Drainage basin pollution discharge transaction knowledge reasoning method based on TransE and Compgcn

Abstract

The invention discloses a drainage basin pollution discharge transaction knowledge reasoning method based on TransE and Compgcn, and relates to the field of knowledge graph embedding and the field of environmental resources. According to the method, the pre-training embedding of TransE is used as the initial representation of the combined graph convolutional network, so that the training efficiency and the final performance of the model are remarkably improved. The objective of the invention is to provide efficient and accurate derivation support for drainage right transactions in a drainage basin. The method comprises the following steps: firstly, carrying out relation modeling on a knowledge graph by adopting a TransE model to generate high-quality initial entity and relation vector representation; secondly, inputting the embedded images into CompGCN, performing graph convolution operation by utilizing the combination operation characteristics of the embedded images, and introducing an attention mechanism to optimize node representation update so as to capture complex graph structure information; and finally, through iteration of the training process and cost function optimization, rapid convergence and performance improvement of the model are realized. The method is suitable for tasks such as knowledge graph completion and system recommendation, so as to help to improve prediction precision and calculation efficiency.

Description

Technical Field

[0001] This application relates to the fields of knowledge graph embedding and environmental resources, and particularly to a method for reasoning about basin pollution discharge trading knowledge based on TransE and Compgcn. Background Art

[0002] As a market-based means, water rights trading realizes the flexible allocation and efficient utilization of water resources by buying, selling or leasing the right to use water resources, and its importance in global water resources management has become increasingly prominent. With the intensification of climate change, population growth and the acceleration of industrialization, the problem of water shortage has become more and more serious, and water rights trading has become an important tool to address the uneven distribution of water resources and alleviate the contradiction between supply and demand. Especially in basin management, water rights trading can optimize the allocation of pollution discharge rights and promote the sustainable utilization of environmental resources. In the Murray-Darling Basin in Australia, the water rights trading market is extremely active. In the 2016-17 fiscal year, allocation transactions accounted for 97% of the country's water rights trading, and entitlement transactions accounted for 77%. The average annual transaction volume in the past decade was nearly 2 billion Australian dollars, mainly used for high-water-consuming industries such as agricultural irrigation. At the same time, the global water rights trading volume shows a rapid growth trend. According to statistics, the water trading volume increased by 2.9 times from 1986 to 2022, with agricultural products dominating. It is expected that by 2100, the virtual water trading volume of renewable water sources may triple. Since the water rights trading case between Dongyang and Yiwu in Zhejiang in 2000 in China, water rights trading pilots have been gradually promoted, covering agricultural, industrial and ecological water use, and the trading volume has been increasing year by year, showing strong development momentum. Against this background, more and more problems have gradually emerged. Since water rights trading involves complex legal, environmental and economic factors, it is often difficult for both parties to obtain transparent transaction data and user information. At the same time, in the current trading mode, the trading efficiency is low. In basin pollution discharge rights trading, it is difficult to accurately match the pollution discharge rights trading between enterprises, and the dynamic and complex nature of trading needs further exacerbates this problem. In order to address problems such as information asymmetry, low trading efficiency and high matching difficulty, how to construct an efficient and accurate trading reasoning and matching mechanism has become an urgent problem to be solved.

[0003] A personalized recommendation system and method based on a knowledge graph and collaborative filtering disclosed in a Chinese patent document with the publication number CN118227884A. This patent utilizes collaborative filtering technology to fuse the representations of users and items in the knowledge graph on the basis of a matrix factorization framework to learn latent factors. A recommendation system based on knowledge graph completion disclosed in a Chinese patent document with the publication number CN119228477A. This patent combines a knowledge graph completion module of a multi-relational graph attention network and multi-scale convolution to complete the knowledge completion task. A method for optimizing few-shot multi-hop reasoning based on a knowledge graph disclosed in a Chinese patent document with the publication number CN118095445A. This patent fuses the feature vectors generated by Transformer with the graph feature vectors processed by GCN to optimize the performance. A method for knowledge graph completion based on DeepPath-CompGCN disclosed in a Chinese patent document with the publication number CN119761477A. This patent embeds entities and relationships in the knowledge graph through a CompGCN convolutional layer and inputs them into the DeepPath model for path reasoning to obtain the optimal reasoning path. A method and system for knowledge graph completion based on multi-view contrast learning disclosed in a Chinese patent document with the publication number CN119312907A. This patent uses view contrast learning. A method and system for processing and querying commodity information disclosed in a Chinese patent document with the publication number CN119377433A. This patent encodes commodity text using a BERT model based on Transformer to obtain the deep semantic representation of the text, improving the processing efficiency of commodity information.

