CN116843400A

CN116843400A - Block chain carbon emission transaction anomaly detection method and device based on graph representation learning

Info

Publication number: CN116843400A
Application number: CN202310333506.3A
Authority: CN
Inventors: 熊翱; 谯陈彬; 李文璟; 郭少勇; 王栋; 李达; 宋雨
Original assignee: State Grid Digital Technology Holdings Co ltd; Beijing University of Posts and Telecommunications
Current assignee: State Grid Digital Technology Holdings Co ltd; Beijing University of Posts and Telecommunications
Priority date: 2023-03-30
Filing date: 2023-03-30
Publication date: 2023-10-03

Abstract

The application provides a blockchain carbon emission transaction anomaly detection method and device based on graph representation learning, wherein the method comprises the following steps: inputting a current adjacency matrix of a blockchain network for carbon emission transactions and an attribute matrix representing attributes of each node in the blockchain network into a graph representation learning model based on a self-encoder and an attention mechanism to output a reconstructed adjacency matrix and a reconstructed attribute matrix; determining abnormal detection scores of all nodes in the blockchain network by using the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix; and determining abnormal nodes with abnormal risks of carbon emission transaction in the blockchain network according to the abnormal detection scores. The method and the device can effectively improve the accuracy and the effectiveness of abnormal detection of the blockchain carbon emission transaction, can effectively filter the influence of abnormal nodes and improve the capability of obtaining the normal characteristics of the nodes, and further can effectively ensure the safety and the reliability of the carbon emission transaction in the blockchain.

Description

Block chain carbon emission transaction anomaly detection method and device based on graph representation learning

Technical Field

The application relates to the technical field of information processing, in particular to a blockchain carbon emission transaction anomaly detection method and device based on graph representation learning.

Background

In the aspect of developing carbon transaction and carbon neutralization management, a plurality of challenges such as weak carbon data circulation trust, difficult carbon emission data identification and tracing, high participation cost of market main bodies, urgent need to perfect an electric carbon cooperative operation mechanism and the like are faced. The blockchain is used as a regularized digital technology, has the characteristics of multiparty consensus, transparent disclosure, tamper resistance, traceability and the like, is suitable for scenes of strong dependence of rules such as carbon monitoring, accounting, transaction and the like, can realize trusted recording of the whole life cycle of the carbon footprint and trusted circulation of the whole elements of carbon emission, and supports and constructs an electric carbon cooperative operation mechanism. Thus, blockchain technology may be combined with carbon emissions transactions, with both nodes and transaction characteristics of each carbon emissions transaction included in the carbon emissions transaction data. Both businesses and individuals have a data indicator of carbon emissions, and if either the business or the individual has a surplus carbon indicator or exceeds the carbon emission indicator, both parties can conduct carbon emission transactions. However, to further ensure the safety and reliability of carbon emission transactions, it is necessary to detect nodes in the blockchain that are conducting abnormal carbon emission transactions.

At present, the existing detection method for the nodes for carrying out abnormal carbon emission transaction in the blockchain ignores the interaction information between the blockchain network structure and the node attribute, so that the blockchain carbon emission transaction abnormality detection capability is limited, and the nodes for carrying out abnormal carbon emission transaction in the blockchain cannot be accurately detected.

Disclosure of Invention

In view of the above, embodiments of the present application provide a blockchain carbon emissions transaction anomaly detection method and apparatus based on graph representation learning to obviate or ameliorate one or more of the disadvantages of the prior art.

One aspect of the present application provides a blockchain carbon emission transaction anomaly detection method based on graph representation learning, comprising:

inputting a current adjacency matrix of a blockchain network for carbon emission transaction and an attribute matrix representing each node attribute in the blockchain network into a preset graph representation learning model based on a self-encoder and an attention mechanism, so that the graph representation learning model outputs a reconstructed adjacency matrix corresponding to the adjacency matrix and a reconstructed attribute matrix corresponding to the attribute matrix;

determining abnormal detection scores of all nodes in the blockchain network by using the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix;

and determining abnormal nodes with abnormal risk of carbon emission transaction in the blockchain network according to the abnormal detection scores of the nodes.

In some embodiments of the application, the self-encoder and attention mechanism based graph representation learning model comprises:

The structure encoder is used for correspondingly outputting the structure embedded representation corresponding to the network structure characteristics of each node in the blockchain network according to the input adjacency matrix of the blockchain network;

the attribute encoder is used for outputting attribute embedded representations corresponding to attribute characteristics of all attribute information in the blockchain network according to an input attribute matrix of the blockchain network, and the attribute encoder and the structure encoder are both the self-encoders;

a graph attention layer for determining importance weights between every two nodes in the blockchain network based on the input adjacency matrix and the attribute matrix by adopting a shared attention mechanism to obtain node embedded representations for representing node characteristics of the nodes;

the structure decoder is used for reconstructing the adjacent matrix according to the input structure embedded representation and the node embedded representation and outputting a corresponding reconstructed adjacent matrix;

and the attribute decoder is used for reconstructing the attribute matrix according to the input attribute embedded representation and the node embedded representation and outputting a corresponding reconstructed attribute matrix.

In some embodiments of the present application, the inputting the current adjacency matrix of the blockchain network for the carbon emission transaction and the attribute matrix representing the attributes of each node in the blockchain network into a preset self-encoder and attention mechanism-based graph representation learning model, so that the graph representation learning model outputs a reconstructed adjacency matrix corresponding to the adjacency matrix and a reconstructed attribute matrix corresponding to the attribute matrix, includes:

Inputting a current adjacency matrix of a blockchain network for carbon emission transaction into the structural encoder so that the structural encoder outputs a structural embedded representation of network structural features of each node in the blockchain network at an hidden layer;

inputting the adjacency matrix and the current attribute matrix of the blockchain network into the graph attention layer so that the graph attention layer adopts a shared attention mechanism to determine importance weight values between every two nodes in the blockchain network, and outputting node embedded representations for representing node characteristics of the nodes;

inputting the attribute matrix into the attribute encoder so that the attribute encoder outputs attribute embedded representations corresponding to attribute features of all attribute information in the blockchain network in an implicit layer;

inputting the structure embedded representation and the node embedded representation into the structure decoder so that the structure decoder reconstructs the adjacent matrix and outputs a corresponding reconstructed adjacent matrix;

and inputting the attribute embedded representation and the node embedded representation into the attribute decoder so that the attribute decoder reconstructs the attribute matrix and outputs a corresponding reconstructed attribute matrix.

In some embodiments of the present application, the inputting the adjacency matrix and the current attribute matrix of the blockchain network into the graph attention layer, so that the graph attention layer uses a shared attention mechanism to determine importance weights between every two nodes in the blockchain network, so as to output node embedded representations for representing node characteristics of the nodes, includes:

inputting the adjacency matrix and the current attribute matrix of the blockchain network into the graph meaning layer, so that the graph meaning layer firstly obtains node feature vectors mapped to potential space based on the attribute matrix; and determining importance weights between every two nodes in the block chain network based on the node feature vectors and the adjacency matrix by adopting a shared attention mechanism, and determining node embedded representations for representing node features of the nodes according to the importance weights between every two nodes.

