Decarburization factor identification and path design method based on path deduction

Abstract

The invention discloses a decarburization factor identification and path design method based on path deduction, which comprises the following steps: S1, collecting multi-source data, carrying out multi-dimensional classification on the data, and constructing a GDIM variable system; s2, constructing a PLS path deduction model based on a GDIM variable system; s3, based on the path structures of the plurality of family samples, performing path feature clustering, and extracting a common carbon locking mechanism and a representative behavior chain; s4, inputting the target family features into the established path database, performing similarity matching, and outputting a personalized decarburization suggestion combination; and S5, constructing a path network diagram and a cause and effect diagram, and realizing visual expression of a path structure. According to the method, the advantages of family behavior heterogeneity, multi-source data driving and path structure recognition can be fused, and deep mining and personalized path deduction of the carbon locking factor are realized.

Description

Technical Field

[0001] The present invention relates to the field of carbon emission reduction, and in particular to a decarbonization factor identification and path design method based on path deduction. Background Art

[0002] As a key unit of final energy use, households have attracted widespread attention for their potential to control carbon emissions. However, existing research has largely focused on building energy efficiency standards, energy consumption structure ratios, or equipment technology upgrades, overlooking the heterogeneity of individual households in terms of cognition, behavior, and feedback mechanisms.

[0003] At present, most traditional energy consumption models are driven by static data, making it difficult to dynamically capture the changing process of household energy consumption behavior and its potential causal chain. At the same time, existing low-carbon path design methods are mostly based on general rules and indicators, lacking targeted modeling of specific family structures, behavioral motivations and response mechanisms, resulting in poor feasibility of design results and low willingness to implement. Although in recent years, path deduction and causal identification methods have gradually been applied to carbon emission reduction behavior modeling. Among them, the partial least squares (PLS) path modeling method has been valued by the academic community because of its applicability to small samples, high dimensions, and multicollinearity problems, but its structural definition often relies on manual settings and lacks logical hierarchical support between factors.

[0004] Therefore, to solve the above problems, a decarbonization factor identification and path design method based on path deduction is needed, which can integrate the advantages of household behavior heterogeneity, multi-source data drive and path structure identification, and realize the in-depth mining of carbon lock-in factors and personalized path deduction. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to overcome the defects in the existing technology and provide a decarbonization factor identification and path design method based on path deduction, which can integrate the advantages of household behavior heterogeneity, multi-source data drive and path structure identification, and realize in-depth mining of carbon locking factors and personalized path deduction.

[0006] The decarbonization factor identification and path design method based on path deduction of the present invention comprises the following steps:

[0007] S1. Collect multi-source data, classify the data in multiple dimensions, and build a GDIM variable system;

[0008] S2. Construct a PLS path deduction model based on the GDIM variable system;

[0009] S3. Based on the path structures of multiple household samples, perform path feature clustering to extract common carbon lock-in mechanisms and representative behavior chains;

[0010] S4. Input the target household characteristics into the established pathway database, perform similarity matching, and output a personalized decarbonization recommendation combination;

[0011] S5. Construct path network diagrams and causal diagrams to achieve visual expression of path structures.

[0012] Furthermore, the step S1 specifically includes:

[0013] Collecting raw data from data sources, including energy bill data, smart device logs, household behavior questionnaires, environmental and building parameters, and indoor sensor data;

[0014] Clean and uniformly code all types of collected data, handle missing values ​​and outliers, and normalize continuous variables;

[0015] The pre-processed variables are divided into four factor groups according to the GDIM logical framework. The four factor groups are driving factors, demand factors, intervention factors, and mechanism factors. Each factor is determined by the annotation method and factor correlation analysis.

[0016] Finally, a GDIM variable matrix containing four dimensions and multiple variable items is formed.

[0017] Furthermore, the step S2 specifically includes:

[0018] The variables in the four dimensions are sorted according to the family samples to form a standardized variable input matrix; each row of the matrix represents a family sample, and each column corresponds to a characteristic variable;

[0019] Each type of latent factor group has corresponding observed variables, and the initial path model framework is set according to the assumptions of family behavior theory;

[0020] Partial least squares structural equation modeling method is used to model the path relationship of variables in the four dimensions;

[0021] After the model is fitted, the path coefficients, loading coefficients and related statistical indicators between each potential factor are output; the significance test of the path coefficients is performed using the Bootstrap method to identify significant path relationships;

[0022] Based on the combination of significant pathways, we reconstruct typical high-carbon causal chains and screen out carbon lock pathways that appear frequently or have high weights.