[0004] The present invention aims to effectively integrate the advantages of the TransE model and the CompGCN model. By collecting basin-related data to construct a knowledge graph and performing preprocessing, a translational distance model is used to model the relationships in the knowledge graph to generate representations of initial entities and relationship vectors, which are input into a model that performs graph convolution operations with combined operation characteristics to initialize node representations. Graph convolution operations are performed in a graph convolutional network and combined with an attention mechanism to optimize the update of node representations. Training and cost function optimization are carried out iteratively to improve the model's reasoning ability. Finally, based on the trained model, the possible emissions trading behavior of users is inferred. The inference results are output in the form of triples and accompanied by confidence scores, and then the results are sorted and screened. The trading triples, associated entities, and recommendation priorities are displayed through a visualization interface to provide users with visual decision-making guidance. Summary of the Invention

[0005] The object of the present invention is to provide a knowledge reasoning method for basin pollution discharge trading based on TransE and CompGCN, aiming to significantly improve the matching efficiency, reasoning accuracy and user decision-making support ability of basin pollution discharge right trading through innovative technical improvements. Based on the existing technology CompGCN, this application proposes an intelligent method integrating TransE model embedding generation and CompGCN graph convolution reasoning. Through efficient knowledge graph modeling and dynamic trading reasoning, it provides actionable trading recommendations and visual guidance for enterprises and managers. The present invention first constructs a basin knowledge graph by integrating multi-source data, including enterprise emission records, pollution discharge right trading history, river topology structure and environmental capacity indicators. These data are cleaned, denoised and triple-formatted to form a structured knowledge graph containing entities and relationships. Subsequently, the present invention uses a translational distance model for relationship modeling and generates initial embeddings through a vector translation mechanism. Specifically, the translational distance model represents each entity and relationship as a low-dimensional vector, ensuring that the relationships between entities satisfy the translation equation enterprise A + trading pollution discharge right X = enterprise B, thereby capturing the core semantic features in basin trading. The generated embeddings have high discrimination and low computational complexity, providing a high-quality starting point for subsequent reasoning. Then, the present invention inputs the embeddings generated by the translational distance model into the graph convolutional network model as the initial feature vectors of nodes and relationships. Through graph convolutional operations, it aggregates the information of neighbor nodes in the knowledge graph and generates richer node representations using weighted sum, matrix multiplication or circular correlation combination operations. To adapt to the dynamics of basin pollution discharge right trading, the present invention designs an attention-based node update mechanism that dynamically assigns weights by calculating the semantic importance of neighbor nodes, thereby highlighting key trading patterns and environmental constraints. This mechanism significantly enhances the model's ability to model complex graph structures and can handle the dynamic changes in the trading demands between enterprises. In the reasoning stage, the present invention trains the model through an optimally designed cost function and iteratively updates the parameters of the graph convolutional network. During the training process, the model learns the potential trading patterns of the likelihood of pollution discharge right buying and selling or the feasibility weights of trading between enterprises in the basin knowledge graph, and the reasoning results are output in the form of triples enterprise B - selling pollution discharge right Y - enterprise A, along with confidence scores for users' reference. To further improve the user experience, the present invention presents the reasoning results through a visual interface, showing trading triples, associated entities and recommendation priorities, helping enterprises intuitively understand potential trading opportunities and assisting managers in formulating precise resource allocation strategies.