In some embodiments of the present application, before the inputting the current adjacency matrix of the blockchain network for the carbon emission transaction and the attribute matrix representing the attributes of each node in the blockchain network into the preset self-encoder and attention mechanism-based graph representation learning model, the graph representation learning model outputs a reconstructed adjacency matrix corresponding to the adjacency matrix and a reconstructed attribute matrix corresponding to the attribute matrix, the method further comprises:

Acquiring current graph data of a blockchain network for carbon emission transaction, wherein the graph data comprises all nodes participating in the carbon emission transaction in the blockchain network, relations among the nodes and attribute information of all the nodes;

generating a corresponding adjacency matrix and an attribute matrix representing attributes of each node in the blockchain network based on the graph data.

In some embodiments of the present application, the determining the anomaly detection score for each node in the blockchain network using the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix, and the reconstructed attribute matrix includes:

constructing an objective function comprising super parameters and parameters respectively representing an adjacency matrix, a reconstructed adjacency matrix, an attribute matrix and a reconstructed attribute matrix by taking the adjacency matrix and the minimum reconstruction error of the attribute matrix representing each node attribute in the blockchain network as targets;

substituting the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix into the objective function and solving to obtain reconstruction error values corresponding to each node in the blockchain network;

and respectively determining the corresponding abnormal detection value of each node based on the reconstruction error value corresponding to each node.

In some embodiments of the present application, the determining, according to the anomaly detection score of each node, an anomaly node in the blockchain network at which there is a risk of carbon emission transaction anomaly includes:

ranking the nodes in order of the anomaly detection score from small;

selecting a preset threshold number of nodes from the first one of the arranged nodes to be marked as abnormal nodes with abnormal risk of carbon emission transaction, or marking nodes larger than the abnormal threshold as abnormal nodes with abnormal risk of carbon emission transaction;

and generating abnormal risk detection result data of the carbon emission transaction aiming at the blockchain network based on each abnormal node, and outputting the detection result data.

Another aspect of the present application provides a blockchain carbon emission transaction anomaly detection device based on graph representation learning, comprising:

the model detection module is used for inputting a current adjacency matrix of a blockchain network for carbon emission transaction and an attribute matrix representing the attribute of each node in the blockchain network into a preset graph representation learning model based on a self-encoder and an attention mechanism, so that the graph representation learning model outputs a reconstructed adjacency matrix corresponding to the adjacency matrix and a reconstructed attribute matrix corresponding to the attribute matrix;

The score calculation module is used for determining the abnormal detection scores of all nodes in the block chain network by applying the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix;

and the abnormality determining module is used for determining abnormal nodes with abnormal risks of carbon emission transaction in the blockchain network according to the abnormality detection scores of the nodes.

In a third aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the blockchain carbon emission transaction anomaly detection method based on graph representation learning when executing the computer program.

A fourth aspect of the present application provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the blockchain carbon emissions transaction anomaly detection method based on graph representation learning.

The application provides a blockchain carbon emission transaction anomaly detection method based on graph representation learning, which is characterized in that a current adjacent matrix of a blockchain network for carbon emission transaction and an attribute matrix representing each node attribute in the blockchain network are input into a preset graph representation learning model based on a self-encoder and an attention mechanism, so that the graph representation learning model outputs a reconstructed adjacent matrix corresponding to the adjacent matrix and a reconstructed attribute matrix corresponding to the attribute matrix; determining abnormal detection scores of all nodes in the blockchain network by using the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix; determining abnormal nodes with abnormal risk of carbon emission transaction in the blockchain network according to the abnormal detection scores of the nodes; by adopting an adjacency matrix for representing the blockchain network structure, an attribute matrix for representing node attributes and a graph representation learning model, cross-modal interaction in a network can be captured, deeper network characteristics are extracted, and accuracy and effectiveness of blockchain carbon emission transaction anomaly detection can be effectively improved; by introducing an attention mechanism, the influence of abnormal nodes can be effectively filtered when the graph representation learning model is extracted from the characteristics, the capability of obtaining normal characteristics of the nodes can be effectively improved, the influence of the abnormal nodes is weakened, the accuracy and reliability of abnormal detection of the blockchain carbon emission transaction can be further improved, and the safety and reliability of the carbon emission transaction in the blockchain can be further effectively ensured.

Additional advantages, objects, and features of the application will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and drawings.

It will be appreciated by those skilled in the art that the objects and advantages that can be achieved with the present application are not limited to the above-described specific ones, and that the above and other objects that can be achieved with the present application will be more clearly understood from the following detailed description.

Drawings

The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate and together with the description serve to explain the application. The components in the figures are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the application. Corresponding parts in the drawings may be exaggerated, i.e. made larger relative to other parts in an exemplary device actually manufactured according to the present application, for convenience in showing and describing some parts of the present application. In the drawings:

fig. 1 is a schematic flow chart of a blockchain carbon emission transaction anomaly detection method based on graph representation learning according to an embodiment of the application.

FIG. 2 is a schematic diagram of a self-encoder and attention mechanism based diagram illustrating a learning model in an embodiment of the present application.

FIG. 3 is a schematic diagram of a second flow chart of a blockchain carbon emissions transaction anomaly detection method based on graph representation learning in an embodiment of the application.

Fig. 4 is a schematic structural diagram of a blockchain carbon emission transaction anomaly detection device based on graph representation learning in another embodiment of the present application.

Detailed Description

The present application will be described in further detail with reference to the following embodiments and the accompanying drawings, in order to make the objects, technical solutions and advantages of the present application more apparent. The exemplary embodiments of the present application and the descriptions thereof are used herein to explain the present application, but are not intended to limit the application.

It should be noted here that, in order to avoid obscuring the present application due to unnecessary details, only structures and/or processing steps closely related to the solution according to the present application are shown in the drawings, while other details not greatly related to the present application are omitted.

It should be emphasized that the term "comprises/comprising" when used herein is taken to specify the presence of stated features, elements, steps or components, but does not preclude the presence or addition of one or more other features, elements, steps or components.

It is also noted herein that the term "coupled" may refer to not only a direct connection, but also an indirect connection in which an intermediate is present, unless otherwise specified.

Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

In one or more embodiments of the application, abnormal node detection is performed in carbon emission transaction data, which includes both nodes and transaction characteristics for each transaction. Both businesses and individuals have a data indicator of carbon emissions, and if either the business or the individual has a surplus carbon indicator or exceeds the carbon emission indicator, both parties can conduct carbon emission transactions. The value of this transaction is one of the carbon emission transaction data.

In one or more embodiments of the present application, a node refers to a transaction node in carbon emission transaction data, two transaction parties like bank transfers, an abnormal node is a node for performing an abnormal carbon emission transaction, and the purpose of the present application is to find a node in carbon emission transaction data where an abnormal carbon emission transaction exists.

