Methods and systems for identifying collusion among power generation groups in the electricity market

Abstract

This invention discloses a method for identifying collusion among power generation groups in the electricity market. The method includes acquiring and preprocessing historical data from the electricity market to construct a training dataset; building an initial model for extracting individual features from power generation groups based on an undirected weighted graph construction scheme, a logical gating mechanism, and a graph neural network model, and training this model to obtain an individual feature extraction model; extracting features of individual power generation groups in the target electricity market using the obtained model; and identifying collusion among power generation groups in the target electricity market based on a clustering algorithm. This invention also discloses a system for implementing the aforementioned method for identifying collusion among power generation groups in the electricity market. This invention not only achieves collusion identification among power generation groups in the electricity market but also offers higher reliability and accuracy.

Description

Technical Field

[0001] This invention belongs to the field of electrical automation, specifically relating to a method and system for identifying collusion among groups on the generation side of the power market. Background Technology

[0002] With economic and technological development and the improvement of people's living standards, electricity has become an indispensable secondary energy source in people's production and daily life, bringing endless convenience. Therefore, ensuring a stable and reliable supply of electricity has become one of the most important tasks of the power system.

[0003] Currently, the electricity market within the power system has developed significantly, playing an increasingly important role, and its mechanisms have become more complex. Ensuring the safe and stable operation of the electricity market has also become one of the key tasks of the power system.

[0004] Currently, covert group collusion has emerged in the electricity market, posing a significant threat to its safe and stable operation. Researchers have proposed corresponding identification schemes for group collusion on the generation side of the electricity market. For example, patent application CN 120217214A proposes a behavior analysis-based identification method that uses an isolation forest to screen for abnormal behavior and calculates the collusion probability using logistic regression; patent application CN 113344589A proposes an intelligent identification method based on the VAE-GMM model, using unsupervised clustering algorithms to separate collusion samples. However, these schemes, in the identification process, ignore the topological relationships between entities, making it difficult to capture hidden collaborative features, and are prone to misjudging high-price behavior under normal competitive conditions as collusion. Therefore, the reliability and accuracy of existing schemes are relatively low. Summary of the Invention

[0005] One of the objectives of this invention is to provide a highly reliable and accurate method for identifying collusion among power generation groups in the electricity market.

[0006] The second objective of this invention is to provide a system for implementing the method for identifying collusion among power generation groups in the electricity market.

[0007] The method for identifying collusion among power generation groups in the electricity market provided by this invention includes the following steps:

[0008] S1. Obtain historical data and information about the electricity market;

[0009] S2. Preprocess the data obtained in step S1 to construct a training dataset;

[0010] S3. Based on the undirected weighted graph construction scheme, logical gating mechanism and graph neural network model, construct an initial model for extracting individual features on the generation side of the power market;

[0011] S4. Using the training dataset obtained in step S2, train the initial model for extracting individual features on the generation side of the electricity market constructed in step S3 to obtain the individual feature extraction model on the generation side of the electricity market.

[0012] S5. Using the power market generation-side individual feature extraction model obtained in step S4, extract the features of the generation-side individuals in the target power market;

[0013] S6. Based on the feature data obtained in step S5, identify collusion among power generation groups in the target electricity market using a clustering algorithm.

[0014] Step S1, which involves obtaining historical data information about the electricity market, specifically includes the following steps:

[0015] Obtain historical data and information about the electricity market;

[0016] The data information includes the price declaration sequence, power declaration sequence, actual cleared power volume and actual cleared price of each market participant on the power generation side, as well as the rated capacity, fuel loss rate and fuel benchmark price of each participant's generating unit.

[0017] Step S2 involves preprocessing the data obtained in step S1 to construct a training dataset, specifically including the following steps:

[0018] Based on the data obtained in step S1, for each trading entity, the corresponding theoretical marginal cost is calculated based on the unit's fuel loss rate and the benchmark fuel price; if it is a new energy unit, the corresponding theoretical marginal cost is set to a fixed value.

[0019] The acquired data is preprocessed; the preprocessing includes missing value repair, outlier handling, unit unification, and data standardization.

[0020] Step S3, which describes the construction of an initial model for extracting individual features on the generation side of the electricity market based on an undirected weighted graph construction scheme, a logical gating mechanism, and a graph neural network model, specifically includes the following steps:

[0021] The initial model for extracting individual features on the generation side of the electricity market includes a transaction declaration behavior association graph module, a clearing result association graph module, a joint graph module, and a graph neural network module.

[0022] The outputs of the transaction declaration behavior association graph module and the clearing result association graph module are connected to the input of the joint graph module; the output of the joint graph module is connected to the graph neural network module.