[0006] The technical solution adopted by the present invention is as follows:

[0007] The embodiment of this application provides a knowledge reasoning method for basin pollution discharge trading based on TransE and CompGCN. The specific steps of this reasoning method are as follows:

[0008] S1: Collect relevant data of the basin in real time, covering enterprise emission records, history of pollution rights trading, river topology, and environmental capacity indicators, etc. Clean the data and construct the data of the basin knowledge graph;

[0009] S2: Use the TransE model to perform relationship modeling on the basin knowledge graph to generate representations of initial entities and relationship vectors;

[0010] S3: Through the development of an embedding vector optimization algorithm for deep feature extraction and dimensionality reduction, and the adoption of an adaptive normalization method, optimize the embedding input and complete the initialization of the CompGCN model;

[0011] S4: Perform graph convolution operations in CompGCN, fuse the advantages of addition, multiplication, and convolution operations to comprehensively aggregate neighbor information, and optimize the update of node representations in combination with the attention mechanism;

[0012] S5: Through an iterative method, expand the function optimization of training and evaluating the performance deviation of the model to improve the reasoning ability of the model;

[0013] S6: Based on the trained model, infer the possible pollution rights trading behaviors of users and generate visual recommendation results.

[0014] As a further solution of this plan, the specific steps of S1 are as follows:

[0015] S1.1: Collect relevant data of the basin in real time from multiple sources such as the environmental protection department's supervision platform, enterprise internal emission monitoring system, and basin management agency database, covering enterprise emission records, history of pollution rights trading, river topology, and environmental capacity indicators, etc. Construct a knowledge graph containing entity enterprises and relationship transactions;

[0016] S1.2: Preprocess the data, including removing noise, filling in missing values, and standardizing the triple format <Enterprise A, purchase / sale, Pollution Right X, Enterprise B> to ensure data quality and consistency;

[0017] As a further solution of this plan, the specific steps of S2 are as follows:

[0018] S2.1: Based on the translation distance model, construct the relationship representation of the basin knowledge graph, map entities and relationships to low-dimensional vectors, and satisfy the translation equation: for the triple (h, r, t), the embedding vectors satisfy h + r ≈ t, where h and t are entity vectors and r is the relationship vector;

[0019] S2.2: Perform pre-training of the TransE model by optimizing the following cost function to generate high-quality embeddings. The cost function formula of the TransE model is as follows:

[0020]

[0021] Among them, S is the set of positive triples, S' is the set of negative triples, d is the Euclidean distance ||x - y||, γ is the margin hyperparameter, and γ = 1;

[0022] S2.3: Save the generated embedding vectors in matrix format and store them in the local system to provide initial feature inputs for CompGCN.

[0023] As a further solution of this scheme, the specific steps of S3 are as follows:

[0024] S3.1: Load the embedding vectors generated by TransE as the initial feature vectors of entities and relationships in CompGCN, ensuring that the dimensions match the input layer of CompGCN;

[0025] S3.2: Perform normalization on the embedding vectors using the z-score normalization formula:

[0026]

[0027] where x is the original vector value, μ is the mean, and σ is the standard deviation, ensuring the numerical stability of the embedding vectors;

[0028] S3.3: Initialize the CompGCN model and set the hyperparameters, including the hidden layer dimension 128, the sub and mult combination operation types, and the initial learning rate 0.001.

[0029] As a further solution of this scheme, the specific steps of S4 are as follows:

[0030] S4.1: Perform graph convolution operations in CompGCN. For node v, its representation update formula is:

[0031]

[0032] where hv(l + 1) represents the embedding vector of node v at the (l + 1)-th layer, σ is the non-linear activation function, N(v) represents the set of neighbor nodes of node v, R represents the set of relationships, Wr(l) is the weight matrix related to relationship r in the l-th layer; φ(hu(l), r) is the combination operation on the embedding hu(l) of neighbor node u and relationship r.