In one or more embodiments of the application, the blockchain is essentially a distributed database technology with features of decentralization, non-tamper-ability, global marking, anonymity, etc. The great innovation of the blockchain technology provides a plurality of challenges for financial services, ecological security and privacy protection. To prevent risks, cryptocurrency is gradually brought into the monitoring domain at the monitoring level; at the technical level, the disclosed data also facilitates more organizations to analyze blockchains. It also shows that applications built solely using blockchain technology are difficult to bring into play, which requires blockchains in combination with other technologies such as artificial intelligence, big data, etc., to form an integrated solution using synergistic effects.

Many common relationships in human society can be abstracted into representations of network structures, such as document citation networks, biomolecular networks, server cluster networks, and the like. These network relationships are formally represented as a topology made up of a number of nodes. The network structure comprises connection relations among all nodes, complex information interaction can be reserved, and meanwhile, each node comprises rich attribute information, so that the network is called an attribute network. In one or more embodiments of the present application, the blockchain network may also be abstracted into an attribute network, and from a structural view, the blockchain data is essentially represented by a graph, where nodes represent addresses, edges represent transactions, and the like, and after a transaction graph is constructed, the model may extract high-dimensional features in the graph structural relationship.

In one or more embodiments of the present application, although it is difficult to implement the supervision of blockchain transactions using conventional means, the complete transaction data is publicly transparent on the blockchain, mining the transaction data on the chain, creating a multidimensional data model, and implementing a data-driven intelligent supervision scheme using techniques such as big data and artificial intelligence has the primary objective of accurately identifying abnormal transactions on the blockchain: a suspicious user (e.g., money laundering organization member) or suspicious transaction (e.g., credit card fraud) therein is identified.

In one or more embodiments of the present application, anomaly detection is a technique for identifying an anomaly pattern that does not conform to an expected behavior, discovering an anomaly connection structure present in a network, a node that presents an anomaly behavior, a node that contains anomaly information, and the like. Therefore, in order to implement blockchain intelligent supervision, to assess the risk of potential money laundering, terrorist financing and other financial crimes for customers, it is necessary to perform anomaly detection on the blockchain network. However, the existing anomaly detection method ignores the interaction information between the network structure and the node attribute, and has limited anomaly detection capability.

The rapid development of blockchain technology has prompted the transformation of information Internet to value Internet, but the application scenario is wide, and many risk behaviors such as money laundering, tax evasion, illegal ICO financing and the like are derived. Abnormal transaction detection has been a positive impetus for the healthy development of the blockchain industry. Conventional approaches to abnormal transaction detection are to design an alarm system based on fixed threshold rules to detect and flag suspicious transactions and then make manual decisions or decisions on suspicious activity. The advent of internet finance such as virtual "digital money" has led to a great challenge for rule-based supervision schemes. Breaking the traditional supervision thinking, constructing an intelligent supervision scheme based on data and by using artificial intelligence, big data analysis and other technologies as means has become a trend. Research work for abnormal transaction detection in intelligent supervision is mainly focused on supervised learning and unsupervised learning. Supervised learning predicts classification of unknown data samples (test sets) by using labeled data (training sets) to learn differentiated bi-classification (e.g., legal versus illegal transactions) or multi-classification machine learning detection models. And searching the structure and the characteristics of unlabeled data in the unsupervised learning, finding out the optimal division of clusters or classes, and taking the orphans far away from other sample points as outliers, namely abnormal data. For example, a learner trains an XGBoost supervised predictive model using information such as the background of the sender/receiver, early transaction actions, and transaction history to identify potential money laundering actions in financial transactions and apply to banks. Also, scholars and the like extract 18 important features from registration information, financial transactions, electronic invoices and other related categories and train an unsupervised deep learning model in combination with an auto-encoder (AE) algorithm to detect export fraud related to back-flushing.

In one or more embodiments of the application, the graph neural network (Graph Nerual Network, GNN) is an application of deep learning on graph structure data and derives a variety of application-specific methods and models. In the graph data, each graph has unordered nodes of different scales, and the number of adjacent nodes of each node is randomly distributed. In recent years, many studies have begun focusing on deep learning methods for graph data, which indirectly optimize graph embedding when reconstructing original spatial data, require that the vector retain as much as possible the structural information, attribute information, etc. of nodes in the graph, and are widely applicable in various structures such as social networks, cooperative networks, protein networks, etc. Transactions on the blockchain can also be mapped into financial networks with users as nodes and transactions among users as edges, such as Weber and the like, map the transactions into a huge and complex graph structure, extract related characteristics of transaction quantity, transaction amount and the like, and then distinguish illegal and legal transactions by adopting a graph roll-up network (GCN, graph convolution network) algorithm.

In one or more embodiments of the present application, the self-encoder is an artificial neural network, which is generally used for dimension reduction, and the final purpose is to generate a high-quality embedded representation of data through learning, the embedded representation is obtained through the encoder, the self-encoding mode is widely applied in a data generation model, and currently, in a deep learning network, feature embedding can be effectively generated by using a technical means of the self-encoder as feature extraction, so that the learning effect and the training speed are improved, and the potential representation of the data in the hidden space is learned in an unsupervised mode through superposition of multiple layers of encoding individual decoding functions, so that good effects are achieved in various fields. The data is input to the self-encoder, the encoder performs feature extraction, and the decoder reconstructs the original data according to the learned input features.

In one or more embodiments of the present application, the attention mechanism is initially designed to be sequence-based, and is mainly used for quickly screening valuable information from a large amount of information, and the biggest effect is to increase the weight of important content in data during information processing. Common sequence tasks such as machine translation, natural language processing produce good results after the introduction of attention mechanisms. The quality of node embedding is obtained to a great extent to determine the effect of matrix reconstruction, the attention mechanism is introduced into the generation process of node embedding, normal node characteristics can be reserved to the maximum extent by using the attention mechanism aiming at the network containing abnormal data, and the influence of abnormal information on the generation of node embedding is weakened.

The GCN model solves the problem of feature extraction of irregular space structure data, but in an anomaly detection task, due to the fact that abnormal nodes exist in a data set, the characteristics of normal nodes are easily affected by anomaly information when the characteristics of the normal nodes are extracted, and therefore embedding quality and reconstruction effects are reduced.

Because the prior art ignores the interaction information between the blockchain network structure and the node attribute, the prior blockchain carbon emission transaction anomaly detection capability is limited, the problems that the nodes for carrying out abnormal carbon emission transaction in the blockchain cannot be accurately detected, and the like are solved. In addition, attention mechanisms are introduced to learn the importance between a node and its neighbors. First the structure encoder converts the observed original node properties into a vector representation of the low-order potential space, then aggregates the embedded representations of all neighboring nodes using a shared-attention mechanism, ultimately generating node embeddings. The attribute encoder uses a multi-layer perceptron to map observed attribute data into a potential attribute embedded representation. The adjacency matrix is then reconstructed using a structure decoder, which reconstructs the attribute matrix, and the reconstruction errors of the nodes are measured from both structure and attribute angles as an objective function of neural network training. Anomaly detection is then implemented based on the reconstruction errors of the two angle measurement nodes for the structure and attributes.