[0023] The model results were cross-validated to improve the robustness of path identification, and finally the path structures identified in multiple family samples were integrated to form the GDIM path database.

[0024] Furthermore, the step S3 specifically includes:

[0025] For each family sample, the path model constructed was used to extract its significant causal paths, including the path start and end variables, path coefficient values, and directional relationships. Each path was vectorized and encoded in the form of variable number-path direction-path weight to form a unified path feature matrix.

[0026] The continuous weight information in the path feature matrix is ​​standardized to eliminate the interference of different variable dimensions on the clustering results;

[0027] Unsupervised learning methods were used to perform cluster analysis on the standardized path feature matrix;

[0028] The frequently occurring path chains in each clustering result were structurally merged and summarized to extract a group of carbon lock-in mechanisms with common characteristics.

[0029] In each type of pathway pattern, we further identified the representative family sample that was most similar to the pathway structure of that type, and extracted its complete GDIM pathway diagram as a typical manifestation of that type of mechanism.

[0030] Furthermore, the step S4 specifically includes:

[0031] The structural characteristics, behavioral data, energy consumption records, and cognitive feedback of the target households were collected, coded and classified according to the GDIM variable system, and normalized to generate the target household GDIM feature vector;

[0032] The path results and cluster structures generated by modeling multiple household samples are used to construct a path database. Each path chain is stored in the form of a GDIM variable sequence and / or path direction and / or weight value, and is associated with its representative sample feature vector and the corresponding decarbonization recommendation combination.

[0033] Measure the similarity between the GDIM vectors of the target family and each representative sample in the pathway database, and perform similarity matching calculations;

[0034] Identify the 1 to 3 pathway chains with the best similarity as the high-risk carbon lock-in behavior pathways of the target households, and identify the carbon lock-in mechanism category to which they belong;

[0035] Based on the intervention strategy template corresponding to the matching path chain and combined with the unique structural and behavioral characteristics of the target family, a recommendation combination is generated.

[0036] Furthermore, the step S5 specifically includes:

[0037] Extract the directed path relationships between GDIM variables from the PLS path deduction modeling results, including variable nodes and path edges between them. The path edges are accompanied by structural weights, significance indicators, and attribution mechanism labels.

[0038] Use a network graph construction tool to map GDIM variables into network nodes, with nodes of different dimensions distinguished by color. Directed edges represent causal paths between variables, with edge thickness indicating path weight. Edge labels can indicate impact intensity and mechanism type. The graph structure supports node aggregation and path focusing, making it easier for users to view critical paths.

[0039] Chain extraction of frequently occurring typical paths is performed and drawn in a flowchart style, emphasizing the temporal progression of causal logic;

[0040] Setting multiple view switching, including full network view, single family view and mechanism summary view;

[0041] Set up interactive operations: Click a node to view variable descriptions, and click a path to view impact details. Users can use the graphical feedback function to evaluate the credibility of the path and the acceptability of the intervention suggestion. Feedback data can be used for subsequent model revisions.

[0042] The generated path map can be exported into multiple image formats for use in patent drawings, report presentations, or online deployment; it can also be set up in the corresponding system for daily carbon emission monitoring and recommendation interaction.

[0043] The present invention has the following beneficial effects: The present invention discloses a path-based decarbonization factor identification and path design method. By dividing household carbon behavior factors into four variable groups: driver (G), demand (D), intervention (I), and mechanism (M), a multidimensional causal chain structure is established. Combined with the PLS structural modeling method, this method achieves quantitative identification of path relationships between factors. This method not only identifies carbon lock-in pathways but also supports the personalized design of behavioral intervention and feedback strategies. It has stronger explanatory power and adaptability, and is particularly suitable for the design and optimization of carbon reduction strategies in heterogeneous household structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0045] Figure 1 This is a flow chart of the decarbonization factor identification and path design method of the present invention. DETAILED DESCRIPTION

[0046] The present invention is further described below with reference to the accompanying drawings, as shown in the drawings:

[0047] This embodiment discloses a decarbonization factor identification and path design method based on path deduction, comprising the following steps:

[0048] S1. Collect multi-source data, classify the data in multiple dimensions, and build a GDIM variable system;

[0049] S2. Construct a PLS path deduction model based on the GDIM variable system;

[0050] S3. Based on the path structures of multiple household samples, perform path feature clustering to extract common carbon lock-in mechanisms and representative behavior chains;

[0051] S4. Input the target household characteristics into the established pathway database, perform similarity matching, and output a personalized decarbonization recommendation combination;

[0052] S5. Construct path network diagrams and causal diagrams to achieve visual expression of path structures.

