Knowledge graph embedded power transaction subject operation evaluation method and system

Abstract

The invention discloses a knowledge graph embedded power transaction subject operation evaluation method and system, relates to the technical field of big data and artificial intelligence, and constructs a power transaction subject operation evaluation system from numerical features and text features. And constructing a knowledge graph based on the literature related to the power transaction subject, and predicting the operation condition of the power transaction subject through a deep learning operation evaluation model embedded in the knowledge graph. According to the model, rich text information in the knowledge graph is effectively combined, the operation condition of the transaction subject can be described more accurately, and support is provided for efficient decision making of an electric power company and the electric power transaction subject.

Description

Technical Field

[0001] The present invention relates to the field of big data and artificial intelligence technology, and in particular to a knowledge graph-embedded power trading entity operation evaluation method and system. Background Art

[0002] With the continuous deepening of the market-oriented reform of electricity, the operating status of the main trading entities in the electricity market has an important impact on market stability and efficiency. The main trading entities in the electricity market include power generation enterprises, power supply companies, power sales companies, and large users participating in market transactions. The multi-dimensional characteristics of these entities, such as operating efficiency, risk control ability, and market adaptability, jointly determine the healthy development of the electricity market. However, the existing technologies mostly use traditional statistical analysis methods, which makes it difficult to fully capture the multi-dimensional operating characteristics of trading entities, and the analysis results are highly subjective and lack real-time and dynamics. Therefore, the development of a new operating status portrait analysis method for power market trading entities is of great significance to improving market supervision efficiency and optimizing market resource allocation.

[0003] Multidimensional capability refers to the evaluation and analysis of the capabilities of power market trading entities from multiple angles and levels. This includes but is not limited to the operational efficiency, risk management, market competitiveness, and innovation capabilities of enterprises. By constructing a multidimensional capability model, the comprehensive performance and potential capabilities of trading entities can be comprehensively evaluated, providing more abundant information for market supervision and decision-making. Knowledge graph is a structured semantic knowledge base that can reveal the relationships and patterns between entities by organizing and representing complex information and knowledge points in the form of graphs. In the field of power market transactions, knowledge graphs can integrate data from different sources and types, build a complex network of connections between trading entities, market rules, policies and regulations, and provide support for in-depth analysis of the operating status of trading entities. Multilayer perceptron (MLP) is a deep learning algorithm composed of multiple neuron layers that can handle complex nonlinear relationships. MLP effectively extracts features and recognizes patterns from data by learning the mapping relationship between input data and output results. In the portrait analysis of the operating status of power market trading entities, MLP can be used to build prediction models to identify and predict the operating trends and risk points of trading entities. Summary of the invention

[0004] In view of the above existing problems, the method proposed in this invention innovatively combines multi-dimensional capability evaluation, knowledge graph and MLP technology to construct a new method for profiling the operation status of power market trading entities. The information of knowledge graph is integrated into the process of power trading entity operation evaluation, and information with predictive effect on trading entity operation rating is mined from unstructured data to improve the accuracy of trading entity operation rating.

[0005] In order to solve the above technical problems, a knowledge graph embedded power trading entity operation evaluation method is proposed, including:

[0006] Construct a numerical feature system of the operating status of the transaction subject from multiple dimensions of transaction history and credit record; construct a knowledge graph tag library of the multi-dimensional capabilities of the operating status of the transaction subject; construct an interactive knowledge graph of the transaction subject-capability dimension based on the transaction subject feature system and expert experience; and construct a rating prediction model for the operating status of the transaction subject based on KG-MLP.

[0007] As a preferred solution of the knowledge graph embedded power trading entity operation evaluation method described in the present invention, the numerical feature system includes constructing a numerical feature data set of power market trading entities, and the features cover structured numerical features of historical trading behaviors, entity credit records, and historical electricity consumption.

[0008] The data sources are market trading platforms, credit rating systems, and historical transaction records.

[0009] The minimum-maximum normalization method is used to convert each feature data into a value in the interval [0,1] to obtain the numerical feature matrix of the transaction subject.