The following examples are provided to illustrate the application in more detail.

The embodiment of the application provides a blockchain carbon emission transaction anomaly detection method based on graph representation learning, referring to fig. 1, the blockchain carbon emission transaction anomaly detection method based on graph representation learning, which can be executed by a blockchain carbon emission transaction anomaly detection device based on graph representation learning, specifically comprises the following contents:

step 100: inputting a current adjacency matrix of a blockchain network for carbon emission transaction and an attribute matrix representing the attribute of each node in the blockchain network into a preset graph representation learning model based on a self-encoder and an attention mechanism, so that the graph representation learning model outputs a reconstructed adjacency matrix corresponding to the adjacency matrix and a reconstructed attribute matrix corresponding to the attribute matrix.

It will be appreciated that adjacency matrix refers to adjacency matrix of graph data of blockchain networks for carbon emissions transactions, which can be written as A, A ε R ^{N×N} Wherein R represents a real set; n represents the number of nodes in the graph. The adjacency matrix is used for providing neighbor node information of each node in the blockchain network to represent network structure characteristics of the blockchain network. And, the attribute matrix refers to all attribute information contained by each node in the blockchain network for carbon emissions transactions.

In one or more embodiments of the present application, the graph representation learning model based on the self-encoder and the attention mechanism refers to a graph representation learning model provided with the self-encoder and the attention mechanism functions.

Step 200: and determining abnormal detection scores of all nodes in the blockchain network by using the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix.

Step 300: and determining abnormal nodes with abnormal risk of carbon emission transaction in the blockchain network according to the abnormal detection scores of the nodes.

In step 200 and step 300, in the anomaly detection stage, the connection information and attribute information of each node are integrated to calculate an anomaly score thereof according to the reconstruction error of the matrix. The nodes are arranged in descending order according to the anomaly scores, and the higher the anomaly score is, the greater the probability that the node has anomalies is. Marking the nodes with the scores higher than a preset threshold as nodes with abnormal prediction, comparing the nodes with the real labels, and judging the detection effect of the algorithm by using various evaluation indexes. The value of the preset threshold is related to the distribution of the anomaly scores.

As can be seen from the above description, according to the blockchain carbon emission transaction anomaly detection method based on graph representation learning provided by the embodiment of the application, by adopting the adjacency matrix for representing the blockchain network structure, the attribute matrix for representing the node attribute and the graph representation learning model, the cross-modal interaction in the network can be captured, the deeper network characteristics can be extracted, and the accuracy and the effectiveness of blockchain carbon emission transaction anomaly detection can be effectively improved; by introducing an attention mechanism, the influence of abnormal nodes can be effectively filtered when the graph representation learning model is extracted from the characteristics, the capability of obtaining normal characteristics of the nodes can be effectively improved, the influence of the abnormal nodes is weakened, the accuracy and reliability of abnormal detection of the blockchain carbon emission transaction can be further improved, and the safety and reliability of the carbon emission transaction in the blockchain can be further effectively ensured.

In order to further improve reliability and safety of blockchain carbon emission transaction anomaly detection based on graph representation learning, in a blockchain carbon emission transaction anomaly detection method based on graph representation learning provided by an embodiment of the present application, referring to fig. 2, a graph representation learning model based on a self-encoder and an attention mechanism in the blockchain carbon emission transaction anomaly detection method based on graph representation learning specifically includes the following contents:

In particular, to capture potential cross-modal interactions between network structure and node attributes in graph data, the present application designs a self-encoder and attention mechanism based deep learning framework, namely the self-encoder and attention mechanism based graph representation learning model shown in FIG. 2. The model consists of two parts, namely an encoder and a decoder. The model takes a node embedded representation learned by a structure encoder in combination with a diagram attention layer and an attribute embedded representation learned by an attribute encoder as inputs, and the structure decoder and the attribute decoder jointly capture interaction between a network structure and node attributes in a training process. Finally, the reconstruction errors of the network structure and the node attributes are utilized to measure the anomalies in the attribute network. The framework extracts the characteristics in the attribute network from multiple angles, fuses the characteristic information of different types, and restores the initial information.

The design of the deep learning framework captures cross-modal interaction of the structure and attribute information through hidden space feature fusion. In the existing method, one embedded representation is used for reconstructing multiple types of information, and the feature extraction capability of a single embedded representation on initial data is limited.

In the application, two decoders of the framework select corresponding feature embedding to perform depth fusion according to the category of the reconstruction information, for example, when the attribute features are reconstructed, the feature extraction is performed on each dimension attribute, and the two decoders are combined with the node features to restore the accurate attribute information of each node, thereby minimizing the difference with the original matrix when the matrix is reconstructed.

The attribute network g= { V, epsilon, X } is defined as having m= |v| nodes each having n= |x|dimension attribute and epsilon edges, G representing the attribute network; v represents a node.

In the present application we focus on anomaly detection problems on attribute networks. The definition of this problem is as follows:

given an attribute network g= { V, epsilon, X }, our goal is to detect rare nodes that differ significantly from most reference nodes, including nodes that differ significantly from most reference nodes in both structural and attribute information. More precisely, formally, our goal is to learn a scoring function f According to threshold lambda pair sample x _i Classification is performed.

Wherein y is _i Representing sample x _i 0 is a normal class, 1 is an abnormal class; vi denotes the i-th node.

In order to capture potential cross-modal interactions between network structures and node attributes in graph data, the application designs a deep learning framework based on a self-encoder and an attention mechanism. The framework extracts the characteristics in the attribute network from multiple angles, fuses the characteristic information of different types, and restores the initial information.

The design of GraphAEAtt captures cross-modal interactions of structure and attribute information through hidden space feature fusion. In the existing method, one embedded representation is used for reconstructing multiple types of information, and the feature extraction capability of a single embedded representation on initial data is limited.

The two decoders of the GraphAEAtt framework select corresponding feature embedding to carry out deep fusion according to the category of the reconstruction information, for example, when the attribute features are reconstructed, feature extraction is carried out on each dimension attribute, and the two decoders are combined with the node features to restore the accurate attribute information of each node, so that the difference between the two decoders and the original matrix is minimized when the matrix is reconstructed.

The frame is composed of an encoder and a decoder. The method takes node embedded representation learned by a structure encoder in combination with a diagram attention layer and attribute embedded representation learned by an attribute encoder as inputs, and the structure decoder and the attribute decoder jointly capture interaction between a network structure and node attributes in the training process. Finally, the reconstruction errors of the network structure and the node attributes are utilized to measure the anomalies in the attribute network.

In order to further improve reliability and safety of blockchain carbon emission transaction anomaly detection based on graph representation learning, in the blockchain carbon emission transaction anomaly detection method based on graph representation learning provided by the embodiment of the application, referring to fig. 3, step 100 in the blockchain carbon emission transaction anomaly detection method based on graph representation learning specifically includes the following contents:

step 110: the current adjacency matrix of the blockchain network for the carbon emissions transaction is input to the structural encoder, such that the structural encoder outputs a structural embedded representation at an implicit layer of the network structural characteristics of each node in the blockchain network.