[0023] Based on an undirected weighted graph construction scheme, a transaction declaration behavior association graph module is constructed; the transaction declaration behavior association graph module is used to obtain the association graph features corresponding to electricity market transaction declaration behaviors;

[0024] Based on the undirected weighted graph construction scheme, a clearing result association graph module is constructed; the clearing result association graph module is used to obtain the association graph features corresponding to the power market clearing results;

[0025] Based on the logic gating mechanism, a joint graph module is constructed. The joint graph module is used to construct joint graph features according to the association graph features corresponding to the electricity market transaction declaration behavior and the association graph features corresponding to the electricity market clearing results.

[0026] Based on the graph neural network model, a graph neural network module is constructed; the graph neural network module is used to extract individual features of the power generation side of the electricity market based on the joint graph features.

[0027] Step S3 specifically includes the following steps:

[0028] The processing steps of the transaction declaration behavior association diagram module include the following steps:

[0029] Based on the data input to the model, the consistency index between subject i and subject j in the electricity market is calculated. The consistency index includes the consistency index of bidding strategy direction, the difference index of bidding level, the correlation strength index of joint bidding, the similarity index of time-of-use electricity structure, and the synchronization index of high-price bidding behavior.

[0030] The consistency index of pricing strategy direction is calculated using the following formula:

[0031] In the formula This is an indicator of the consistency of pricing strategy direction between subject i and subject j. For the set of trading sessions; Number the time period; The bid price of subject i in time period t-1; The bid price of subject j in time period t-1; For symbolic functions, but , but , but ; For indicator functions, conditions If established ,condition If not true ;

[0032] The price level difference index is calculated using the following formula:

[0033] In the formula The indicator of the difference in pricing levels between subject i and subject j; This is a set, extremely small positive number, used to prevent the denominator from being 0;

[0034] The joint declaration association strength index is calculated using the following formula:

[0035] In the formula The joint declaration correlation strength index between subject i and subject j; Let i be the state variable indicating whether subject i participates in the declaration during time period t. If subject i participates in the declaration during time period t, then... If subject i did not participate in the declaration during time period t, then ; The state variable is whether subject j participates in the declaration during time period t. If subject j participates in the declaration during time period t, then... If entity j did not participate in the declaration during time period t, then ;

[0036] The time-of-use electricity structure similarity index is calculated using the following formula:

[0037] In the formula The similarity index of time-sharing electricity structure between subject i and subject j; The electricity consumption reported by subject i during time period t; The electricity volume declared by subject j in time period t;

[0038] The synchronization index of high-price bidding behavior is calculated using the following formula:

[0039] In the formula This is an indicator of the synchronicity of high-price bidding behavior between subject i and subject j. Let be the state variable indicating whether subject i engages in high-price bidding behavior during time period t. If subject i's bid price during time period t is greater than or equal to the set high-price bidding threshold, then... If the bid price of subject i in time period t is less than the set high bid threshold, then ; Let j be the state variable indicating whether subject j engages in high-price bidding behavior during time period t. If subject j's bid price during time period t is greater than or equal to the set high-price bidding threshold, then... If the bid price of subject j in time period t is less than the set high bid threshold, then... ;

[0040] The obtained index values Perform min-max normalization to obtain the normalized index value. ; s is the index number, and s takes a value from 1 to 5;

[0041] The overall consistency assessment value between entity i and entity j in the electricity market is calculated using the following formula. :

[0042] In the formula Let be the weight value of the s-th indicator, and ;

[0043] Set a significance threshold for consistency Construct an adjacency matrix for the relationship graph of transaction declaration behavior. , The element in the i-th row and j-th column The rules for determining the value are as follows: , For indicator functions, If established , If not true ;

[0044] Construct a relational graph corresponding to electricity market transaction declaration behavior. for ;in For point set, , Corresponding to the Nth entity in the electricity market; This is an edge set, corresponding to the collaborative reporting relationships between entities in the electricity market, and ;side The corresponding correlation strength value is ;

[0045] Clearing Result Relationship Module

[0046] Based on the data input to the model, calculate the common impact indicators between entity i and entity j in the electricity market; the common impact indicators include the marginal clearing contribution rate indicator, the joint winning bid rate indicator in the high price range indicator, and the bid premium rate indicator in the key period.

[0047] The marginal clearing contribution rate is calculated using the following formula:

[0048] In the formula The marginal clearing contribution rate indicator between subject i and subject j; As an indicator variable for whether subject i is in a marginally cleared state in time period t, and , Let t be the market clearing price for time period t. The set marginal clearing price tolerance, The amount of electricity won by subject i in time period t;

[0049] The joint winning bid rate for the high-price range is calculated using the following formula:

[0050] In the formula The joint success rate indicator for the high-price range between subject i and subject j; The state variable is whether time period t belongs to the high price range. When the market clearing price of time period t is greater than or equal to the set high price range threshold, ... When the market clearing price in time period t is less than the set high price range threshold ; Let be the state variable indicating whether subject i wins the bid in time period t. If the amount of electricity won by subject i in time period t is greater than 0, then... If the winning bid amount of subject i in time period t is equal to or less than 0, then ;