[0010] As a preferred solution of the knowledge graph embedded power trading entity operation evaluation method described in the present invention, the knowledge graph label library includes a label library that constructs a trading entity operation status portrait through five capability dimensions, including market characteristics, operation characteristics, financial characteristics, compliance and risk control characteristics, and innovation and technology application capability characteristics.

[0011] The interactive relationship of the unstructured characteristics of transaction subjects is characterized by subject relationship labels in five capability dimensions.

[0012] As a preferred solution of the knowledge graph embedded power trading entity operation evaluation method described in the present invention, the trading entity-capability dimension interactive knowledge graph includes, after obtaining the label library, screening out the documents with correlation with power trading or power trading entities lower than 0.2, performing special symbol filtering and text segmentation on the collected text.

[0013] Construct a training data set, use text data combined with active learning and RoBERTa-Bi-Attention methods to train the entity and relationship extraction model of power trading entities, annotate the high-quality samples extracted by active learning, perform word embedding processing on the input text fragments, and extract semantic features through deep neural networks to perform entity relationship extraction tasks. The trained model will extract entity relationships for all selected trading entity documents.

[0014] As a preferred solution of the knowledge graph embedded power trading subject operation evaluation method described in the present invention, the rating prediction model construction includes setting the operating status rating set of the trading subject, giving the numerical feature set of the trading subject, recording the triple table in the knowledge graph obtained by query, constructing a cross compression unit, and constructing the following interaction matrix:

[0015]

[0016] Where M is the numerical feature set of a given transaction subject, H represents the embedding of the knowledge graph head entity set corresponding to the transaction subject, and M l ∈R d , H l ∈R k is the lth hidden layer vector corresponding to M and H.

[0017] Based on the interaction matrix, the variable vector of the transaction subject and the variable vector of the associated parameters at the l+1 layer are:

[0018]

[0019] Among them, w ·· , b · is a trainable parameter matrix.

[0020] The given set of triples (h, r, t) in the knowledge graph, where h and t are the head entity and the tail entity respectively, and r is the corresponding relationship, is used to embed the obtained triples using the pre-trained language model, and the corresponding feature vector is represented as (H, R, T). The head entity and the relationship representation are respectively input into the model and the prediction of the tail entity is finally output through a K-layer multi-layer perceptron.

[0021]

[0022] Among them, H L Represents the header entity representation after L-layer cross-compression units, R L represents the relationship after passing through the same L number of multi-layer perceptrons, f L (.) represents the cross compression unit of L layer, is the prediction vector represented by the tail entity feature.

[0023] The similarity function is used to calculate the similarity score between the predicted value and the actual value.

[0024] As a preferred solution of the method for evaluating the operation of power trading entities embedded in a knowledge graph according to the present invention, the rating prediction model construction further includes extracting dense features from the feature vector M of the trading entity through a cross compression unit of the L layer while embedding the knowledge graph:

[0025] M L =f L (M, H)[M]

[0026] Combining the interaction between the original features of the transaction subject and the features embedded in the knowledge graph, the potential features of the transaction subject are obtained, and the multi-classification prediction function f R (·) Output the final operation rating prediction of the transaction entity, the operation status rating g i The prediction is:

[0027]

[0028] Using the learning algorithm of multi-objective mapping, adding a trainable parameter regularization term after the loss to solve the overfitting problem is:

[0029]

[0030] Among them, γ(.) in the first term of the loss function is the cross entropy loss function, and λ1 and λ2 are adjustable hyperparameters.

[0031] Based on the constructed model and loss function, the model is forward trained and back-propagated until the model training converges to obtain the final prediction model of the transaction entity's operational rating.

[0032] Another object of the present invention is to provide a knowledge graph embedded power trading entity operation evaluation system. This system involves a method for constructing a portrait of the operation status of power trading entities based on multi-dimensional capabilities and knowledge graph-multi-layer perceptron; it aims to conduct a comprehensive and in-depth analysis and evaluation of the operation status of power market trading entities through big data and artificial intelligence technology.

[0033] As a preferred solution of the knowledge graph embedded power trading entity operation evaluation system described in the present invention, it is characterized by including a data preprocessing module, a knowledge graph construction module, and an operation status rating prediction model module.

[0034] The data preprocessing module is responsible for collecting and processing raw data and constructing a numerical feature data set of the transaction subject.