In the structure encoder, all attribute information in the data is ignored, and only node structure features are extracted.

The goal of the structural encoder is to derive an embedded representation by learning that can reflect global structural features in the network. The input is the adjacency matrix A, A epsilon R of the graph data ^{N×N} . Obtaining an embedded representation Z of a hidden layer by nonlinear feature transformation ^s As shown in equation 2:

in the case of the formula 2 of the present application,an embedded representation representing a hidden layer; sigma represents an activation function;Representing a weight matrix obtained by training in the first layer;Representing the bias of the first layer.

It is input to the second feature transformation layer as shown in equation 3:

in the case of the formula 3 of the present invention,a weight matrix representing a second layer;Representing the bias of the second layer.

In the data set with the anomalies, the excessively dense aggregation connection or the near independent nodes have obvious characteristic expression in the graph structure, but in the attribute graphs with the numerous nodes, the proportion of the anomalies is small, and the structural characteristics of most normal nodes are easier to capture when the low-dimensional embedding is generated. Therefore, the weight parameter matrix can be dynamically optimized through the loss function in the model training process, so that the extracted characteristic is embedded to the maximum extent to contain the normal structural characteristic.

Step 120: and inputting the adjacency matrix and the current attribute matrix of the blockchain network into the graph attention layer so that the graph attention layer adopts a shared attention mechanism to determine importance weight values between every two nodes in the blockchain network, and outputting node embedded representations for representing node characteristics of the nodes.

In order to further increase the application effectiveness of the attention layer, the step 120 may further include: step 121: inputting the adjacency matrix and the current attribute matrix of the blockchain network into the graph meaning layer, so that the graph meaning layer firstly obtains node feature vectors mapped to potential space based on the attribute matrix; and determining importance weights between every two nodes in the block chain network based on the node feature vectors and the adjacency matrix by adopting a shared attention mechanism, and determining node embedded representations for representing node features of the nodes according to the importance weights between every two nodes.

Specifically, to obtain a representative high level node feature representation, the observed original node attributes X are first converted into a vector representation of a low dimensional potential space, and the output is subjected to nonlinear feature transformation by using an activation function to obtain node feature vectors mapped to the potential space by the attribute features X, and the embedded representation of the graph dataExpressed as equation 4:

where σ (·) is the activation function,and->Representing the weights and biases learned by the attention layer. And->X and ∈X, respectively>Is a dimension of (c).

Obtaining an embedded representation of graph dataThen, in combination with the adjacency matrix, the embedded representations of all neighboring nodes are aggregated by utilizing a shared attention mechanism for the nodes. The adjacency matrix provides neighbor information for the node. First, the degree of association between two nodes is defined as formula 5:

wherein e _i,j Is node v _i To v _j Is independent of the network structure of other nodes in the graph data. attn (·) represents the sum of the values for a εR ^D Andparameterized neural network,/->Is->Dimension of-> D represents->And->The dimension after concatenation, where the weights are shared by all nodes. The i represents a cascading operation of the node feature vectors;An embedded representation representing an i-th node; / >An embedded representation representing a j-th node; a represents the implementation of the attention mechanism for each node, a is a single layer feedforward neural network. Here, two feature vectors are spliced and remapped into a scalar;The representation is a weight matrix trained on the nodes.

In order to facilitate comparison and calculation of the correlation coefficients, the attention of the node to each neighboring node needs to be normalized. Importance weight gamma _i,j Normalization was performed by the Softmax function.

Node V _i Can be obtained by learning importance weights, and weighting and summing the embedded representations for all nodes. After the graph attention layer, the final node embedded z is obtained ^v As shown in equation 6.

In the case of the formula 6 of the present invention,represents the i-th node V _i Is embedded into the node of the node; k represents a kth node; n (N) _i Representing node V _i Is defined between all adjacent nodes of the network; gamma ray _i,k Representing importance weights of the ith node to the kth node;Represents the kth node V _k Is embedded in the node of (a).

Step 130: and inputting the attribute matrix into the attribute encoder so that the attribute encoder outputs an attribute embedded representation corresponding to the attribute characteristics of all attribute information in the blockchain network in an implicit layer.

The goal of the attribute encoder is to obtain an embedded representation of all attribute information in the attribute network in hidden space. In an attribute encoder, two non-linear feature transformation layers are used to map observed attribute data into a potential attribute embedded representation. The formula is as follows:

In the formula (7) and the formula (8),representing an attribute embedding matrix obtained by the first feature extraction; x is X ^T A transpose representing the input attribute matrix;And->Parameters representing the first feature transformation layer; z is Z ^A Representing an attribute embedding matrix obtained by the secondary feature extraction;And->Representing parameters of the second feature transformation layer.

The attribute matrix is all attribute information contained in each node, the transposed attribute matrix takes each dimension attribute as an analysis object, hidden space information of the attribute is extracted through nonlinear feature transformation, and the obtained embedded is used for attribute matrix reconstruction.

Step 140: and inputting the structure embedded representation and the node embedded representation into the structure decoder so that the structure decoder reconstructs the adjacent matrix and outputs a corresponding reconstructed adjacent matrix.

The goal of the structural decoder is to reconstruct the new adjacency matrix, and to achieve cross-modal information fusion, inner product operations are done using node embedding and structure embedding:

in the case of the formula 9 of the present invention,is a reconstructed adjacency matrix; z is Z ^s Comprises global structural features, Z ^v The method includes the steps of considering node characteristics of attribute similarity among nodes, and fusing the node characteristics and the node characteristics to enable each node to be endowed with proper adjacent relation, and restoring the same structure as the initial network.

In the above, the node is embedded with Z ^v ∈R ^N×D Structural embedding Z ^s ∈R ^N×D And finally, carrying out normalization operation on the matrix by using a Sigmoid activation function, so that the connection relation between the nodes is approximated to 0 and 1 as much as possible, and the reconstructed matrix and the original matrix are approximated infinitely.

The application calculates the inner product of the embedded representation vectors of two nodes as the probability of the link between the two nodes as shown in formula 10:

in the formula 10 of the present application,representing a reconstruction matrix;An embedded representation representing the node Vi;Representing an embedded representation of node Vj.

Step 150: and inputting the attribute embedded representation and the node embedded representation into the attribute decoder so that the attribute decoder reconstructs the attribute matrix and outputs a corresponding reconstructed attribute matrix.

The attribute encoder is used for reconstructing an attribute matrix, performing inner product operation on node embedding and attribute embedding, and generating attribute information conforming to node characteristics for each node through information fusion of hidden space.

Finally, the attribute decoder embeds the node into Z ^v And attribute embedding Z ^A As input for original node attribute decoding:

Z ^v ∈R ^N×D ,Z ^A ∈R ^N×D ,X∈R ^N×F . Wherein,,representing a reconstructed attribute matrix; r is R ^N×D Representing an N x D dimensional real matrix; r is R ^N×F Representing an N x F dimensional real matrix.

