby generating the carbon flow network; Specifically, based on carbon flow weights, the four types of entity nodes in the carbon footprint knowledge graph—equipment, process, energy consumption, and carbon emissions—are mapped to carbon flow nodes in the carbon flow network according to their corresponding roles in the LCA life cycle assessment method. The semantic relationship edges connecting entities in the carbon footprint knowledge graph are combined with their corresponding carbon flow weights to convert them into weighted edges with weight values ​​in the carbon flow network. Through the full life cycle material and energy transfer paths defined in the network topology layer, stage affiliation and flow direction constraints are applied to carbon flow nodes and carbon flow weighted edges to ensure that the carbon flow direction is consistent with the actual production process, ultimately generating a carbon flow network that reflects the dynamic distribution of carbon flow throughout the entire life cycle of the gold mine.

[0047] S2.6: Based on the carbon flow network, extract the carbon load attributes of carbon flow nodes and the weight attributes of carbon flow weighted edges, calculate the influence value of each carbon flow node, and identify carbon flow nodes whose influence values ​​exceed the preset influence value threshold as key influence nodes. Specifically, based on the carbon flow network, the carbon load attribute associated with each carbon flow node is extracted. The carbon load attribute comes from the actual observation values ​​of the corresponding equipment, process, energy consumption, or carbon emission entities in the multi-source data pool within the current time window. At the same time, the weight attributes of all carbon flow weighted edges connected to the carbon flow node are extracted. The weight attributes come from the carbon flow weights. The carbon load attribute of the carbon flow node is summed with the weight attributes of all its adjacent carbon flow weighted edges to calculate the impact value of each carbon flow node. After obtaining the impact values ​​of all carbon flow nodes, the carbon flow nodes with impact values ​​greater than or equal to the preset impact value threshold are identified as key impact nodes.

[0048] The influence value of each carbon flow node is calculated using the following expression: ; In the formula, Indicates carbon flow node The impact value; Indicates the carbon flow node index; Indicates carbon flow node Carbon loading properties; Indicates carbon flow node A connected carbon flow weighted edge; Represents all nodes related to carbon flow. The set consisting of directly connected carbon flow weighted edges; Indicates carbon flow weighted edge The weight attribute.

[0049] It should be noted that the impact threshold is set based on the statistical distribution characteristics of the impact values ​​of carbon flow nodes in the carbon flow network during the historical operation of the gold mine. The specific setting steps are as follows: Select multiple typical production cycles (such as complete weeks or months under different seasons, mining depths, or production capacity levels) from the historical time period covered by the multi-source data pool; construct a carbon flow network within each cycle and calculate the impact values ​​of all carbon flow nodes; summarize the impact value data for all cycles and calculate their overall mean and standard deviation; use the sum of the mean and one standard deviation as the initial value of the preset impact threshold, and combine it with the mine's carbon... Emission management targets are calibrated to form impact thresholds; the exemplary value range is 500 to 2000; the value is based on the carbon emission levels generated by major energy-consuming equipment or core processes in a typical gold mine on a daily or shift scale; if the impact value of a carbon flow node is less than 500, it indicates that its contribution to global carbon emissions is weak, and including it in optimization decisions will bring redundant costs and reduce coordination efficiency; if it is higher than 2000, it usually corresponds to the main energy consumption or large-scale material handling process, which is a concentrated area of ​​emission reduction potential, and decision agents must be prioritized to achieve effective control.

[0050] S2.7: Extract process paths associated with key influencing nodes in the carbon flow network using a path traversal algorithm; Specifically, using a path traversal algorithm, starting from the key impact nodes already selected in the carbon flow network, the algorithm traverses forward and backward along the directed connections of the carbon flow weighted edges until it reaches the starting carbon source node or the ending carbon sink node of the carbon flow network. During the traversal, the sequence of carbon flow nodes traversed and the carbon flow weighted edges they connect are recorded to form a complete process path. Each process path reflects the entire process from resource input and production activities to carbon emission output, and includes key impact nodes and their upstream and downstream related links.