[0035] The knowledge graph construction module is responsible for constructing the knowledge graph of the transaction subject and describing the interactive relationship of unstructured features.

[0036] The operation status rating prediction model module is responsible for predicting the operation status rating of the transaction subject based on numerical features and knowledge graph embedding.

[0037] The numerical feature matrix generated by the data preprocessing module is input into the operation status rating prediction model module, and the knowledge graph embedding representation generated by the knowledge graph construction module is input into the operation status rating prediction model module. The operation status rating prediction model module outputs the operation rating prediction of the transaction subject based on the numerical feature matrix and the knowledge graph embedding representation.

[0038] As a preferred solution of the knowledge graph embedded power trading entity operation evaluation system described in the present invention, the data preprocessing module collects information on the data source of the market trading platform, credit evaluation system, and historical transaction records, constructs original features, and uses the minimum-maximum normalization method to normalize the feature data to form a numerical feature matrix of the trading entity.

[0039] The knowledge graph construction module includes a tag library construction unit, a document retrieval and preprocessing unit, an entity relationship extraction unit, and a knowledge graph storage unit.

[0040] The tag library construction unit constructs the tag library of the knowledge graph based on the literature and domain knowledge. The document retrieval and preprocessing unit retrieves the relevant literature and performs preprocessing to provide text data for entity relationship extraction. The entity relationship extraction unit uses the preprocessed text data to extract entities and relationships and generate triples. The knowledge graph storage unit receives the triple data from the entity relationship extraction unit and stores it in the Neo4j database.

[0041] The operation status rating prediction model module includes a cross compression unit, a status analysis module, a knowledge graph embedding unit, and a feature extraction and rating prediction unit.

[0042] The cross-compression unit receives the numerical feature matrix of the data preprocessing module and the knowledge graph embedding representation of the knowledge graph construction module, performs cross-compression and is analyzed by the state analysis module. The knowledge graph embedding unit uses the output of the feature interaction unit to embed the knowledge graph through a multi-layer perceptron model. The feature extraction and rating prediction unit combines the numerical features and the knowledge graph embedding results to extract potential features and perform rating predictions.

[0043] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method for evaluating the operation of an electric power trading entity embedded with a knowledge graph are implemented.

[0044] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the method for evaluating the operation of an electric power trading entity embedded with a knowledge graph are implemented.

[0045] Beneficial effects of the present invention: The present invention adopts the method of predicting the tail entity by head entity and relationship, effectively extracts the correlation information between the transaction subject and the capability dimension in the knowledge graph, thereby realizing the embedding of the knowledge graph. The cross compression unit is used to integrate the information of the knowledge graph into the operation evaluation task of the power transaction subject, effectively combining the multi-dimensional information sources of the transaction subject, and improving the information richness and prediction accuracy of the transaction subject operation rating task. The knowledge graph is used to perform capability dimension diagnosis and analysis to provide guidance for the operation of the transaction subject. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0047] Figure 1 An overall flow chart of a knowledge graph-embedded power trading entity operation evaluation method provided for an embodiment of the present invention.

[0048] Figure 2 A flowchart of constructing a portrait of the operating status of a power trading entity based on multi-dimensional capabilities and a knowledge graph-multi-layer perceptron, which is provided as a knowledge graph-embedded power trading entity operation evaluation method according to an embodiment of the present invention.

[0049] Figure 3 A system solution module diagram of a knowledge graph-embedded power trading entity operation evaluation system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is mutually exclusive with other embodiments, either individually or selectively.

[0053] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0054] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0055] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] Example 1, reference Figure 1-Figure 2 , which is the first embodiment of the present invention, and provides a knowledge graph embedded power trading subject operation evaluation method, comprising:

[0057] Based on the information of Guangxi power trading entities:

[0058] S1: Construct a numerical feature system of the operating status of the transaction subject from multiple dimensions of transaction history and credit record.

[0059] Furthermore, a numerical feature data set of power market trading entities is established, covering structured numerical original features such as historical trading behavior, entity credit records, historical electricity consumption, etc. Data sources include market trading platforms, credit evaluation systems, historical trading records, etc. Derivative features describing the operating conditions of trading entities are constructed based on the original features.