[0053] In this embodiment, in step S1, heterogeneous data reflecting household carbon behavior is first collected through multiple sources to construct a complete GDIM (Driver-Demand-Intervention-Mechanism) variable system. Specifically, the following sub-steps are included:

[0054] Raw data is collected from data sources, including energy bill data, smart device logs, household behavior questionnaires, environmental and building parameters, and indoor sensor data. Energy bill data includes historical energy consumption records for electricity, water, and gas, spanning no less than 12 months, with a sampling frequency primarily monthly. Smart device logs include start and stop frequencies, operating hours, and load intensity for air conditioners, water heaters, heating systems, lighting, and other equipment. The household behavior questionnaire covers subjective behavioral variables such as energy usage habits, comfort preferences, energy-saving cognition, and feedback perception. Environmental and building parameters primarily collect structural characteristics such as household type, building age, window and wall materials, insulation measures, orientation, and regional climate. Indoor sensor data includes indicators such as temperature and humidity, carbon dioxide concentration, and light intensity, used to assist in verifying behavioral responses and environmental feedback.

[0055] Clean and uniformly encode all types of collected data, handle missing values ​​and outliers, and normalize continuous variables to ensure that variables of different dimensions can be modeled in parallel;

[0056] The above pre-processed variables are divided into four factor groups according to the set GDIM logical framework; the four factor groups are driving factors, demand factors, intervention factors and mechanism factors; driving factors (Generator): describe the external background or internal motivation that induces household energy consumption behavior, such as cognitive level, income level, constraints, incentives, etc.; demand factors (Demand): refer to the specific life function needs of family members, such as temperature comfort, lighting brightness, bathing frequency, spatial activity density, etc.; intervention factors (Intervention): include the intervention of external measures or technical means, such as energy-saving subsidies, equipment energy efficiency rating, intelligent control strategy, etc.; mechanism factors (Mechanism): describe the feedback and adjustment mechanism of the family after the energy consumption behavior occurs, such as behavioral adjustment ability, feedback system perception, equipment response efficiency, etc. Among them, each type of factor is determined by the labeling method and factor correlation analysis to ensure that the variable classification logic is rigorous and the structure is clear;

[0057] Ultimately, a GDIM variable matrix consisting of four dimensions and multiple variables is formed, providing a structured input foundation for subsequent path modeling and causal reasoning. The four dimensions correspond to four types of factors: driving factor dimension, demand factor dimension, intervention factor dimension, and mechanism factor dimension; and the multiple variables include no fewer than 80.

[0058] In this embodiment, in step S2, a PLS path deduction model is constructed based on the GDIM variable system to achieve quantitative modeling and identification of the causal path of household decarbonization behavior. The implementation process includes the following specific steps:

[0059] After completing the construction of the GDIM variable system, the variables in the four dimensions are organized according to the household samples to form a standardized variable input matrix. Each row of the matrix represents a household sample, and each column corresponds to a characteristic variable. The variable types can include quantitative (such as the duration of air conditioning use), ordinal (such as cognitive level classification), and categorical (such as family structure type, processed after dummy variable conversion).

[0060] To identify the causal path structure, based on the GDIM theoretical logic, we initially set up four groups of potential factors (latent variables): driving factors (G), demand (D), intervention (I), and mechanism (M). Each type of potential factor group has corresponding observed variables. Based on the assumptions of family behavior theory, we set up the initial path model framework. For example, it is assumed that driving factors affect demand and intervention, and then act on the mechanism and affect the final carbon emission behavior.

[0061] The partial least squares structural equation modeling method (PLS-SEM) is used to model the path relationship of variables in the four dimensions; the PLS method is particularly suitable for dealing with practical data problems such as limited sample size and multicollinearity between variables. The modeling process is implemented using tools such as SmartPLS or R (plspm package). The operating parameters include the setting of the minimum fitting error of the path weight, the number of factors extracted from the latent variables, and the verification indicators (such as AVE, R 2 , path significance) settings, etc.