The attribute decoder enables the reconstructed attribute matrix to more accurately and completely reflect the basic semantics of the normal node characteristics in the original data space through the fusion of the two embedded information.

In order to further improve the effectiveness and reliability of the blockchain carbon emission transaction anomaly detection based on graph representation learning, in the blockchain carbon emission transaction anomaly detection method based on graph representation learning provided by the embodiment of the application, referring to fig. 3, before step 100 in the blockchain carbon emission transaction anomaly detection method based on graph representation learning, the method specifically comprises the following contents:

step 010: acquiring current graph data of a blockchain network for carbon emission transaction, wherein the graph data comprises all nodes participating in the carbon emission transaction in the blockchain network, relations among the nodes and attribute information of all the nodes;

step 020: generating a corresponding adjacency matrix and an attribute matrix representing attributes of each node in the blockchain network based on the graph data.

In order to further improve the reliability and effectiveness of the calculation of the anomaly detection score, in the blockchain carbon emission transaction anomaly detection method based on graph representation learning provided by the embodiment of the present application, referring to fig. 3, step 200 in the blockchain carbon emission transaction anomaly detection method based on graph representation learning further specifically includes the following contents:

Step 210: and constructing an objective function comprising super parameters and parameters respectively representing the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix by taking the minimum reconstruction error of the adjacency matrix and the attribute matrix representing each node attribute in the blockchain network as an objective.

Step 220: substituting the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix into the objective function and solving to obtain the reconstruction error value corresponding to each node in the blockchain network.

Step 230: and respectively determining the corresponding abnormal detection value of each node based on the reconstruction error value corresponding to each node.

In order to further improve the accuracy and reliability of detecting abnormal nodes with abnormal risk of carbon emission transaction, in the blockchain carbon emission transaction abnormal detection method based on graph representation learning provided by the embodiment of the application, referring to fig. 3, step 300 in the blockchain carbon emission transaction abnormal detection method based on graph representation learning further specifically includes the following contents:

step 310: the nodes are arranged in order of the abnormality detection score from small.

Step 320: and selecting a preset threshold number of nodes from the first node to be marked as abnormal nodes with abnormal risk of carbon emission transaction, or marking nodes larger than the abnormal threshold as abnormal nodes with abnormal risk of carbon emission transaction.

Step 330: and generating abnormal risk detection result data of the carbon emission transaction aiming at the blockchain network based on each abnormal node, and outputting the detection result data.

In terms of software, based on the foregoing embodiment of the blockchain carbon emission transaction anomaly detection method based on graph representation learning, the present application further provides an embodiment of a blockchain carbon emission transaction anomaly detection device based on graph representation learning, referring to fig. 4, for implementing the blockchain carbon emission transaction anomaly detection method based on graph representation learning, where the blockchain carbon emission transaction anomaly detection device based on graph representation learning specifically includes the following contents:

the model detection module 10 is configured to input a current adjacency matrix of a blockchain network for carbon emission trading and an attribute matrix representing attributes of each node in the blockchain network into a preset graph representation learning model based on a self-encoder and an attention mechanism, so that the graph representation learning model outputs a reconstructed adjacency matrix corresponding to the adjacency matrix and a reconstructed attribute matrix corresponding to the attribute matrix.

And the score calculation module 20 is used for determining the abnormality detection scores of all nodes in the blockchain network by applying the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix and the reconstructed attribute matrix.

An anomaly determination module 30 for determining an anomaly node in the blockchain network that is at risk of carbon emissions transaction anomalies based on the anomaly detection scores of the respective nodes.

The blockchain carbon emission transaction anomaly detection device based on graph representation learning provided by the application can be particularly used for executing the processing flow of the embodiment of the blockchain carbon emission transaction anomaly detection method based on graph representation learning in the embodiment, and the functions of the device are not repeated herein, and can be referred to for the detailed description of the embodiment of the blockchain carbon emission transaction anomaly detection method based on graph representation learning.

The part of the blockchain carbon emission transaction abnormality detection device based on graph representation learning for detecting blockchain carbon emission transaction abnormality based on graph representation learning can be executed in a server or a client device, and specifically can be selected according to the processing capability of the client device, the limitation of a user use scene and the like. The application is not limited in this regard. If all operations are completed in the client device, the client device may further include a processor for representing specific processing of the learned blockchain carbon emissions transaction anomaly detection based on the graph.

The client device may have a communication module (i.e. a communication unit) and may be in communication connection with a remote chat server and a blockchain server, so as to implement data transmission with the server. The server may include a server on the side of the task scheduling center, and in other implementations may include a server of an intermediate platform, such as a server of a third party server platform having a communication link with the task scheduling center server. The server may include a single computer device, a server cluster formed by a plurality of servers, or a server structure of a distributed device.

Any suitable network protocol may be used between the server and the client device, including those not yet developed on the filing date of the present application. The network protocols may include, for example, TCP/IP protocol, UDP/IP protocol, HTTP protocol, HTTPS protocol, etc. Of course, the network protocol may also include, for example, RPC protocol (Remote Procedure Call Protocol ), REST protocol (Representational State Transfer, representational state transfer protocol), etc. used above the above-described protocol.

As can be seen from the above description, the blockchain carbon emission transaction anomaly detection device based on graph representation learning provided by the embodiment of the application can capture cross-modal interaction in a network and extract deeper network features by adopting the adjacency matrix for representing the blockchain network structure, the attribute matrix for representing the node attribute and the graph representation learning model, so that the accuracy and the effectiveness of blockchain carbon emission transaction anomaly detection can be effectively improved; by introducing an attention mechanism, the influence of abnormal nodes can be effectively filtered when the graph representation learning model is extracted from the characteristics, the capability of obtaining normal characteristics of the nodes can be effectively improved, the influence of the abnormal nodes is weakened, the accuracy and reliability of abnormal detection of the blockchain carbon emission transaction can be further improved, and the safety and reliability of the carbon emission transaction in the blockchain can be further effectively ensured.

In order to further explain the scheme, the application also provides a graph representation learning model based on a self-encoder and an attention mechanism and a specific application example of a blockchain carbon emission transaction abnormality detection method based on graph representation learning, and the application designs a deep learning framework based on the self-encoder and the attention mechanism as a blockchain transaction abnormality detection method aiming at the problem that the existing abnormality detection method ignores interaction information between a network structure and node attributes and has limited abnormality detection capability. The framework is composed of a structure automatic encoder and an attribute automatic encoder to learn node and attribute feature vector representations together. In addition, attention mechanisms are introduced to learn the importance between a node and its neighbors. First the structure encoder converts the observed original node properties into a vector representation of the low-order potential space, then aggregates the embedded representations of all neighboring nodes using a shared-attention mechanism, ultimately generating node embeddings. The attribute encoder uses a multi-layer perceptron to map observed attribute data into a potential attribute embedded representation. The adjacency matrix is then reconstructed using a structure decoder, which reconstructs the attribute matrix, and the reconstruction errors of the nodes are measured from both structure and attribute angles as an objective function of neural network training. Anomaly detection is then implemented based on the reconstruction errors of the two angle measurement nodes for the structure and attributes.