[0060] Table 1

[0061]

[0062] It should be noted that the minimum-maximum normalization method is used to convert each feature data into a value in the [0,1] interval, that is,

[0063]

[0064] After the above steps, the numerical feature matrix of the transaction subject is finally obtained:

[0065]

[0066] S2: Construct a knowledge graph tag library of multi-dimensional capabilities of the operating status of transaction entities.

[0067] Furthermore, based on the existing literature on the operation evaluation of power market trading entities, a relational subject label library of the knowledge graph is constructed from the dimensions of subject attributes, market characteristics, innovation and technology application capability characteristics, etc. The interactive relationship of the unstructured characteristics of the trading entities is characterized by the subject relationship labels of the above dimensions.

[0068] Table 2

[0069]

[0070] S3: An interactive knowledge graph of transaction subject-capability dimensions based on the transaction subject feature system and expert experience.

[0071] Furthermore, after obtaining the label library in S2, S3 combines the label library and expert experience to build an interactive knowledge graph of transaction subjects and capability dimensions, and establishes the association between transaction subjects and capability dimensions with the help of the knowledge graph:

[0072] Retrieve the literature related to power trading knowledge and power trading subject information, and screen out the literature with a given power trading or power trading subject keyword appearance ratio less than 0.2. Filter the collected literature with special symbols, delete irrelevant information text such as URLs, pictures, and timestamps, and use the text cutting library in Huggingface for text segmentation.

[0073] A labeled entity relationship training dataset is constructed for text data, and the RoBERTa-Bi-Attention method is used based on the labeled data to train the entity and relationship joint extraction model of the power trading subject. Define the entity label set Y e With the relationship label set Y r , then the label space of the model is Y = Y r ∪Y e ∪{⊥}. Define entity label set Y e With the relationship label set Y r , then the label space of the model is Y = Y r∪Y e ∪{⊥}. For a text segment with an input length of |s|, the text is processed by word embedding and semantic features are extracted through a deep neural network. The final model will output a vector of |s|×|s|×|y| to represent the entity relationship label prediction of each position in the matrix, thereby achieving the task of entity relationship extraction. Finally, the trained model will extract entity relationships from all selected transaction subject documents to form "entity-relationship-entity" triples.

[0074] Build a knowledge graph. Store the identified entity-relationship-entity triples in the Neo4j database for subsequent queries on the transaction subject capability dimension.

[0075] S4: Construction of a rating prediction model based on the operating status of trading entities of KG-MLP.

[0076] Furthermore, after obtaining the knowledge graph of the transaction subject's capability dimension, S4 constructed a rating prediction model for the transaction subject's operating status. This method consists of three modules: a cross-compression unit, a status analysis module, and a knowledge graph embedding module.

[0077] First, set the transaction subject's operating status rating set G = {g1, g2, ...g n}, where n is the level of operation status evaluation, given the numerical feature set of the transaction subject M = {m1, m2, ..., m s}, s is the number of transaction subject characteristics.

[0078] At the same time, the triples in the knowledge graph obtained by the record query are represented as (h, r, t), where h and t are the head entity and the tail entity respectively, and r is the corresponding relationship.

[0079] It should be noted that the L-layer cross compression unit is constructed. In the l-th layer of the feature extraction module and the knowledge graph embedding module, the numerical feature set M of the given transaction subject and the embedding representation H of the knowledge graph head entity set corresponding to the transaction subject, as well as the l-th hidden layer vector representation Ml∈R corresponding to M and H d , H l ∈R k , construct the following interaction matrix:

[0080]

[0081] Based on the interaction matrix, the variable vector of the transaction subject and the variable vector of the associated parameters at the l+1 layer are:

[0082]

[0083] Among them, w MM , wHM , b M is the transformation parameter matrix of the hidden representation of the numerical feature, w MH , w HH , b H is the transformation parameter matrix of the entity relationship feature hidden representation. In one embodiment, L is set to 2, and L in the subsequent steps is set to this value.

[0084] Knowledge graph embedding module: For a given set of triples (h, r, t) in the knowledge graph, we first use the pre-trained language model to embed the obtained triples into words, and obtain the corresponding feature vector represented as (H, R, T).