[0062] After the model is fitted, the path coefficients, load coefficients, and related statistical indicators between each potential factor are output. The significance test of the path coefficients is performed using the Bootstrap method to identify significant path relationships, such as: cognitive level (G) → intensity of heating and cooling demand (D) → frequency of air conditioning equipment use (I) → missing energy consumption feedback (M).

[0063] Based on significant pathway combinations, we reconstruct typical high-carbon causal chains and screen out frequently occurring or highly weighted carbon-locking pathways. For example, we identified the pathway: complex household structure (G) → high comfort needs (D) → frequent use of old appliances (I) → no energy consumption feedback mechanism (M) → high carbon emissions, which we then prioritized for intervention.

[0064] The model results were cross-validated to improve the robustness of path identification, and finally the path structures identified in multiple family samples were integrated to form the GDIM path database for subsequent path clustering and personalized design.

[0065] The above steps not only achieve scientific modeling of the causal chain of household carbon behavior, but also provide a structural basis and quantitative support for the formulation and deduction of subsequent behavioral intervention strategies.

[0066] In this embodiment, in step S3, based on the PLS path model deduction results of multiple household samples, a path feature cluster analysis is further performed to extract representative common carbon lock-in mechanisms and typical behavior chains. Specifically, the following steps are included:

[0067] For each family sample, the path model constructed was used to extract its significant causal paths, including the path start and end variables, path coefficient values, and directional relationships. Each path was vectorized and encoded in the form of variable number-path direction-path weight to form a unified path feature matrix.

[0068] The continuous weight information in the path feature matrix is ​​standardized to eliminate the interference of different variable dimensions on the clustering results. If the dimension is too high, the principal component analysis (PCA) or t-SNE method can be used to reduce the dimension of the feature space.

[0069] Cluster analysis of the standardized path feature matrix is ​​performed using unsupervised learning methods such as K-means, DBSCAN, or hierarchical clustering. The number of clusters can be determined by using indicators such as the silhouette coefficient or the Calinski-Harabasz index.

[0070] By merging and summarizing the frequently occurring path chains in each clustering result, a set of carbon lock-in mechanisms with common characteristics was extracted. For example, in cluster result A, a large number of households exhibited the path of low awareness → strong comfort needs → frequent equipment operation → abnormal energy consumption, which can be summarized as a high-carbon behavior chain driven by insufficient information.

[0071] In each type of pathway pattern, we further identified representative family samples that were most similar to the pathway structure of that type, and extracted their complete GDIM pathway diagrams as typical manifestations of that type of mechanism for subsequent comparative analysis and intervention pathway design.

[0072] Through the above steps, we can systematically summarize the potential carbon locking mechanism groups in multiple family samples, providing common support and structural templates for subsequent personalized design strategies.

[0073] In this embodiment, in step S4, to implement personalized decarbonization path design for a single target household, similarity matching and decarbonization suggestion output are performed based on the previously constructed path database and clustering summary results. The specific steps are as follows:

[0074] The target households' structural characteristics, behavioral data, energy consumption records, and cognitive feedback were collected, coded and classified according to the GDIM variable system, and normalized to generate the target household's GDIM feature vector. Structural characteristics include apartment size, building year, and household size; behavioral data includes temperature control setting frequency and equipment usage habits; energy consumption records include monthly electricity consumption and peak and valley loads; and cognitive feedback includes the environmental awareness questionnaire score.

[0075] The path results and cluster structures generated by modeling multiple household samples are used to construct a path database. Each path chain is stored in the form of a GDIM variable sequence and / or path direction and / or weight value, and is associated with its representative sample feature vector and the corresponding decarbonization recommendation combination.

[0076] Indicators such as cosine similarity, Euclidean distance, or weighted Manhattan distance are used to measure the similarity between the target household and the GDIM vectors of each representative sample in the path database, and similarity matching calculations are performed. To improve the effectiveness of matching, matching can be combined with the feature space after dimensionality reduction using principal component analysis, or the Top-k nearest neighbor algorithm can be introduced to select the most similar path patterns.