The graph representation learning model based on the self-encoder and the attention mechanism provided by the application example of the application, referring to fig. 2, specifically comprises the following contents:

encoder (one)

(1) Structure encoder

it is input to the second feature transformation layer as shown in equation 3:

(2) Drawing attention layer

The goal of the graph attention layer is to extract the features of each node in the network data in hidden space. The network structure information and the node attribute information of each node can be directly read through the real data. Because the information carried by the nodes is different, the abstract representation of each feature can be extracted by combining the information of the two aspects. The accurate extraction of the node characteristics is the key for reconstructing the original network, and the node characteristics can reflect the local connection structure and the attributes of similar nodes because the information of the node is from two information sources, and each node has unique expression.

The graph attention layer extracts potential characteristics of each node and generates node embedding, wherein the node embedding is high-level characteristic summarization combining node self attributes, adjacent node attributes and node connection relations, and in the reconstruction process of the adjacency matrix and the attribute matrix, the adjacency relations and the attribute information are reconstructed according to the specific characteristics of each node.

The input of the attention layer is a node feature vector, and then the feature of each node is reduced in dimension again, so that a more abstract feature representation is obtained. When the correlation degree of two nodes is calculated, the node features are spliced to form a scalar, and the scalar is subjected to nonlinear feature transformation to obtain potential representation combining the two node features. After the center node and all the neighbor nodes execute the same operation, the Softmax layer is used for normalizing all the potential representations, so that the importance degree of the neighbor nodes is distributed among [0,1 ].

To obtain a representative high level node feature representation, the observed original node attribute X is first converted into a vector representation of a low dimensional potential space, and nonlinear feature transformation is performed using an activation function at the time of output to obtain a node feature vector mapped to the potential space by the attribute feature X, which is expressed as formula 4:

Where σ (·) is the activation function,and->Representing the weights and offsets learned by the encoder.

wherein e _i,j Is node v _i To v _j Is independent of the network structure of other nodes in the graph data. attn (·) represents the sum of the values for a εR ^D Anda parameterized neural network in which weights are shared by all nodes. The term "represents a concatenation operation of node feature vectors.

To facilitate the comparison and calculation of the correlation coefficients, a node and each neighbor are required to beThe attention of the node is normalized. Importance weight gamma _i,j Normalization was performed by the Softmax function.

(3) Attribute encoder

(II) decoder

(1) Structure decoder

Z ^s comprises global structural features, Z ^v Including consideration of inter-node attributesThe node characteristics of the similarity are fused, so that proper adjacency relationship can be given to each node, and the same structure as the initial network is restored.

(3) Attribute decoder

Z ^v ∈R ^N×D ,Z ^A ∈R ^N×D ,X∈R ^N×F . The attribute decoder enables the reconstructed attribute matrix to more accurately and completely reflect the basic semantics of the normal node characteristics in the original data space through the fusion of the two embedded information.

In summary, in the application example of the present application, the test result of the experimental system shows that the anomaly detection method designed and implemented according to the present application can effectively improve the capability of detecting anomalies. The attention introducing mechanism can filter the influence of the abnormal node during feature extraction, can effectively improve the quality of the normal feature of the obtained node and weaken the influence of the abnormal node. The design of the multi-path encoder captures cross-modal interaction in the network, extracts deeper network characteristics, and accordingly improves the capability of detecting abnormality of the model.

Block chain carbon emission transaction anomaly detection method based on graph representation learning

In order to use the deep learning framework provided by the application for detecting abnormal block chain transaction, a corresponding abnormal detection method is designed. The goal of the framework is to reconstruct a training network that is as similar as possible to the original network in both network structure and attribute information. Because of the oppositivity of normal and abnormal features, and because all nodes share a set of embeddings in reconstruction, reconstruction errors can be minimized when the embeddings contain as little normal features as possible.

The objective function is to minimize the reconstruction error of structures and attributes, expressed as:

wherein L is _rec Representing a reconstruction error; f and F' represent two-norms of matrix differences; alpha, theta, eta are super parameters, the parameters alpha control the weight of the structural reconstruction error and the attribute reconstruction error, and the parameters alpha control the Hadamard product, and the matrix difference two norms are used as the measurement of the network similarity, including the structure and the attribute information of each node; the definition of eta and theta is as follows:

in the formula, θ>1 and eta>1, as the matrix representing the real data set is sparse, the attribute or connection of the original matrix should be preserved during reconstruction; θ _i,j And eta _i,j Values representing subscripts of θ and η parameter matrices (i, j); a is that _i,j Values representing subscripts of the adjacency matrix (i, j); x is X _i,j Values representing indices of the attribute matrix (i, j).

The definition of outlier nodes refers to those nodes that deviate severely in structure and properties from other nodes. Through the above analysis, v of the node _i The anomaly score of (2) can be measured by the network structure and attribute reconstruction errors.

The method for embedding reconstruction can filter out nodes with a lower topological structure or rare node attributes, and the reconstructed network is generated by the feature induction of most normal nodes in the original data. Thus, if the structure information and attribute information of a certain node are reconstructed to still approximate the original state, it is a normal node with a high probability. Otherwise it may be quite different from most nodes, i.e. abnormal nodes.

In the abnormality detection stage, according to the reconstruction error of the matrix, the connection information and attribute information of each node are integrated to calculate the abnormality score. The nodes are arranged in descending order according to the anomaly scores, and the higher the anomaly score is, the greater the probability that the node has anomalies is. Marking the nodes with higher scores as nodes with abnormal prediction, comparing the nodes with the real labels, and judging the detection effect of the algorithm by using various evaluation indexes. The value of (2) is related to the distribution of anomaly scores.

The performance evaluation tables of the blockchain carbon emission transaction anomaly detection method based on graph representation learning, the conventional anomaly detection method Radar, anomalous and Dominant based on Cora, citeseer and Pubmed data sets are shown in table 1.

TABLE 1 AUC scores of anomaly detection methods in three data sets

Abnormality detection method	Cora	Citeseer	Pubmed
				Radar	72.08％	72.86％	71.68％
Anomalous	71.60％	72.67％	73.01％
				Dominant	76.81％	75.43％	77.49％
The application relates to a blockchain carbon emission transaction anomaly detection method based on graph representation learning	85.68％	87.03％	90.92％

As shown in Table 1, the recall rate of classical methods Radar, anomalous and Dominant is low, the detection capability of abnormal nodes is insufficient, and the recall rate of the algorithm of the application is obviously improved. This means that the probability of the abnormal node being identified is high for the original sample, which is reflected from the side, and the information of the abnormal node is obviously changed in the original space through feature extraction and reconstruction.

Comparison of the data in Table 1, with increasing AUC scores, demonstrates the effectiveness of the self-encoder and the effectiveness of the attention-directed mechanism in the proposed method.