[0085] After the head entity and relation representation are input into L-layer cross compression unit and L-layer multilayer perceptron model respectively, the output results are concatenated and finally the prediction of the tail entity is output through K-layer multilayer perceptron.

[0086]

[0087] Among them, H L Represents the header entity representation after L-layer cross-compression units, R L represents the relationship after passing through the same L number of multi-layer perceptrons, f L (·) represents the cross compression unit of L layer, MLP L With MLP K Represent the multi-layer perceptron model of L layer and K layer respectively, is the prediction vector represented by the tail entity feature. Finally, the similarity function is used to calculate the similarity score between the predicted value and the actual value.

[0088]

[0089] Here, the normalized vector inner product is used as the measure of similarity score.

[0090] The specific construction of the multi-layer perceptron is as follows:

[0091] First, an L-layer multilayer perceptron is constructed, with 2 linear layers, the input and output dimensions are both 64, and a Dropout layer is added between layers, with the Dropout rate set to 0.5; secondly, a K-layer multilayer perceptron is constructed, with 1 linear layer, the input dimension is 64, and the output dimension is the word embedding dimension.

[0092] Feature extraction and rating prediction of the transaction subject's operating status. While embedding the knowledge graph, the feature vector M of the transaction subject is also extracted through the cross-compression unit of the L layer to extract its dense features:

[0093] M L =f L(M,H)[M]

[0094] Among them, f L The meaning of (.) is the same as above. After combining the interaction between the original features of the transaction subject and the features embedded in the knowledge graph, the potential features of the transaction subject are finally obtained, and then the multi-classification prediction function f R (.) Output the final operational rating forecast of the transaction subject. Operational status rating g i Prediction It can be expressed as:

[0095]

[0096] The multi-classification prediction model here still uses K-layer multilayer perceptron for discrimination and classification, constructs 1 linear layer, the input dimension is 64, and the output dimension is the number of levels of the operating status rating.

[0097] The model uses a multi-objective mapping learning algorithm, and the loss function is composed of the following two parts: one is the operation status prediction loss, and the other is the loss of the prediction knowledge graph embedding module. At the same time, a trainable parameter regularization term is added after the loss to solve the overfitting problem. The final loss function can be expressed as:

[0098]

[0099] Among them, γ(.) in the first term of the loss function is the cross entropy loss function, the second term is the entity prediction accuracy penalty, the third term is the trainable parameter regularization term, and λ1 and λ2 are adjustable hyperparameters.

[0100] Based on the above-mentioned model and loss function, forward training and back propagation of the model are carried out until the model training converges to obtain the final prediction model of the transaction entity's operation rating.

[0101] Example 2, reference Figure 1-Figure 2 , which is an embodiment of the present invention, provides a knowledge graph embedded power trading entity operation evaluation method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0102] Build a knowledge graph. Store the identified entity-relationship-entity triples in the Neo4j database for subsequent queries on the transaction subject capability dimension.

[0103] After obtaining the knowledge graph of the transaction subject's capability dimension, a rating prediction model for the transaction subject's operating status is constructed. This method includes three modules: cross compression unit, status analysis module, and knowledge graph embedding module.

[0104] First, set the transaction subject's operating status rating set G = {g1, g2, ...g n}, where n is the level of operation status evaluation, given the numerical feature set of the transaction subject M = {m1, m2, ..., m s}, s is the number of transaction subject features. At the same time, the triples in the knowledge graph obtained by record query are represented as (h, r, t). Among them, h, t are the head entity and the tail entity respectively, and r is the corresponding relationship.

[0105] Construct L-layer cross compression units. In the l-th layer of the feature extraction module and the knowledge graph embedding module, given the numerical feature set M of the transaction subject and the embedding representation H of the knowledge graph head entity set corresponding to the transaction subject, as well as the l-th layer hidden layer vector representation M corresponding to M and H l ∈R d , H l ∈R k , construct the following interaction matrix:

[0106]

[0107] Based on the interaction matrix, the variable vector of the transaction subject and the variable vector of the associated parameters at the l+1 layer are:

[0108]

[0109] Among them, w MM , w HM , b M is the transformation parameter matrix of the hidden representation of the numerical feature, w MH , w HH , b H Transformation parameter matrix for hidden representation of entity relationship features.