[0033] S4.2: Introduce the attention mechanism to calculate the attention weight of neighbor node u to node v:

[0034]

[0035] Among them, α(u, v, r) represents the attention weight of neighbor node u to node v, a is the attention parameter vector used to calculate the attention score, || represents the vector concatenation operation; hu and hv are the embedding vectors of node u and node v respectively, r is the representation vector of the relationship vector connecting u and v, N(v) represents the set of neighbor nodes of node v, and LeakyReLU is a non-linear activation function used to introduce non-linearity;

[0036] S4.3: Update the node representation to generate a feature vector that captures the transaction pattern and environmental constraints.

[0037] As a further solution of this scheme, the specific steps of S5 are as follows:

[0038] S5.1: Design a contrastive cost function based on triple scoring to optimize the model inference ability:

[0039]

[0040] Among them, f(h, r, t) is the triple scoring function, S represents the set of positive triples, S′ represents the set of negative triples, f(h, r, t) is the scoring function of the model for the triple (h, r, t), and γ is the margin hyperparameter used to control the minimum distance between positive and negative triples;

[0041] S5.2: Conduct cyclic training in the local computing environment, use the Adam optimizer, set the batch size to 128, set the number of training epochs to 200, and save the model checkpoint regularly;

[0042] S5.3: Monitor the Mean Reciprocal Rank (MRR) and Hits@10 metrics on the validation set, dynamically adjust the learning rate, and ensure fast convergence.

[0043] As a further solution of this scheme, the specific steps of S6 are as follows:

[0044] S6.1: Based on the trained CompGCN model, infer potential transaction triples and calculate the confidence score of the triple (h, r, t):

[0045] s(h, r, t) = sigmoid(W s [h|r||t]+b s ) (6)

[0046] Among them, s(h, r, t) represents the confidence score of the triple (h, r, t), with a value range of [0, 1], indicating the feasibility of the transaction. Ws is the scoring weight matrix for linear transformation, bs is the bias term, h, r, and t are the embedding vectors of the head entity, relation, and tail entity respectively, and sigmoid(x) is the activation function that maps the score to the interval [0, 1].

[0047] S6.2: Sort the inference results by confidence, filter out high-confidence triples (Enterprise A - Purchase - Emission Rights X - Enterprise B, s = 0.95), and generate a transaction recommendation list.

[0048] S6.3: Construct a basin knowledge graph to display transaction triples, associated entities, and priorities, providing intuitive decision-making guidance for users and supporting the matching optimization and resource allocation of basin emission rights trading.

[0049] The beneficial effects of this application are as follows:

[0050] 1. In traditional basin emission rights trading, information asymmetry makes it difficult for both trading parties to obtain transparent trading data, such as emission rights quotas and trading prices, increasing trading risks. By constructing a basin knowledge graph, this invention uses the translational distance model to generate high-quality representations of entity and relation vectors, representing trading entity enterprises and trading types in the form of structured triples. These embeddings are further optimized through graph convolutional operations to ensure that the knowledge graph can accurately reflect the semantic associations of transactions. The generated knowledge graph is publicly available to all participants, and trading information can be queried at any time, significantly improving the transparency of transactions and the reliability of data, and enhancing the supervision effects of regulatory authorities and the public.

[0051] 2. This invention uses knowledge graph technology to convert basin emission rights trading data into a standard triple format, constructing a structured semantic network to visually display the associations between trading entities, trading relationships, and environmental information. Users can query trading histories, potential trading opportunities, and basin water quality conditions through a visualization interface, clearly understanding market dynamics. Compared with traditional data display methods, the hierarchical presentation method of the knowledge graph greatly facilitates users' information acquisition and analysis, providing a scientific decision-making basis for enterprises to formulate trading strategies and managers to optimize resource allocation.