In summary, the blockchain is used as a regularized digital technology, has the characteristics of multiparty consensus, transparent disclosure, tamper resistance and the like, and is naturally suitable for the scenes of strong rule dependence such as carbon transaction and carbon neutralization management. However, the user scale is dynamic and the participating identities are anonymous, which results in a more concealed, complex, intelligent financial crime. Thus, to implement intelligent blockchain transaction supervision, it is desirable to accurately identify abnormal transactions on the blockchain. The existing anomaly detection method ignores the interaction information between the network structure and the node attribute, and has limited anomaly detection capability. Based on this, the present application proposes a deep learning framework based on a self-encoder and a attention mechanism. From the comprehensive experimental data, the attention introducing mechanism can filter the influence of abnormal nodes during feature extraction, can effectively improve the quality of obtaining normal features of the nodes, and weakens the influence of the abnormal nodes. The design of the multi-path encoder captures cross-mode interaction in the network, and extracts deeper network characteristics, so that the capability of detecting abnormality of the model is improved.

The embodiment of the application also provides an electronic device (i.e. a computer device), which may include a processor, a memory, a receiver and a transmitter, where the processor is configured to execute the blockchain carbon emission transaction anomaly detection method based on graph representation learning mentioned in the above embodiment, and the processor and the memory may be connected by a bus or other manners, for example, through a bus connection. The receiver may be connected to the processor, memory, by wire or wirelessly.

The processor may be a central processing unit (Central Processing Unit, CPU). The processor may also be any other general purpose processor, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof.

The memory is used as a non-transitory computer readable storage medium and can be used for storing non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions/modules corresponding to the blockchain carbon emission transaction anomaly detection method based on graph representation learning in the embodiment of the application. The processor executes various functional applications and data processing of the processor by running non-transitory software programs, instructions and modules stored in the memory, i.e., implementing the blockchain carbon emission transaction anomaly detection method based on graph representation learning in the above method embodiments.

The memory may include a memory program area and a memory data area, wherein the memory program area may store an operating system, at least one application program required for a function; the storage data area may store data created by the processor, etc. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory may optionally include memory located remotely from the processor, the remote memory being connectable to the processor through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The one or more modules are stored in the memory that, when executed by the processor, perform the blockchain carbon emissions transaction anomaly detection method of the embodiment that is based on graph representation learning.

In some embodiments of the present application, a user equipment may include a processor, a memory, and a transceiver unit, which may include a receiver and a transmitter, the processor, the memory, the receiver, and the transmitter may be connected by a bus system, the memory being configured to store computer instructions, the processor being configured to execute the computer instructions stored in the memory to control the transceiver unit to transmit and receive signals.

As an implementation manner, the functions of the receiver and the transmitter in the present application may be considered to be implemented by a transceiver circuit or a dedicated chip for transceiver, and the processor may be considered to be implemented by a dedicated processing chip, a processing circuit or a general-purpose chip.

As another implementation manner, a manner of using a general-purpose computer may be considered to implement the server provided by the embodiment of the present application. I.e. program code for implementing the functions of the processor, the receiver and the transmitter are stored in the memory, and the general purpose processor implements the functions of the processor, the receiver and the transmitter by executing the code in the memory.

Embodiments of the present application also provide a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the foregoing blockchain carbon emissions transaction anomaly detection method based on graph representation learning. The computer readable storage medium may be a tangible storage medium such as Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disk, a removable memory disk, a CD-ROM, or any other form of storage medium known in the art.

Those of ordinary skill in the art will appreciate that the various illustrative components, systems, and methods described in connection with the embodiments disclosed herein can be implemented as hardware, software, or a combination of both. The particular implementation is hardware or software dependent on the specific application of the solution and the design constraints. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application. When implemented in hardware, it may be, for example, an electronic circuit, an Application Specific Integrated Circuit (ASIC), suitable firmware, a plug-in, a function card, or the like. When implemented in software, the elements of the application are the programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine readable medium or transmitted over transmission media or communication links by a data signal carried in a carrier wave.

It should be understood that the application is not limited to the particular arrangements and instrumentality described above and shown in the drawings. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the present application are not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after appreciating the spirit of the present application.

In this disclosure, features that are described and/or illustrated with respect to one embodiment may be used in the same way or in a similar way in one or more other embodiments and/or in combination with or instead of the features of the other embodiments.

The above description is only of the preferred embodiments of the present application and is not intended to limit the present application, and various modifications and variations can be made to the embodiments of the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. The blockchain carbon emission transaction anomaly detection method based on graph representation learning is characterized by comprising the following steps of:

2. The blockchain carbon emissions transaction anomaly detection method based on graph representation learning of claim 1, wherein the graph representation learning model based on the self-encoder and the attention mechanism comprises:

3. The blockchain carbon emission transaction anomaly detection method based on graph representation learning of claim 2, wherein the inputting the current adjacency matrix of a blockchain network for carbon emission transactions and an attribute matrix representing attributes of each node in the blockchain network into a preset graph representation learning model based on a self-encoder and an attention mechanism, so that the graph representation learning model outputs a reconstructed adjacency matrix corresponding to the adjacency matrix and a reconstructed attribute matrix corresponding to the attribute matrix, comprises:

Inputting the attribute matrix into the attribute encoder so that the attribute encoder outputs attribute embedded representation corresponding to the attribute characteristics of all attribute information in the blockchain network in an implicit layer;

4. The blockchain carbon emissions transaction anomaly detection method based on graph representation learning of claim 3, wherein the inputting the adjacency matrix and the current attribute matrix of the blockchain network into the graph attention layer such that the graph attention layer determines importance weights between every two nodes in the blockchain network using a shared attention mechanism to output a node embedded representation representing node characteristics of each of the nodes comprises:

5. The blockchain carbon emission transaction anomaly detection method based on graph representation learning of claim 1, wherein before the current adjacency matrix of a blockchain network for carbon emission transactions and an attribute matrix representing attributes of each node in the blockchain network are input into a preset graph representation learning model based on a self-encoder and an attention mechanism, so that the graph representation learning model outputs a reconstructed adjacency matrix corresponding to the adjacency matrix and a reconstructed attribute matrix corresponding to the attribute matrix, further comprising:

6. The blockchain carbon emissions transaction anomaly detection method based on graph representation learning of any of claims 1-5, wherein the applying the adjacency matrix, the reconstructed adjacency matrix, the attribute matrix, and the reconstructed attribute matrix to determine anomaly detection scores for each node in the blockchain network comprises:

7. The blockchain carbon emission transaction anomaly detection method based on graph representation learning of any one of claims 1 to 5, wherein the determining anomaly nodes in the blockchain network that are at risk of carbon emission transaction anomalies from the anomaly detection scores of the respective nodes comprises:

ranking the nodes in order of the anomaly detection score from small;

8. A blockchain carbon emission transaction anomaly detection device based on graph representation learning, comprising:

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the computer program, implements the blockchain carbon emissions transaction anomaly detection method based on graph representation learning of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, which when executed by a processor, implements the blockchain carbon emission transaction anomaly detection method based on graph representation learning as claimed in any one of claims 1 to 7.