[0051] S3. Assign a decision agent to each key impact node and associated process path, and all decision agents constitute a distributed decision network. S3.1: Based on key impact nodes and related process paths, initialize a decision agent for each key impact node; Specifically, the configuration information of the decision agent includes the carbon load attributes of the key impact node in the carbon flow network, the type and order of upstream and downstream carbon flow nodes in its associated process path, and the actual physical entity corresponding to the key impact node (such as equipment number or energy consumption medium type). The decision agent obtains the current operating status by subscribing to real-time sensor data, equipment operation logs and process parameters related to the key impact node in a multi-source data pool. At the same time, the decision agent binds a preset impact value threshold as the benchmark for its activation and response, forming an initial decision agent with perception, recognition and response capabilities.

[0052] S3.2: Connect all decision agents through a carbon flow network topology to construct a distributed decision network with a hierarchical communication structure; Specifically, through the directed connections between carbon flow nodes in the carbon flow network, each decision agent establishes a communication link with its upstream and downstream decision agents adjacent to it in the carbon flow network. Based on the flow direction of the carbon flow network, the decision agent at the beginning of the process path is set as the bottom-level agent, the decision agent in the middle link is set as the middle-level agent, and the decision agent at the end or the aggregation link is set as the top-level agent. All decision agents form a hierarchical communication structure from bottom to top and converges step by step according to their position in the carbon flow network. The upper-level decision agent can receive the status information of the lower-level decision agent and send coordination instructions to the lower level, thereby constructing a distributed decision network with a hierarchical communication structure.

[0053] S4. Through the virtual carbon quota trading mechanism, decision-making agents within the distributed decision-making network can make collaborative decisions and generate collaborative decision-making results. S4.1: Assign initial virtual carbon allowances and trading strategies to each decision agent via smart contracts; Specifically, the initial virtual carbon allowance value is determined based on the average carbon load attribute of the key influencing node corresponding to the decision agent in the historical typical operating cycle, and is scaled proportionally in combination with a preset influence value threshold; the trading strategy includes pricing rules, trading direction (buy or sell) and allowance adjustment response mechanism, which are automatically configured by the smart contract according to the hierarchical position of the decision agent in the distributed decision network and the carbon flow characteristics of its associated process path; the smart contract writes the initial virtual carbon allowance and trading strategy into the blockchain ledger in an immutable form.

[0054] S4.2: Based on the allocated virtual carbon allowances and trading strategies, multiple rounds of allowance trading are conducted through a two-way auction mechanism, and the trading strategies are dynamically updated during the trading process; Specifically, based on the allocated virtual carbon allowances and trading strategies, each decision-making agent publishes buy or sell intentions on the blockchain platform, including the trading quantity and price. Through a two-way auction mechanism, all buy intentions are sorted from high to low price, and all sell intentions are sorted from low to high price. Trading pairs that meet the condition that the buyer's price is not lower than the seller's price are matched, and allowance delivery is completed at a unified settlement price. After each round of trading, each decision-making agent adjusts the price range or trading direction for the next round based on the deviation between the actual transaction results and the current carbon load attributes, forming a dynamically updated trading strategy. Multiple rounds of allowance trading continue until no new matching pairs appear, ultimately realizing the redistribution of virtual carbon allowances in the distributed decision-making network.

[0055] S4.3: When quota trading reaches equilibrium, generate collaborative decision results based on the current trading strategy; Specifically, when quota trading reaches equilibrium, i.e., when there are no new transactions in two consecutive rounds of two-way auctions or when the price ranges of the buying and selling intentions of all decision-making agents no longer overlap, each decision-making agent determines its own operational adjustment actions based on the final bid direction, quota holding amount, and carbon load attribute matching relationship recorded in the current trading strategy. The high-level decision-making agent summarizes the operational adjustment actions of its subordinate intermediate and low-level decision-making agents, and combines them with the carbon flow constraints in the associated process path to form a consistent operation instruction covering the entire path. All levels of decision-making agents combine their respective operation instructions into a collaborative decision result, which is reflected in the joint optimized response to the carbon emissions of the gold mine throughout its entire life cycle under the virtual carbon quota allocation equilibrium.