[0052] 3. As the core inference tool, the graph convolutional network takes advantage of its ability to process complex graph structures to deeply mine the multi-dimensional information in the basin knowledge graph. Through graph convolutional operations and attention mechanisms, the model can capture the implicit connections between entities, such as the potential trading preferences between enterprises or the relevance of emission behaviors, and infer trading opportunities that are not directly observable. The inference results are presented in the form of triples, accompanied by confidence scores, and visually displayed to users using visualization tools. Users can make trading decisions or policy adjustments based on the inference results, significantly improving the accuracy and efficiency of decision-making. This invention is particularly applicable to the knowledge graph completion and trading recommendation tasks in basin pollution rights trading, providing intelligent support for water resource management and environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application.

[0054] Figure 1 : It is a flowchart of the basin knowledge graph inference method based on TransE and CompGCN of the present application, showing the complete process from data collection, knowledge graph construction to trading inference and visualization recommendation;

[0055] Figure 2 : It is an example diagram of the TransE embedding generation model of the present application, showing the generation process of the representations of entity and relationship vectors and their representations in the basin knowledge graph;

[0056] Figure 3 : It is an example diagram of the CompGCN inference model of the present application, showing the implementation structure of graph convolutional operations; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments.

[0058] Embodiment 1

[0059] Refer to Figures 1 - 3 , the embodiments of the present application provide a basin knowledge graph inference method based on TransE and CompGCN. The specific steps of this method are as follows:

[0060] S1: Construct basin knowledge graph data to provide support for pollution rights trading inference. The specific steps are as follows:

[0061] S1.1: Collect relevant data on the basin, including enterprise emission records, pollution rights trading history, river topology, and environmental capacity indicators, to form an initial dataset containing entities and relationships.

[0062] S1.2: Preprocess the data, including removing duplicate records, filling in missing emission data, and converting trading records into triple format. The cleaned data is stored in tabular form, and example data is as follows:

[0063] Entity ID Entity type Emission Transaction amount Relationship Target entity a1 Enterprise 400 100 Purchase a4 b2 Enterprise 263 60 Sell b1 a3 Enterprise 300 80 Purchase a8 ... ... ... ... ... ...

[0064] S2: Use the translational distance model to perform relationship modeling on the basin knowledge graph, generating representations of initial entities and relationships. The specific steps are as follows:

[0065] S2.1: Based on the translational distance model, map the triples in the knowledge graph into low-dimensional vector representations to ensure effective capture of semantic associations between enterprises, pollution rights, and trading relationships.

[0066] S2.2: Generate embeddings by optimizing the translational distance model, set the embedding dimension to 100, train for 500 rounds, and the negative sampling ratio to 1:253 to generate high-quality entity and relationship vectors. The trained embeddings record partial results in tabular form, and example data is as follows:

[0067] Entity / Relationship Type Embedding vector a1 Enterprise [0.14,-0.65,0.48,0.13,0.56] Purchase Relationship [-0.25,0.23,0.62,0.57,0.15] ... ... ...

[0068] S2.3: Save the embedding vectors in the form of NumPy arrays and store them in the local file system (such as / data / entity_embedding_transe.pt) to provide initial feature inputs for the combined graph convolutional network model.

[0069] S3: Input the pre-trained embeddings generated by the TransE model into the combined graph convolutional network to initialize the node representations of the graph neural network. The specific steps are as follows:

[0070] S3.1: Load the embedding vectors generated by the TransE model as the initial feature vectors of entities and relationships in the combined graph convolutional network, and check the dimension consistency (100 dimensions);

[0071] S3.2: Standardize the embedding vectors to ensure stable numerical distribution and prevent gradient anomalies during training. The standardized vectors are recorded in tabular form for easy tracking of the data processing process.

[0072] S3.3: Initialize the combined graph convolutional network model, set the hidden layer dimension to 128, the combination operation type to weighted sum, the initial learning rate to 0.001, and use the ReLU activation function.

[0073] S4: Perform graph convolution operations in the combined graph convolutional network, and optimize the update of node representations by combining the attention mechanism. The specific steps are as follows:

[0074] S4.1: Through the graph convolution operation of the combined graph convolutional network, aggregate the information of neighbor nodes in the knowledge graph, and use the weighted sum combination operation to generate new node representations;

[0075] S4.2: Introduce the attention mechanism to dynamically calculate the weights of neighbor nodes. The updated node representations are recorded in tabular form. The example data is as follows:

[0076]

[0077]

[0078] S4.3: Generate node representations that capture trading patterns and environmental constraints, and store them in the local file system.