[0110] In this embodiment, L is set to 2, and L in the subsequent descriptions is set to this value.

[0111] Knowledge graph embedding module: For a given set of triples (h, r, t) in the knowledge graph, we first use the pre-trained language model to embed the obtained triples into words, and obtain the corresponding feature vector represented as (H, R, T).

[0112] In this embodiment, the dimension of word embedding is 32.

[0113] Then, the head entity and the relationship representation are input into the L-layer cross compression unit and the L-layer multilayer perceptron model respectively, the output results are concatenated, and the prediction of the tail entity is finally output through the K-layer multilayer perceptron.

[0114]

[0115] Among them, H L Represents the header entity representation after L-layer cross-compression units, R Lrepresents the relationship after passing through the same L number of multi-layer perceptrons, f L (.) represents the cross compression unit of L layer, MLP L With MLP K Represent the multi-layer perceptron model of L layer and K layer respectively, is the prediction vector represented by the tail entity feature.

[0116] In this embodiment, K is set to 1.

[0117] Finally, the similarity function is used to calculate the similarity score between the predicted value and the actual value.

[0118]

[0119] Here, the normalized vector inner product is used as the measure of similarity score.

[0120] In the embodiment, the specific construction of the multi-layer perceptron is as follows:

[0121] First, an L-layer multilayer perceptron is constructed, with 2 linear layers, the input and output dimensions are both 64, and a Dropout layer is added between layers, with the Dropout rate set to 0.5; secondly, a K-layer multilayer perceptron is constructed, with 1 linear layer, the input dimension is 64, and the output dimension is the word embedding dimension.

[0122] S4.4 Feature extraction and rating prediction of the transaction subject's operating status. While embedding the knowledge graph, the feature vector M of the transaction subject is also extracted through the cross-compression unit of the L layer to extract its dense features:

[0123] M L =f L (M,H)[M]

[0124] Among them, f L The meaning of (.) is the same as above. After combining the interaction between the original features of the transaction subject and the features embedded in the knowledge graph, the potential features of the transaction subject are finally obtained, and then the multi-classification prediction function f R (.) Output the final operational rating forecast of the transaction subject. Operational status rating g i Prediction It can be expressed as:

[0125]

[0126] The multi-classification prediction model here still uses K-layer multilayer perceptron for discrimination and classification, constructs 1 linear layer, the input dimension is 64, and the output dimension is the number of levels of the operating status rating.

[0127] The model uses a multi-objective mapping learning algorithm, and the loss function is composed of the following two parts: one is the operation status prediction loss, and the other is the loss of the prediction knowledge graph embedding module. At the same time, a trainable parameter regularization term is added after the loss to solve the overfitting problem. The final loss function can be expressed as:

[0128]

[0129] Among them, γ(.) in the first term of the loss function is the cross entropy loss function, the second term is the entity prediction accuracy penalty, the third term is the trainable parameter regularization term, and λ1 and λ2 are adjustable hyperparameters.

[0130] In this embodiment, the hyperparameters are set in the range of [0.001, 0.01, 0.05, 0.1], and the optimal parameters are searched according to cross-validation and are 0.05.

[0131] Based on the above-mentioned model and loss function, the model is forward trained and back-propagated until the model training converges, and the prediction model of the final transaction entity operation rating is obtained. The performance of the final model in the validation set and test set is as follows:

[0132] Table 3

[0133] Evaluation indicators Validation set Test Set Accuracy (%) 94.87 93.53

[0134] Embodiment 3, the third embodiment of the present invention, is different from the first two embodiments in that:

[0135] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0136] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0137] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0138] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0139] Example 4, reference Figure 3 , which is the fourth embodiment of the present invention, provides a knowledge graph embedded power trading entity operation evaluation system, including a data preprocessing module 10, a knowledge graph construction module 20, and an operation status rating prediction model module 30.

[0140] The data preprocessing module 10 collects information from the market trading platform, credit rating system, and data sources of historical transaction records, constructs original features, and uses the minimum-maximum normalization method to normalize the feature data to form a numerical feature matrix of the transaction subject and input it into the operation status rating prediction model module.