[0056] S5. Perform multi-objective optimization on the collaborative decision-making results, solve the non-dominated solution set and select the optimal equilibrium solution, and generate an optimization strategy sequence and the corresponding expected carbon emissions. S5.1: Map the collaborative decision-making results to a multi-dimensional objective space, construct a multi-objective optimization framework, and use the Pareto front contraction algorithm to solve the non-dominated solution set based on the multi-objective optimization framework; Specifically, the operational instructions generated by each decision agent in the collaborative decision-making results are mapped into target vectors in a multi-dimensional target space based on the role of their corresponding key influencing nodes in the entire life cycle of the gold mine. The multi-dimensional target space includes four dimensions: total carbon emissions, energy intensity, production efficiency, and quota utilization rate. Based on the multi-dimensional target space, a multi-objective optimization framework is constructed, in which each combination of operational instructions corresponds to a multi-objective solution. The Pareto front contraction algorithm is used to iteratively select the solution set that is not dominated by other solutions in the multi-objective optimization framework, i.e., the non-dominated solution set. Each solution in the non-dominated solution set represents a set of feasible collaborative strategies that achieve a balance between carbon emissions, energy consumption, efficiency, and quotas.

[0057] S5.2: The non-dominated solution set is screened by the fuzzy preference ensemble method to obtain the optimal equilibrium solution. The optimal equilibrium solution is converted into an optimization strategy sequence and the corresponding expected carbon emissions are generated by simulation through the dynamic carbon emission benchmark model. Specifically, using the fuzzy preference integration method, qualitative preferences in mine management objectives regarding total carbon emissions, energy intensity, production efficiency, and quota utilization (such as "prioritizing carbon emission reduction" and "allowing moderate efficiency losses") are transformed into membership functions for each objective dimension. The membership functions are then used to perform fuzzy evaluations of the performance of each solution in the multi-dimensional objective space within the non-dominated solution set, yielding a comprehensive preference score for each solution. The solution with the highest comprehensive preference score is selected as the optimal equilibrium solution. The decision-making agent operation instructions contained in the optimal equilibrium solution are arranged according to time sequence and process path logic to form an optimization strategy sequence. This optimization strategy sequence is input into a dynamic carbon emission benchmark model, driving the synchronous evolution of the data fusion layer, graph feature layer, and network topology layer under a unified spatiotemporal coordinate system, generating the expected carbon emissions corresponding to the optimization strategy sequence.

[0058] The membership function is used to perform a fuzzy evaluation of the performance of each solution in the non-dominated solution set in the multidimensional objective space, resulting in a comprehensive preference score for each solution, expressed as: ; In the formula, Indicates the first non-dominated solution set The overall preference score for each solution; Indicates the solution index in the non-dominated solution set; This represents the total number of target dimensions in the multidimensional target space; Indicates an index on the target dimension; Indicates the first Preference weights for each objective dimension; Indicates the first Membership function corresponding to each target dimension (decreasing trapezoidal membership function); Indicates the first The solution is at the th solution. Actual performance values ​​across each target dimension.

[0059] S6. Input the optimized strategy sequence into the mine production scheduling process for execution, monitor the actual carbon emission data in real time and compare it with the expected carbon emission, and correct the dynamic carbon emission benchmark model based on the comparison results. S6.1: Decompose the optimization strategy sequence into control commands input to the mine production scheduling process, and collect actual carbon emission data in real time through sensors; Specifically, the operation instructions corresponding to each decision agent in the optimization strategy sequence are decomposed into executable control instructions according to the equipment control interface, process parameter adjustment range, and scheduling execution sequence. The control instructions are then sent to the corresponding actuators in the mine production scheduling process, including frequency converters, valve controllers, pump start-stop devices, and transportation scheduling terminals. During the execution of the control instructions, sensors deployed in key carbon emission links continuously collect actual operating data, including power consumption, fuel consumption, material flow rate, and exhaust gas composition, and convert them into actual carbon emission data based on carbon emission factors.

[0060] It should be noted that the carbon emission factor refers to the carbon dioxide equivalent emissions corresponding to a unit activity level (such as consuming 1 kWh of electricity, burning 1 liter of diesel, or using 1 kilogram of reagent), and is used to convert actual energy or material consumption data into carbon emissions.

[0061] S6.2: Compare the actual carbon emission data with the expected carbon emission amount, generate the comparison results, use a hybrid filtering algorithm to perform data fusion and error analysis on the comparison results, and generate correction parameters; Specifically, actual carbon emission data and expected carbon emission amounts are aligned according to the same time window and spatial location, and the differences between the two values ​​are compared item by item to form a comparison result that includes the direction and magnitude of the deviation. A hybrid filtering algorithm is used, which combines the smoothing ability of Kalman filtering for high-frequency small deviations with the robustness of particle filtering for non-Gaussian large deviations. The comparison results are then fused to identify offset and random disturbance components. Based on the fused deviation sequence, its mean and variance characteristics are statistically analyzed to generate correction parameters, including the carbon load attribute scaling factor and the spatiotemporal correlation intensity attenuation factor.