[0051] The premium rate for key periods is calculated using the following formula:

[0052] In the formula The premium rate indicator for key time periods between subject i and subject j; The state variable is whether time period t belongs to the critical time period. If time period t belongs to the critical time period, then... If time period t is not a critical time period, then ; Let i be the premium rate of the bid of subject i in time period t, and ;

[0053] The obtained index values Perform min-max normalization to obtain the normalized index value. ; s1 is the index number, and the value of s1 ranges from 1 to 3;

[0054] The comprehensive assessment value of the combined impact between subject i and subject j in the electricity market is calculated using the following formula. :

[0055] In the formula Let be the weight value of the s1th indicator, and ;

[0056] Set a threshold for the significance of common effects Construct the adjacency matrix of the clearing result association graph. , The element in the i-th row and j-th column The rules for determining the value are as follows: ;

[0057] Construct a correlation diagram corresponding to the results of the electricity market clearing. for ;in This is an edge set, corresponding to the abnormal relationships between entities in the electricity market at the level of transaction outcomes, and ;side The corresponding correlation strength value is ;

[0058] The processing procedure of the union graph module includes the following steps:

[0059] Based on the adjacency matrix of the obtained transaction declaration behavior association graph Adjacency matrix of the graph relating to clearing results Construct a joint correlation strength matrix ; The element in the i-th row and j-th column This represents the strength of the joint association between entity i and entity j in the electricity market, and ;

[0060] right For each element greater than 0 in the vector, construct the corresponding edge feature vector. Used to represent the participants in the electricity market System characteristics at both the behavioral and outcome levels ;

[0061] Constructing a joint graph for ;in, Let be the set of edges, and ; The node feature matrix, The i-th row Represented as , The rated installed capacity of the unit. This refers to the unit's fuel loss rate. For theoretical marginal cost, The code corresponding to the unit type; For edge feature vectors, ;

[0062] The graph neural network module includes a linear transformation layer, a multi-head attention hidden layer, and an output embedding layer;

[0063] Will , and Dimension alignment is performed through a linear transformation layer, feature extraction is performed through a multi-head attention hidden layer, and finally, the feature extraction vector is output through an output embedding layer; the feature extraction vector is used to represent the individual characteristics of the power generation side in the electricity market.

[0064] The output embedding layer is a fully connected mapping layer used to map the node features output from the last layer of the graph neural network to individual feature vectors of the power generation side entities in the electricity market; the processing of the output embedding layer is represented as follows:

[0065] In the formula Let i be the individual feature vector of the power generation side entity i in the electricity market; For activation functions; To output the weight matrix of the embedding layer; This is the bias vector for the output embedding layer; The node features of subject i are the output of the last layer of the graph neural network;

[0066] The processing steps of the multi-head attention hidden layer include the following:

[0067] A graph neural network is used to construct a multi-head attention hidden layer;

[0068] Let the current layer be layer l, and the characteristic of the center node i in the current layer be: The characteristics of node i's neighbor node j in the current layer are: The static edge feature vector between the center node i and its neighbor node j is: ;

[0069] The unnormalized influence weight of neighbor node j on center node i is calculated using the following formula. :

[0070] In the formula To define a function, and the expression is , The range is set as Parameters; This is the attention weight vector; The weight matrix of node features to be learned in the current layer; The edge features are the weight matrix to be learned in the current layer; This is a vector concatenation operation;

[0071] Using the softmax function Perform a normalization operation to obtain the attention coefficient. ;

[0072] The features at layer l+1 are calculated using the following formula. :

[0073] In the formula It is the sigmoid function; This is the edge feature transformation matrix, used to map edge features to the same dimensional space as node features; Subject i in the joint graph The set of neighboring nodes in;

[0074] The graph neural network used has L layers; the output features of the last layer are... After processing by the output embedding layer, individual characteristics of the power generation side in the electricity market are obtained. .

[0075] The training described in step S4 specifically includes the following steps:

[0076] The following formula is used as the loss function for the training process:

[0077] In the formula The value of the loss function; The set of positive samples, specifically The set of edges corresponding to the non-zero elements in the set; For the negative sample set, having in Randomly generated in the middle does not exist The set of edges in;

[0078] During training, the goal is to minimize the loss function value. The target is to be trained using the backpropagation algorithm.

[0079] Step S5 involves using the power market generator-side individual feature extraction model obtained in step S4 to extract features of generator-side individuals in the target power market. Specifically, this includes the following steps:

[0080] Acquire and preprocess data from the target electricity market;

[0081] The preprocessed data is input into the power market generator-side individual feature extraction model obtained in step S4 to obtain the features of the generator-side individuals in the target power market. and the corresponding intermediate parameters;

[0082] The intermediate parameters include the joint correlation strength between entity i and entity j on the generation side of the target electricity market. .