[0077] Identify the 1-3 pathway chains with the best similarity as the high-risk carbon lock-in behavior pathways for target households, and identify the carbon lock-in mechanism categories they belong to, such as feedback failure, high demand inertia, and structural coupling blockage, to provide a mechanistic basis for subsequent intervention plans;

[0078] Based on the intervention strategy template corresponding to the matching path chain and combining it with the target household's unique structural and behavioral characteristics, a combination of recommendations is generated. The intervention strategy template might be: Cognitive Enhancement → Behavioral Reminders → Equipment Replacement → Load Peak Shaving. Recommendations might include the following intervention types: Behavioral: Adjusting air conditioning temperature settings, limiting hot water usage, and fostering energy consumption feedback habits; Equipment: Replacing high-energy-consuming older equipment and introducing smart control terminals; Structural: Improving ventilation and insulation conditions and adjusting the layout of functional areas in the home; Feedback: Adding a real-time energy consumption feedback interface and delivering regular energy consumption recommendation reports.

[0079] All suggested combinations are presented through user interfaces, such as mobile apps or home energy management platforms, as illustrated paths and suggestion cards, enhancing understanding and implementation. User feedback is collected and used for dynamic optimization of the path database and revision of design strategies.

[0080] Through the above steps, the full process deduction from structured path mechanism to individualized intervention recommendations was achieved, ensuring the accuracy, feasibility and user acceptance of the decarbonization strategy.

[0081] In this embodiment, in step S5, to improve the comprehensibility and operability of the household carbon lock-in mechanism identification results, the interaction relationship between GDIM variables is visualized in the form of a path network diagram and a cause-effect diagram. The specific operation process is as follows:

[0082] Directed path relationships between GDIM variables are extracted from the PLS path deduction modeling results, including variable nodes and path edges (i.e., causal relationships) between them. Path edges are accompanied by structural weights, significance indicators, and attribution mechanism labels (e.g., demand-driven). Variable nodes can include low cognition, high demand, frequent air conditioning use, etc.

[0083] Use a network graph construction tool to map GDIM variables into network nodes, with nodes of different dimensions distinguished by color. Directed edges represent causal paths between variables, with edge thickness indicating path weight. Edge labels can indicate impact intensity and mechanism type. The graph structure supports node aggregation and path focusing, making it easier for users to view critical paths.

[0084] For typical, frequently occurring pathways, such as insufficient awareness → increased comfort temperature expectations → frequent air conditioning use → increased energy consumption, chain extraction is performed and drawn in a flowchart style, emphasizing the temporal progression of causal logic. Path diagrams can be used to explain user feedback, promote strategies, or enable intelligent push notifications.

[0085] Multiple view switching is available, including full network view, single-family view, and mechanism summary view. The full network view displays the aggregated path network of all family samples, identifying the core nodes of carbon lock factors. The single-family view displays the matching paths of target families and their high-weight causal chains, highlighting the individualized behavioral characteristics of users. The mechanism summary view displays high-frequency path mechanisms according to their categories, facilitating the summary of intervention strategies.

[0086] Set up interactive operations: Click a node to view variable descriptions, and click a path to view impact details. Users can use the graphical feedback function to evaluate the credibility of the path and the acceptability of the intervention suggestion. Feedback data can be used for subsequent model revisions.

[0087] The generated path map can be exported into multiple image formats for use in patent drawings, report presentations, or online deployment; it can also be set up in the existing home energy management system for daily carbon emission monitoring and recommendation interaction.

[0088] Through the above steps, the interpretability and communicability of the path structure and causal logic are effectively enhanced, and the user's visual cognition of the model results and the willingness to implement strategies are improved.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A decarbonization factor identification and path design method based on path deduction, characterized by: The steps include: S1. Collect multi-source data, classify the data in multiple dimensions, and build a GDIM variable system; S2. Construct a PLS path deduction model based on the GDIM variable system; S3. Based on the path structures of multiple household samples, perform path feature clustering to extract common carbon lock-in mechanisms and representative behavior chains; S4. Input the target household characteristics into the established pathway database, perform similarity matching, and output a personalized decarbonization recommendation combination; S5. Construct path network diagrams and causal diagrams to achieve visual expression of path structures.