[0079] S5: Optimize the function for training and evaluating the performance deviation of the model in an iterative manner to improve the inference ability of the model. The specific steps are as follows:

[0080] S5.1: Design a contrast cost function based on triple scoring to optimize the prediction ability of the combined graph convolutional network model for trading triples. Set the margin parameter to 0.5. The cost function enhances the model's ability to distinguish potential trading relationships by comparing the scores of positive and negative triples;

[0081] S5.2: Train the model with a batch size of 128 for 200 rounds. Decay the learning rate by 0.1 times every 50 rounds, and save the best model checkpoint. The performance metrics during training are recorded in tabular form. The example data is as follows:

[0082] Number of rounds MRR Hits@10 Loss value Training time 50 0.31 0.622 0.77 xxxx-xx-xx 12:00 100 0.40 0.75 0.43 xxxx-xx-xx 16:00 ... ... ... ... ...

[0083] S5.3: Evaluate the Mean Reciprocal Rank (MRR) and Hits@10 metrics on the validation set. The target is MRR > 0.7 and Hits@10 > 0.85 to confirm the convergence of the model.

[0084] S6: Based on the trained model, infer the possible emission rights trading behaviors of users and generate visual recommendation results. The specific steps are as follows:

[0085] S6.1: Use the trained CompGCN model to infer potential trading triples, generate confidence scores, and filter triples with a confidence > 0.8. The inference results are recorded in tabular form. The example data is as follows:

[0086] Triple Confidence score Recommendation priority River section number Transaction time (a2, Purchase, a3) 0.95 1 1 xxxx-xx-xx (a7, Purchase, a5) 0.90 2 2 xxxx-xx-xx ... ... ... ... ...

[0087] S6.2: Sort by confidence level, generate a transaction recommendation list, and store it in a local database (such as SQLite) for users to query;

[0088] S6.3: Use Neo4j to visualize the knowledge graph, display transaction triples, associated entities (the buying and selling relationship between enterprise A and pollution discharge right X), and recommendation priorities, present the transaction network and water quality status, and assist enterprises in formulating transaction strategies and managers in optimizing resource allocation.

[0089] In summary, this example is a knowledge reasoning method for watershed pollution discharge trading based on TransE and CompGCN. It generates high-quality embeddings through the TransE model and conducts accurate transaction reasoning through the combined graph convolutional network model, constructs an efficient watershed pollution discharge right trading support system, and uses the visualization of the knowledge graph to provide users with intuitive transaction recommendations and decision-making guidance.

[0090] Note that the above content only describes the preferred implementation manner of the embodiments of this application and the technical principles applied. The embodiments of this application are not limited to the specific examples detailed here. For those skilled in the art, various obvious modifications, reconfigurations, and replacements can be made without departing from the core idea of the embodiments of this application. Although the embodiments of this application have been described in more detail through the above specific embodiments, the embodiments of this application go far beyond this. Without departing from the basic concept of the embodiments of this application, more other equivalent implementation manners can be covered. The protection scope of the embodiments of this application is defined by the appended claims.

[0091] The above has described an embodiment of this application in detail, but the content described is only the preferred embodiment of this application and cannot be considered as limiting the implementation scope of this application. All equivalent changes and improvements made in accordance with the scope of this application's application shall still fall within the scope covered by the patent of this application.

Claims

1. A knowledge reasoning method for basin pollution discharge trading based on TransE and Compgcn, characterized in that, The specific steps of this method are as follows: S1: Construct the basin knowledge graph data and prepare the data processing and model training environment; S2: Use the TransE model to perform relationship modeling on the basin knowledge graph to generate initial entity and relationship vector representations; S3: Combine the pre-trained input generated by the translation distance model with the graph convolutional network model to initialize the node representations of the graph neural network; S4: The model that performs graph convolutional operations using the combined operation characteristics performs graph convolutional operations, and combines the attention mechanism to optimize the update of node representations; S5: Through iterative training and cost function optimization, improve the inference ability of the model; S6: Based on the trained model, infer the possible emission rights trading behaviors of users and output the recommended results.