[0083] Step S6, which involves identifying collusion among power generation groups in the target electricity market based on the feature data obtained in step S5 and using a clustering algorithm, specifically includes the following steps:

[0084] The characteristics of the generation-side individuals in the target electricity market obtained in step S5 Density clustering algorithm is used for clustering: during clustering, each feature is calculated. The distance between them, and the distance is less than the set radius. Features whose density meets the set minimum number of points are grouped into the same cluster;

[0085] Treat each cluster as a group;

[0086] For the kth group The number of market entities it includes is The interconnected edges between various market entities are set as follows: ; Calculate the group Connection density for ; Represents the set of interconnected edges The total number of edges in;

[0087] The group is calculated using the following formula. Conspiracy Risk Index :

[0088] In the formula This is the first weighting coefficient set. This is the second weighting coefficient that is set; , and ;

[0089] Set a low threshold for the risk of collusion. High threshold for risk of collusion And make a judgment:

[0090] like Then the group is judged. The risk of collusion among the entities within the organization is high.

[0091] like Then the group is judged. The risk of collusion among the entities within the organization is medium.

[0092] like Then the group is judged. The risk of collusion among the entities involved is low.

[0093] This invention also provides a system for implementing the aforementioned method for identifying collusion among power generation groups in the electricity market, comprising a data acquisition module, a data processing module, a model building module, a model training module, a feature extraction module, and a collusion identification module; the data acquisition module, data processing module, model building module, model training module, feature extraction module, and collusion identification module are connected in series; the data acquisition module is used to acquire historical data information of the electricity market and upload the data information to the data processing module; the data processing module is used to preprocess the acquired data information according to the received data information to construct a training dataset and upload the data information to the model building module; the model building module is used to construct a scheme based on an undirected weighted graph, a logical gating mechanism, and graph neural networks according to the received data information. The network model constructs an initial model for extracting individual features from the power generation side of the electricity market and uploads the data to the model training module. The model training module trains the initial model using the received data and the obtained training dataset to obtain a power generation side individual feature extraction model, and uploads the data to the feature extraction module. The feature extraction module extracts features of individual power generation entities in the target electricity market using the obtained power generation side individual feature extraction model and uploads the data to the collusion identification module. The collusion identification module identifies collusion among power generation groups in the target electricity market based on the received data and the obtained feature data, using a clustering algorithm.

[0094] The present invention provides a method and system for identifying collusion among power generation groups in the power market. Based on an undirected weighted graph construction scheme, a logic gating mechanism, and a graph neural network model, a feature extraction model is constructed. Feature extraction and feature identification are performed on the target power market, and collusion identification is performed based on a clustering algorithm. This not only achieves collusion identification of power generation groups in the power market, but also has higher reliability and better accuracy. Attached Figure Description

[0095] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0096] Figure 2 This is a schematic diagram of the functional modules of the system of the present invention. Detailed Implementation

[0097] like Figure 1 The diagram shown is a flowchart of the method of the present invention: The method for identifying collusion among power generation groups in the power market disclosed in this invention includes the following steps:

[0098] S1. Obtain historical data information from the electricity market; specifically including the following steps:

[0099] To acquire historical data on the electricity market to characterize the trading behavior and clearing results of market participants;

[0100] The data information includes the price declaration sequence, power declaration sequence, actual cleared power volume and actual cleared price of each market participant on the power generation side, as well as the rated capacity, fuel loss rate and fuel benchmark price of each participant's generating unit;

[0101] S2. Preprocess the data obtained in step S1 to construct the training dataset; specifically, this includes the following steps:

[0102] Based on the data obtained in step S1, for each trading entity, the corresponding theoretical marginal cost is calculated based on the unit's fuel loss rate and the benchmark fuel price. If it is a new energy unit, the corresponding theoretical marginal cost is set to a fixed value. The marginal cost is used to characterize the entity's reasonable bidding benchmark under normal competitive conditions.

[0103] The acquired data is preprocessed; the preprocessing includes missing value repair, outlier handling, unit unification, and data standardization.

[0104] S3. Based on an undirected weighted graph construction scheme, a logical gating mechanism, and a graph neural network model, an initial model for extracting individual features from the generation side of the electricity market is constructed; specifically, the following steps are included:

[0105] The initial model for extracting individual features on the generation side of the electricity market includes a transaction declaration behavior association graph module, a clearing result association graph module, a joint graph module, and a graph neural network module.

[0106] The outputs of the transaction declaration behavior association graph module and the clearing result association graph module are connected to the input of the joint graph module; the output of the joint graph module is connected to the graph neural network module.