[0062] S6.3: Use the correction parameters to correct the dynamic carbon emission benchmark model to form a closed-loop iterative optimization; Specifically, the carbon load attribute and spatiotemporal correlation strength weights in the dynamic carbon emission benchmark model are updated using correction parameters. The carbon load attribute scaling factor is applied to the observation value mapping relationship of the corresponding entity in the data fusion layer, and the spatiotemporal correlation strength attenuation factor is applied to the weight attribute of the relevant edge in the graph feature layer. The updated dynamic carbon emission benchmark model regenerates the expected carbon emissions in the next time window and is compared with the actual carbon emission data collected by the sensor again to form a new round of comparison results. This process is continuously cyclical, so that the simulation output of the dynamic carbon emission benchmark model gradually approaches the actual operating state, realizing closed-loop iterative optimization.

[0063] In summary, this invention achieves semantic understanding and correlation analysis of carbon emissions throughout the entire process of mine exploration and closure by constructing a carbon footprint knowledge graph; within the framework of product lifecycle management, the carbon footprint knowledge graph intuitively reveals the key sources and transmission paths of carbon flows across the lifecycle, providing a unified and semantically rich knowledge foundation for carbon flow network generation, key node identification, and distributed collaborative decision-making based on the product lifecycle management concept; and by associating and integrating unstructured document data from each stage with real-time sensor data, it ensures that the dynamic carbon emission benchmark model possesses both profound full-lifecycle process background knowledge and can dynamically respond to changes in actual production status, thereby laying a precise and reasonable data foundation for the entire carbon improvement potential quantification method.

[0064] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method of quantifying the improvement potential of carbon emissions for a mine life cycle process, characterized by: comprising, Collecting real-time sensor data, equipment operation logs, process parameters and unstructured documents of each stage in the whole life cycle of gold mines to build a dynamically updated multi-source data pool; Based on the multi-source data pool, extract carbon emission entities and semantic relationships of carbon emission entities, and build a carbon footprint knowledge graph; Build a dynamic carbon emission benchmark model, map the carbon footprint knowledge graph to the dynamic carbon emission benchmark model, generate a carbon flow network, and identify key impact nodes and associated process paths in the carbon flow network that affect global carbon emissions; Assign a decision agent to each key impact node and associated process path, and all decision agents form a distributed decision network; Through a virtual carbon quota trading mechanism, the decision agents in the distributed decision network make collaborative decisions to generate collaborative decision results; Multi-objective optimization is performed on the collaborative decision results, a non-dominated solution set is obtained, and the optimal equilibrium solution is selected to generate an optimization strategy sequence and the corresponding expected carbon emissions; The optimization strategy sequence is input into the mine production scheduling process for execution, real-time monitoring of actual carbon emissions data is performed, and the actual carbon emissions data are compared with the expected carbon emissions, and the dynamic carbon emission benchmark model is corrected based on the comparison results.

2. A method of quantifying the improvement potential of carbon emissions of a mine life cycle process as claimed in claim 1, characterised in that: The construction of the dynamically updated multi-source data pool is as follows, Through the Internet of Things protocol, data interface and document parser, the real-time sensor data, equipment operation logs, process parameters and unstructured documents of each stage in the whole life cycle of gold mines are converted into standardized time series data and stored in the established time series database; Through information entropy theory, the standardized time series data are subjected to value quantification analysis to generate value marked data stream; According to the pre-defined data value weight table, the value marked data stream is prioritized to form a hierarchical data stream; The hierarchical data stream is mapped to a unified space-time coordinate system, quality verification is performed using a dynamic calibration algorithm, and the time series database is updated through a dynamic updating mechanism to build a dynamically updated multi-source data pool.

3. A method of quantifying the improvement potential of carbon emissions of a mine life cycle process as claimed in claim 2, characterised in that: The construction of the carbon footprint knowledge graph is as follows, Based on the text data in the multi-source data pool, a multi-task learning model is used to simultaneously identify four types of entities and entity semantic relationships, including equipment, process, energy consumption and carbon emissions, to generate a set of entity semantic relationship triples; Align the set of entity semantic relationship triples with the space-time coordinate system, and calculate the space-time correlation strength between entities using a graph attention network to form a knowledge graph pattern layer; Based on the knowledge graph pattern layer, the entities are taken as nodes, the entity semantic relationships are taken as edges, and the space-time correlation strength is taken as the weight attribute of the edges to build a carbon footprint knowledge graph.