2. The decarbonization factor identification and path design method based on path deduction according to claim 1 is characterized by: The step S1 specifically includes: Collecting raw data from data sources, including energy bill data, smart device logs, household behavior questionnaires, environmental and building parameters, and indoor sensor data; Clean and uniformly code all types of collected data, handle missing values ​​and outliers, and normalize continuous variables; The pre-processed variables are divided into four factor groups according to the GDIM logical framework. The four factor groups are driving factors, demand factors, intervention factors, and mechanism factors. Each factor is determined by the annotation method and factor correlation analysis. Finally, a GDIM variable matrix containing four dimensions and multiple variable items is formed.

3. The decarbonization factor identification and path design method based on path deduction according to claim 1 is characterized in that: The step S2 specifically includes: The variables in the four dimensions are sorted according to the family samples to form a standardized variable input matrix; each row of the matrix represents a family sample, and each column corresponds to a characteristic variable; Each type of latent factor group has corresponding observed variables, and the initial path model framework is set according to the assumptions of family behavior theory; Partial least squares structural equation modeling method is used to model the path relationship of variables in the four dimensions; After the model is fitted, the path coefficients, loading coefficients and related statistical indicators between each potential factor are output; the significance test of the path coefficients is performed using the Bootstrap method to identify significant path relationships; Based on the combination of significant pathways, we reconstruct typical high-carbon causal chains and screen out carbon lock pathways that appear frequently or have high weights. The model results were cross-validated to improve the robustness of path identification, and finally the path structures identified in multiple family samples were integrated to form the GDIM path database.

4. The decarbonization factor identification and path design method based on path deduction according to claim 1 is characterized in that: The step S3 specifically includes: For each family sample, the path model constructed was used to extract its significant causal paths, including the path start and end variables, path coefficient values, and directional relationships. Each path was vectorized and encoded in the form of variable number-path direction-path weight to form a unified path feature matrix. The continuous weight information in the path feature matrix is ​​standardized to eliminate the interference of different variable dimensions on the clustering results; Unsupervised learning methods were used to perform cluster analysis on the standardized path feature matrix; The frequently occurring path chains in each clustering result were structurally merged and summarized to extract a group of carbon lock-in mechanisms with common characteristics. In each type of pathway pattern, we further identified the representative family sample that was most similar to the pathway structure of that type, and extracted its complete GDIM pathway diagram as a typical manifestation of that type of mechanism.

5. The decarbonization factor identification and path design method based on path deduction according to claim 1 is characterized in that: The step S4 specifically includes: The structural characteristics, behavioral data, energy consumption records, and cognitive feedback of the target households were collected, coded and classified according to the GDIM variable system, and normalized to generate the target household GDIM feature vector; The path results and cluster structures generated by modeling multiple household samples are used to construct a path database. Each path chain is stored in the form of a GDIM variable sequence and / or path direction and / or weight value, and is associated with its representative sample feature vector and the corresponding decarbonization recommendation combination. Measure the similarity between the target family and the GDIM vectors of each representative sample in the pathway database, and perform similarity matching calculations; Identify the 1 to 3 pathway chains with the best similarity as the high-risk carbon lock-in behavior pathways of the target households, and identify the carbon lock-in mechanism category to which they belong; Based on the intervention strategy template corresponding to the matching path chain and combined with the unique structural and behavioral characteristics of the target family, a recommendation combination is generated.

6. The decarbonization factor identification and path design method based on path deduction according to claim 1 is characterized by: The step S5 specifically includes: Extract the directed path relationships between GDIM variables from the PLS path deduction modeling results, including variable nodes and path edges between them. The path edges are accompanied by structural weights, significance indicators, and attribution mechanism labels. Use a network graph construction tool to map GDIM variables into network nodes, with nodes of different dimensions distinguished by color. Directed edges represent causal paths between variables, with edge thickness indicating path weight. Edge labels can indicate impact intensity and mechanism type. The graph structure supports node aggregation and path focusing, making it easier for users to view critical paths. Chain extraction of frequently occurring typical paths is performed and drawn in a flowchart style, emphasizing the temporal progression of causal logic; Setting multiple view switching, including full network view, single family view and mechanism summary view; Set up interactive operations: Click a node to view variable descriptions, and click a path to view impact details. Users can use the graphical feedback function to evaluate the credibility of the path and the acceptability of the intervention suggestion. Feedback data can be used for subsequent model revisions. The generated path map can be exported into multiple image formats for use in patent drawings, report presentations, or online deployment; it can also be set up in the corresponding system for daily carbon emission monitoring and recommendation interaction.

Images