[0107] Based on an undirected weighted graph construction scheme, a transaction declaration behavior association graph module is constructed; the transaction declaration behavior association graph module is used to obtain the association graph features corresponding to electricity market transaction declaration behaviors;

[0108] Based on the undirected weighted graph construction scheme, a clearing result association graph module is constructed; the clearing result association graph module is used to obtain the association graph features corresponding to the power market clearing results;

[0109] Based on the logic gating mechanism, a joint graph module is constructed. The joint graph module is used to construct joint graph features according to the association graph features corresponding to the electricity market transaction declaration behavior and the association graph features corresponding to the electricity market clearing results.

[0110] Based on the graph neural network model, a graph neural network module is constructed; the graph neural network module is used to extract individual features of the generation side of the electricity market based on the joint graph features.

[0111] In practice, the following steps can be taken:

[0112] The processing steps of the transaction declaration behavior association diagram module include the following steps:

[0113] Based on the data input to the model, the consistency index between subject i and subject j in the electricity market is calculated. The consistency index includes the consistency index of bidding strategy direction, the difference index of bidding level, the correlation strength index of joint bidding, the similarity index of time-of-use electricity structure, and the synchronization index of high-price bidding behavior. The consistency index is used to characterize the potential strategy consistency relationship between subject i and j.

[0114] The consistency index of pricing strategy direction is calculated using the following formula:

[0115] In the formula This is an indicator of the consistency of pricing strategy direction between subject i and subject j. For the set of trading sessions; Number the time period; The bid price of subject i in time period t-1; The bid price of subject j in time period t-1; For symbolic functions, but , but , but ; For indicator functions, conditions If established ,condition If not true ;

[0116] The price level difference index is calculated using the following formula:

[0117] In the formula The indicator of the difference in pricing levels between subject i and subject j; This is a set, extremely small positive number, used to prevent the denominator from being 0;

[0118] The joint declaration association strength index is calculated using the following formula:

[0119] In the formula The joint declaration correlation strength index between subject i and subject j; Let i be the state variable indicating whether subject i participates in the declaration during time period t. If subject i participates in the declaration during time period t, then... If subject i did not participate in the declaration during time period t, then ; The state variable is whether subject j participates in the declaration during time period t. If subject j participates in the declaration during time period t, then... If entity j did not participate in the declaration during time period t, then ;

[0120] The time-of-use electricity structure similarity index is calculated using the following formula:

[0121] In the formula The similarity index of time-sharing electricity structure between subject i and subject j; The electricity consumption reported by subject i during time period t; The electricity volume declared by subject j in time period t;

[0122] The synchronization index of high-price bidding behavior is calculated using the following formula:

[0123] In the formula This is an indicator of the synchronicity of high-price bidding behavior between subject i and subject j. Let be the state variable indicating whether subject i engages in high-price bidding behavior during time period t. If subject i's bid price during time period t is greater than or equal to the set high-price bidding threshold, then... If the bid price of subject i in time period t is less than the set high bid threshold, then ; Let j be the state variable indicating whether subject j engages in high-price bidding behavior during time period t. If subject j's bid price during time period t is greater than or equal to the set high-price bidding threshold, then... If the bid price of subject j in time period t is less than the set high bid threshold, then... ;

[0124] The obtained index values Perform min-max normalization to obtain the normalized index value. ; s is the index number, and s takes a value from 1 to 5;

[0125] The overall consistency assessment value between entity i and entity j in the electricity market is calculated using the following formula. :

[0126] In the formula Let be the weight value of the s-th indicator, and ;

[0127] Set a significance threshold for consistency Construct an adjacency matrix for the relationship graph of transaction declaration behavior. , The element in the i-th row and j-th column The rules for determining the value are as follows: , For indicator functions, If established , If not true ;

[0128] Construct a relational graph corresponding to electricity market transaction declaration behavior. for ;in For point set, , Corresponding to the Nth entity in the electricity market; This is an edge set, corresponding to the collaborative reporting relationships between entities in the electricity market, and ;side The corresponding correlation strength value is ;

[0129] Clearing Result Relationship Module

[0130] Based on the data input to the model, the common influence indicators between subject i and subject j in the electricity market are calculated. The common influence indicators include the marginal clearing contribution rate indicator, the joint winning bid rate indicator in the high price range, and the bid premium rate in the key period. The common influence indicators are used to characterize the degree of common influence of the coordinated behavior between subject i and j on the market clearing results.

[0131] The marginal clearing contribution rate is calculated using the following formula:

[0132] In the formula The marginal clearing contribution rate indicator between subject i and subject j; As an indicator variable for whether subject i is in a marginally cleared state in time period t, and , Let t be the market clearing price for time period t. The set marginal clearing price tolerance, The amount of electricity won by subject i in time period t;

[0133] The joint winning bid rate for the high-price range is calculated using the following formula:

[0134] In the formula The joint success rate indicator for the high-price range between subject i and subject j; The state variable is whether time period t belongs to the high price range. When the market clearing price of time period t is greater than or equal to the set high price range threshold, ... When the market clearing price in time period t is less than the set high price range threshold ; Let be the state variable indicating whether subject i wins the bid in time period t. If the amount of electricity won by subject i in time period t is greater than 0, then... If the winning bid amount of subject i in time period t is equal to or less than 0, then ;