[0141] The knowledge graph construction module 20 includes a label library construction unit 21, a document retrieval and preprocessing unit 22, an entity relationship extraction unit 23, and a knowledge graph storage unit 24; the label library construction unit 21 constructs a label library of the knowledge graph based on documents and domain knowledge, the document retrieval and preprocessing unit 22 retrieves relevant documents and performs preprocessing to provide text data for entity relationship extraction, the entity relationship extraction unit 23 uses the preprocessed text data to extract entities and relationships and generate triples, the knowledge graph storage unit 24 receives the triple data of the entity relationship extraction unit 23 and stores it in the Neo4j database, and inputs it into the operation status rating prediction model module.

[0142] The operation status rating prediction model module 30 includes a cross-compression unit 31, a status analysis module 32, a knowledge graph embedding unit 33, and a feature extraction and rating prediction unit 34; the cross-compression unit 31 receives the numerical feature matrix of the data preprocessing module and the knowledge graph embedding representation of the knowledge graph construction module, performs cross-compression and analyzes it through the state analysis module 32, the knowledge graph embedding unit 33 uses the output of the feature interaction unit to embed the knowledge graph through a multi-layer perceptron model, and the feature extraction and rating prediction unit 34 combines the numerical features and the knowledge graph embedding results to extract potential features and perform rating predictions.

[0143] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A knowledge graph embedded power trading entity operation evaluation method, characterized by: This includes building a numerical feature system of the operating status of transaction entities from multiple dimensions of transaction history and credit records; Build a knowledge graph tag library of multi-dimensional capabilities of the operating status of transaction entities; An interactive knowledge graph of transaction subject-capability dimensions based on the transaction subject feature system and expert experience; Construction of rating prediction model based on the operating status of KG-MLP trading entities.

2. The method for evaluating the operation of power trading entities embedded with knowledge graph according to claim 1, characterized in that: The numerical feature system includes constructing a numerical feature data set of power market transaction entities, the features of which include structured numerical features of historical transaction behaviors, entity credit records, and historical electricity consumption; The data sources are market trading platforms, credit rating systems, and historical transaction records; The minimum-maximum normalization method is used to convert each feature data into a value in the interval [0,1] to obtain the numerical feature matrix of the transaction subject.

3. The method for evaluating the operation of power trading entities embedded with knowledge graph according to claim 2, characterized in that: The knowledge graph tag library includes a tag library that constructs a profile of the operating status of a transaction subject through five capability dimensions, including market characteristics, operational characteristics, financial characteristics, compliance and risk control characteristics, and innovation and technology application capability characteristics; The interactive relationship of the unstructured characteristics of transaction subjects is characterized by subject relationship labels in five capability dimensions.

4. The method for evaluating the operation of power trading entities embedded with knowledge graph according to claim 3, characterized in that: The transaction subject-capability dimension interactive knowledge graph includes, after obtaining the tag library, screening out the literature with a correlation with power transactions or power transaction subjects lower than 0.2, performing a special symbol filtering text cleaning operation on the collected text and performing text segmentation; Construct a training data set, use text data combined with active learning and RoBERTa-Bi-Attention methods to train the entity and relationship extraction model of power trading entities, annotate the high-quality samples extracted by active learning, perform word embedding processing on the input text fragments, and extract semantic features through deep neural networks to perform entity relationship extraction tasks. The trained model will extract entity relationships for all selected trading entity documents.