2. The knowledge reasoning method for basin pollution discharge trading based on TransE and Compgcn according to claim 1, wherein The S1 includes: S1.1: Collect basin-related data, including emission rights trading records, enterprise emission data, river topology structure, and environmental capacity indicators, and construct a basin knowledge graph; S1.2: Preprocess the knowledge graph data, including denoising, missing value filling, and triple formatting, to ensure data consistency and quality.

3. A knowledge reasoning method for basin pollution discharge trading based on TransE and Compgcn according to claim 1, characterized in that, The S2 includes: S2.1: Based on the translation distance model, construct the relationship representation of the basin knowledge graph, and use translation vectors to represent enterprise entities and relationships such as transactions to capture semantic associations; S2.2: Pre-train the translation distance model to optimize the embedding representation to ensure that the representations of entity and relationship vectors can accurately reflect the characteristics of basin emission rights trading; S2.3: Save the pre-trained embeddings in the format of a vector matrix and store them in the local system to provide high-quality initial representations for the model training that performs graph convolutional operations using the combined operation characteristics.

4. A knowledge inference method for basin pollution discharge trading based on TransE and Compgcn according to claim 1, characterized in that, The S3 includes: S3.1: Load the pre-trained embeddings generated by the translation distance model as the initial feature vectors of basin enterprise entities and relationships in the combined convolutional network; S3.2: Perform dimension alignment and normalization processing on the embedding vectors to ensure compatibility with the input layer of CompGCN and adapt to the characteristics of basin data; S3.3: Initialize the CompGCN model, set hyperparameters such as the hidden layer dimension, combined operation type, and learning rate, and prepare for graph convolutional operations and transaction inference.

5. A knowledge reasoning method for basin pollution discharge trading based on TransE and Compgcn according to claim 1, characterized in that, The S4 includes: S4.1: Perform graph convolutional operations in CompGCN, and aggregate the information of neighbor nodes in the basin knowledge graph through addition and multiplication combined operations; S4.2: Introduce the attention mechanism to dynamically allocate weights to neighbor nodes and enhance the representation ability for key transaction patterns and environmental features; S4.3: Update the node representations to generate feature vectors that reflect the complex relationships of basin emission rights trading, and capture transaction patterns and semantic information in the graph structure.

6. The method for inferring knowledge of basin pollution discharge trading based on TransE and Compgcn according to claim 1, wherein, The S5 includes: S5.1: Design a function for judging the performance deviation of the model suitable for basin knowledge graph inference, such as logarithmic loss or triple scoring loss based on negative sampling, to optimize the accuracy of transaction inference; S5.2: Conduct repeated training in the local computing environment, use the gradient descent algorithm to optimize the model parameters, and regularly save the training checkpoints; S5.3: Monitor the convergence speed and validation set performance during training, dynamically adjust the learning rate and batch size to ensure the model converges quickly and adapts to the emissions trading scenario.

7. A knowledge inference method for basin pollution discharge trading based on TransE and Compgcn according to claim 1, characterized in that The above S6 includes: S6.1: Based on the trained CompGCN model, infer potential trading behaviors in the watershed knowledge graph according to historical transaction data, generate possible emissions trading triples for users, and clearly show enterprise A - purchase emissions rights X - enterprise B; S6.2: Sort and filter the inference results, and use metrics such as triple scores or trading feasibility to output high-confidence trading recommendations; S6.3: Apply the inference results to the watershed emissions trading system, construct a watershed emissions trading knowledge graph, support knowledge graph completion and trading recommendations, optimize the trading matching efficiency and the accuracy of environmental resource allocation, and give users visual guidance.

Images