[0135] The premium rate for key periods is calculated using the following formula:

[0136] In the formula The premium rate indicator for key time periods between subject i and subject j; The state variable is whether time period t belongs to the critical time period. If time period t belongs to the critical time period, then... If time period t is not a critical time period, then ; Let i be the premium rate of the bid of subject i in time period t, and ;

[0137] The obtained index values Perform min-max normalization to obtain the normalized index value. ; s1 is the index number, and the value of s1 ranges from 1 to 3;

[0138] The comprehensive assessment value of the combined impact between subject i and subject j in the electricity market is calculated using the following formula. :

[0139] In the formula Let be the weight value of the s1th indicator, and ;

[0140] Set a threshold for the significance of common effects Construct the adjacency matrix of the clearing result association graph. , The element in the i-th row and j-th column The rules for determining the value are as follows: ;

[0141] Construct a correlation diagram corresponding to the results of the electricity market clearing. for ;in This is an edge set, corresponding to the abnormal relationships between entities in the electricity market at the level of transaction outcomes, and ;side The corresponding correlation strength value is ;

[0142] The processing procedure of the union graph module includes the following steps:

[0143] Based on the adjacency matrix of the obtained transaction declaration behavior association graph Adjacency matrix of the graph relating to clearing results Construct a joint correlation strength matrix ; The element in the i-th row and j-th column This represents the strength of the joint association between entity i and entity j in the electricity market, and ;

[0144] right For each element greater than 0 (indicating the existence of its corresponding edge), construct the corresponding edge feature vector. Used to represent the participants in the electricity market System characteristics at both the behavioral and outcome levels ;

[0145] Constructing a joint graph for ;in, Let be the set of edges, and This represents a potentially high-risk collusion relationship that has been verified by both the declaration behavior and the clearing results; The node feature matrix, The i-th row Represented as , The rated installed capacity of the unit. This refers to the unit's fuel loss rate. For theoretical marginal cost, The code corresponding to the unit type; For edge feature vectors, ;

[0146] The graph neural network module includes a linear transformation layer, a multi-head attention hidden layer, and an output embedding layer;

[0147] Will , and Dimension alignment is performed through a linear transformation layer, feature extraction is performed through a multi-head attention hidden layer, and finally, the feature extraction vector is output through an output embedding layer; the feature extraction vector is used to represent the individual characteristics of the power generation side in the electricity market.

[0148] The output embedding layer is a fully connected mapping layer used to map the node features output from the last layer of the graph neural network to individual feature vectors of the power generation side entities in the electricity market; the processing of the output embedding layer is represented as follows:

[0149] In the formula Let i be the individual feature vector of the power generation side entity i in the electricity market; For activation functions; To output the weight matrix of the embedding layer; This is the bias vector for the output embedding layer; The node features of subject i are the output of the last layer of the graph neural network;

[0150] The processing steps of the multi-head attention hidden layer include the following:

[0151] A graph neural network is used to construct a multi-head attention hidden layer;

[0152] Let the current layer be layer l, and the characteristic of the center node i in the current layer be: The characteristics of node i's neighbor node j in the current layer are: The static edge feature vector between the center node i and its neighbor node j is: ;

[0153] The unnormalized influence weight of neighbor node j on center node i is calculated using the following formula. :

[0154] In the formula To define a function, and the expression is , The range is set as Parameters; This is the attention weight vector, used to map the concatenated high-dimensional features to scalar attention coefficients; The weight matrix of node features to be learned in the current layer; The edge features are the weight matrix to be learned in the current layer; This is a vector concatenation operation; and Used to map node features and edge features of different dimensions to a unified feature space;

[0155] Using the softmax function Perform a normalization operation to obtain the attention coefficient. ;

[0156] The features at layer l+1 are calculated using the following formula. :

[0157] In the formula It is the sigmoid function; This is the edge feature transformation matrix, used to map edge features to the same dimensional space as node features; Subject i in the joint graph The set of neighboring nodes in;

[0158] The graph neural network used has L layers; the output features of the last layer are... After processing by the output embedding layer, individual characteristics of the power generation side in the electricity market are obtained. ;

[0159] S4. Using the training dataset obtained in step S2, train the initial model for extracting individual features on the generation side of the electricity market constructed in step S3 to obtain the individual feature extraction model on the generation side of the electricity market.