5. The method for evaluating the operation of power trading entities embedded with knowledge graph according to claim 4, characterized in that: The rating prediction model construction includes setting the operational status rating set of the transaction subject, giving the numerical feature set of the transaction subject, recording the triple table in the knowledge graph obtained by query, constructing a cross compression unit, and constructing the following interaction matrix: Among them, M is the numerical feature set of a given transaction subject, H represents the embedding of the knowledge graph head entity set corresponding to the transaction subject, and M l ∈R d , H l ∈R k is the lth hidden layer vector corresponding to M and H; Based on the interaction matrix, the variable vector of the transaction subject and the variable vector of the associated parameters at the l+1 layer are: Among them, w ·· , b · is a trainable parameter matrix; Through the given triples (h, r, t) in the knowledge graph, where h and t are the head entity and the tail entity respectively, and r is the corresponding relationship; the pre-trained language model is used to embed the obtained triples, and the corresponding feature vector is represented as (H, R, T). The head entity and the relationship representation are respectively input into the model and the prediction of the tail entity is finally output through a K-layer multi-layer perceptron; Among them, H L Represents the header entity representation after L-layer cross-compression units, R L represents the relationship after passing through the same L number of multi-layer perceptrons, f L (.) represents the cross compression unit of L layer, is the prediction vector represented by the tail entity feature; The similarity function is used to calculate the similarity score between the predicted value and the actual value.

6. A knowledge graph embedded power trading entity operation evaluation method as claimed in claim 5, characterized in that: The rating prediction model construction also includes extracting dense features from the feature vector M of the transaction subject through the cross compression unit of the L layer while embedding the knowledge graph: Bao = f L (M, H)[M] Combining the interaction between the original features of the transaction subject and the features embedded in the knowledge graph, the potential features of the transaction subject are obtained, and the multi-classification prediction function f R (.) Output the final operational rating forecast of the transaction entity, the operational status rating g i The prediction is: Using the learning algorithm of multi-objective mapping, adding a trainable parameter regularization term after the loss to solve the overfitting problem is: Among them, γ(.) in the first term of the loss function is the cross entropy loss function, and λ1 and λ2 are adjustable hyperparameters; Based on the constructed model and loss function, the model is forward trained and back-propagated until the model training converges to obtain the final prediction model of the transaction entity's operational rating.

7. A system using a method for evaluating the operation of a power trading entity embedded in a knowledge graph as claimed in any one of claims 1 to 6, characterized in that: It includes data preprocessing module, knowledge graph construction module, and operation status rating prediction model module; The data preprocessing module is responsible for collecting and processing raw data and constructing a numerical feature data set of the transaction subject; The knowledge graph construction module is responsible for constructing the knowledge graph of the transaction subject and describing the interactive relationship of the unstructured features; The operation status rating prediction model module is responsible for predicting the operation status rating of the transaction subject based on numerical features and knowledge graph embedding; The numerical feature matrix generated by the data preprocessing module is input into the operation status rating prediction model module, and the knowledge graph embedding representation generated by the knowledge graph construction module is input into the operation status rating prediction model module. The operation status rating prediction model module outputs the operation rating prediction of the transaction subject based on the numerical feature matrix and the knowledge graph embedding representation.

8. The method for evaluating the operation of power trading entities embedded with knowledge graph according to claim 7, characterized in that: The data preprocessing module collects information from the market trading platform, credit evaluation system, and data sources of historical transaction records, constructs original features, and uses the minimum-maximum normalization method to normalize the feature data to form a numerical feature matrix of the transaction subject; The knowledge graph construction module includes a tag library construction unit, a document retrieval and preprocessing unit, an entity relationship extraction unit, and a knowledge graph storage unit; The tag library construction unit constructs the tag library of the knowledge graph based on the literature and domain knowledge. The literature retrieval and preprocessing unit retrieves the relevant literature and performs preprocessing to provide text data for entity relationship extraction. The entity relationship extraction unit uses the preprocessed text data to extract entities and relationships and generate triples. The knowledge graph storage unit receives the triple data of the entity relationship extraction unit and stores it in the Neo4j database. The operation status rating prediction model module includes a cross compression unit, a status analysis module, a knowledge graph embedding unit, and a feature extraction and rating prediction unit; The cross-compression unit receives the numerical feature matrix of the data preprocessing module and the knowledge graph embedding representation of the knowledge graph construction module, performs cross-compression and is analyzed by the state analysis module. The knowledge graph embedding unit uses the output of the feature interaction unit to embed the knowledge graph through a multi-layer perceptron model. The feature extraction and rating prediction unit combines the numerical features and the knowledge graph embedding results to extract potential features and perform rating predictions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a knowledge graph-embedded power trading entity operation evaluation method described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a knowledge graph-embedded power trading entity operation evaluation method according to any one of claims 1 to 6 are implemented.

Images