4. The method for quantifying the potential for improving carbon emissions throughout the mining lifecycle as described in claim 1, characterized in that: The construction of the dynamic carbon emission benchmark model is as follows, Based on the multi-source data pool, a data fusion layer is built, a graph feature layer is built based on the carbon footprint knowledge graph, and a network topology layer is built based on LCA life cycle assessment method; Through a data stream pipeline, the data fusion layer, the graph feature layer and the network topology layer are connected in real time to build a dynamic carbon emission benchmark model.

5. A method of quantifying the improvement potential of carbon emissions of a mine life cycle process as claimed in claim 4, characterised in that: The generation of the carbon flow network is as follows, Based on the data fusion layer of the dynamic carbon emission benchmark model, real-time monitoring data are provided, and combined with the entity semantic relationships in the carbon footprint knowledge graph, a space-time correlation strength weight matrix is extracted; According to the spatiotemporal correlation strength weight matrix, carbon flow characteristics are extracted through a graph feature layer, and the carbon flow weight of each edge in the carbon flow network is calculated; Based on the carbon flow weight, the entity nodes and entity semantic relationship edges in the carbon footprint knowledge graph are mapped into the carbon flow nodes and carbon flow weighted edges of the carbon flow network through a network topology layer, and a carbon flow network is generated.

6. A method of quantifying the improvement potential of carbon emissions of a mine life cycle process as claimed in claim 5, characterised in that: The key influence nodes and associated process paths in the carbon flow network that affect global carbon emissions are identified, and the steps are as follows, Based on the carbon flow network, the carbon load attribute of the carbon flow node and the weight attribute of the carbon flow weighted edge are extracted, the influence value of each carbon flow node is calculated, and the carbon flow node with an influence value exceeding a preset influence value threshold is regarded as a key influence node; The process path associated with the key influence node is extracted in the carbon flow network through a path traversal algorithm.

7. A method of quantifying the improvement potential of carbon emissions of a mine life cycle process as claimed in claim 6, characterised in that: All decision agents constitute a distributed decision network, and the steps are as follows, Based on the key influence nodes and associated process paths, a decision agent is initialized for each key influence node; All decision agents are connected through the carbon flow network topology to build a distributed decision network with a hierarchical communication structure.

8. A method of quantifying the improvement potential of carbon emissions of a mine life cycle process as claimed in claim 7, characterised in that: The steps for generating a collaborative decision result are as follows, An initialization virtual carbon quota and a transaction strategy are assigned to each decision agent through a smart contract; Based on the assigned virtual carbon quota and transaction strategy, a multi-round quota transaction is performed through a two-way auction mechanism, and the transaction strategy is dynamically updated during the transaction process; When the quota transaction reaches an equilibrium state, a collaborative decision result is generated according to the current transaction strategy.

9. A method of quantifying the improvement potential of carbon emissions of a mine life cycle process as claimed in claim 8, characterised in that: The steps for generating an optimization strategy sequence and corresponding expected carbon emissions are as follows, The collaborative decision result is mapped to a multi-dimensional target space to build a multi-objective optimization framework, and a non-dominated solution set is solved based on the multi-objective optimization framework using a Pareto front contraction algorithm; The non-dominated solution set is screened by a fuzzy preference integration method to obtain an optimal equilibrium solution, which is converted into an optimization strategy sequence, and the corresponding expected carbon emissions are generated through a dynamic carbon emission benchmark model simulation.

10. A method of quantifying the improvement potential of carbon emissions of a mine life cycle process as claimed in claim 9, characterised in that: The steps for correcting the dynamic carbon emission benchmark model by comparing the results are as follows, The optimization strategy sequence is decomposed into control instructions input into the mine production scheduling process, and actual carbon emission data is collected in real time through sensors; The actual carbon emission data is compared with the expected carbon emissions to generate a comparison result, a mixed filtering algorithm is used to perform data fusion and error analysis on the comparison result to generate a correction parameter; The dynamic carbon emission benchmark model is corrected using the correction parameter to form a closed-loop iterative optimization.

Images