[0160] An unsupervised learning strategy based on link prediction is adopted. During training, the following formula is used as the loss function:

[0161] In the formula The value of the loss function; The set of positive samples, specifically The set of edges corresponding to the non-zero elements in the set; For the negative sample set, having in Randomly generated in the middle does not exist The set of edges in;

[0162] During training, the goal is to minimize the loss function value. The target is to be trained using the backpropagation algorithm;

[0163] S5. Using the power market generator-side individual feature extraction model obtained in step S4, extract the features of generator-side individuals in the target power market; specifically including the following steps:

[0164] Acquire and preprocess data from the target electricity market;

[0165] The preprocessed data is input into the power market generator-side individual feature extraction model obtained in step S4 to obtain the features of the generator-side individuals in the target power market. and the corresponding intermediate parameters;

[0166] The intermediate parameters include the joint correlation strength between entity i and entity j on the generation side of the target electricity market. ;

[0167] S6. Based on the feature data obtained in step S5, and using a clustering algorithm, identify collusion among power generation groups in the target electricity market; specifically, this includes the following steps:

[0168] The characteristics of the generation-side individuals in the target electricity market obtained in step S5 Density-based clustering algorithms (such as DBSCAN clustering algorithm) are used for clustering: during clustering, each feature is calculated. The distance between them, and the distance is less than the set radius. Features whose density meets the set minimum number of points are grouped into the same cluster;

[0169] Treat each cluster as a group;

[0170] For the kth group The number of market entities it includes is The interconnected edges between various market entities are set as follows: ; Calculate the group Connection density for ; Represents the set of interconnected edges The total number of edges in;

[0171] The group is calculated using the following formula. Conspiracy Risk Index :

[0172] In the formula This is the first weighting coefficient set. This is the second weighting coefficient that is set; , and ;

[0173] Set a low threshold for the risk of collusion. High threshold for risk of collusion And make a judgment:

[0174] like Then the group is judged. The risk of collusion among the entities within the organization is high.

[0175] like Then the group is judged. The risk of collusion among the entities within the organization is medium.

[0176] like Then the group is judged. The risk of collusion among the entities involved is low.

[0177] In practice, the analysis can be conducted by combining the transaction declaration behavior and transaction clearing results with the abnormal situation of relevant indicators, and finally a power market risk analysis report including the list of members, risk level, core abnormal indicators and recommended disposal measures can be generated.

[0178] This invention introduces a fusion mechanism, establishing connections only when market participants demonstrate significant correlation in both "collaborative bidding strategies" and "substantial impact on clearing results." This mechanism effectively addresses the problem of traditional methods misjudging high-price bids in normal market competition as collusion. By adding "outcome impact" as a verification condition, the false positive rate is significantly reduced, ensuring the objectivity and fairness of the identification results. Addressing the shortcomings of traditional statistical methods such as SVM and Isolation Forest, which treat market participants as independent samples and ignore inter-participant relationships, this invention constructs a joint graph containing node attributes and edge weights, and utilizes graph neural networks for representation learning. This enables the capture of hidden, non-linear topological collaborative relationships between market participants. Even if some participants' individual transaction characteristics are not obvious, as long as they occupy a core collaborative position in the network structure, the model can still identify them through the aggregation of neighbor information, significantly improving the sensitivity to identify complex group collusion. The present invention not only aggregates node features, but also uses the calculated specific indicators as edge feature vectors to directly participate in the calculation of attention weights. This enables the model to adaptively focus on connection edges with "high collaboration strength" or "high risk indicators" when learning node embeddings. This mechanism gives the model stronger interpretability and feature extraction capabilities, so that the final generated node embedding vector can more accurately reflect the collusion tendency of the subject.

[0179] like Figure 2The diagram shows the functional modules of the system of this invention: The system for implementing the method for identifying collusion among power generation groups in the power market disclosed in this invention includes a data acquisition module, a data processing module, a model building module, a model training module, a feature extraction module, and a collusion identification module; the data acquisition module, data processing module, model building module, model training module, feature extraction module, and collusion identification module are connected in series; the data acquisition module is used to acquire historical data information of the power market and upload the data information to the data processing module; the data processing module is used to preprocess the acquired data information according to the received data information to construct a training dataset and upload the data information to the model building module; the model building module is used to construct a scheme and logic based on an undirected weighted graph according to the received data information. A gating mechanism and a graph neural network model are used to construct an initial model for extracting individual features on the power generation side of the electricity market, and the data information is uploaded to the model training module. The model training module is used to train the initial model for extracting individual features on the power generation side of the electricity market using the received data information and the obtained training dataset, resulting in an individual feature extraction model for the power generation side of the electricity market. The data information is then uploaded to the feature extraction module. The feature extraction module is used to extract the features of individual power generation side users in the target electricity market using the received data information and the obtained feature data, based on a clustering algorithm, to identify collusion among power generation side groups in the target electricity market.

Claims

1. A method for identifying collusion among power generation groups in the electricity market, comprising the following steps: S1. Obtain historical data and information about the electricity market; S2. Preprocess the data obtained in step S1 to construct a training dataset; S3. Based on the undirected weighted graph construction scheme, logical gating mechanism and graph neural network model, construct an initial model for extracting individual features on the generation side of the power market; S4. Using the training dataset obtained in step S2, train the initial model for extracting individual features on the generation side of the electricity market constructed in step S3 to obtain the individual feature extraction model on the generation side of the electricity market. S5. Using the power market generation-side individual feature extraction model obtained in step S4, extract the features of the generation-side individuals in the target power market; S6. Based on the feature data obtained in step S5, identify collusion among power generation groups in the target electricity market using a clustering algorithm.

2. The method of claim 1, wherein Step S1, which involves obtaining historical data information about the electricity market, specifically includes the following steps: Obtain historical data and information about the electricity market; The data information includes the price declaration sequence, power declaration sequence, actual cleared power volume and actual cleared price of each market participant on the power generation side, as well as the rated capacity, fuel loss rate and fuel benchmark price of each participant's generating unit.

3. The method of claim 2, wherein Step S2 involves preprocessing the data obtained in step S1 to construct a training dataset, specifically including the following steps: Based on the data obtained in step S1, for each trading entity, the corresponding theoretical marginal cost is calculated based on the unit's fuel loss rate and the benchmark fuel price; if it is a new energy unit, the corresponding theoretical marginal cost is set to a fixed value. The acquired data is preprocessed; the preprocessing includes missing value repair, outlier handling, unit unification, and data standardization.

4. The method for identifying collusion among power generation groups in the electricity market according to claim 3, characterized in that... Step S3, which describes the construction of an initial model for extracting individual features on the generation side of the electricity market based on an undirected weighted graph construction scheme, a logical gating mechanism, and a graph neural network model, specifically includes the following steps: The initial model for extracting individual features on the generation side of the electricity market includes a transaction declaration behavior association graph module, a clearing result association graph module, a joint graph module, and a graph neural network module. The outputs of the transaction declaration behavior association graph module and the clearing result association graph module are connected to the input of the joint graph module; the output of the joint graph module is connected to the graph neural network module. Based on an undirected weighted graph construction scheme, a transaction declaration behavior association graph module is constructed; the transaction declaration behavior association graph module is used to obtain the association graph features corresponding to electricity market transaction declaration behaviors; Based on the undirected weighted graph construction scheme, a clearing result association graph module is constructed; the clearing result association graph module is used to obtain the association graph features corresponding to the power market clearing results; Based on the logic gating mechanism, a joint graph module is constructed; The joint graph module is used to construct joint graph features based on the association graph features corresponding to electricity market transaction declaration behavior and the association graph features corresponding to electricity market clearing results; Based on the graph neural network model, a graph neural network module is constructed; the graph neural network module is used to extract individual features of the power generation side of the electricity market based on the joint graph features.

5. The method for identifying collusion among power generation groups in the electricity market according to claim 4, characterized in that... Step S3 specifically includes the following steps: The processing steps of the transaction declaration behavior association diagram module include the following steps: Based on the data input to the model, the consistency index between subject i and subject j in the electricity market is calculated. The consistency index includes the consistency index of bidding strategy direction, the difference index of bidding level, the correlation strength index of joint bidding, the similarity index of time-of-use electricity structure, and the synchronization index of high-price bidding behavior. The consistency index of pricing strategy direction is calculated using the following formula: In the formula This is an indicator of the consistency of pricing strategy direction between subject i and subject j. For the set of trading sessions; Number the time period; The bid price of subject i in time period t-1; The bid price of subject j in time period t-1; For symbolic functions, but , but , but ; For indicator functions, conditions If established ,condition If not true ; The price level difference index is calculated using the following formula: In the formula The indicator of the difference in pricing levels between subject i and subject j; This is a set, extremely small positive number, used to prevent the denominator from being 0; The joint declaration association strength index is calculated using the following formula: In the formula The joint declaration correlation strength index between subject i and subject j; Let i be the state variable indicating whether subject i participates in the declaration during time period t. If subject i participates in the declaration during time period t, then... If subject i did not participate in the declaration during time period t, then ; The state variable is whether subject j participates in the declaration during time period t. If subject j participates in the declaration during time period t, then... If entity j did not participate in the declaration during time period t, then ; The time-of-use electricity structure similarity index is calculated using the following formula: In the formula The similarity index of time-sharing electricity structure between subject i and subject j; The electricity consumption reported by subject i during time period t; The electricity volume declared by subject j in time period t; The synchronization index of high-price bidding behavior is calculated using the following formula: In the formula This is an indicator of the synchronicity of high-price bidding behavior between subject i and subject j. Let be the state variable indicating whether subject i engages in high-price bidding behavior during time period t. If subject i's bid price during time period t is greater than or equal to the set high-price bidding threshold, then... If the bid price of subject i in time period t is less than the set high bid threshold, then ; Let j be the state variable indicating whether subject j engages in high-price bidding behavior during time period t. If subject j's bid price during time period t is greater than or equal to the set high-price bidding threshold, then... If the bid price of subject j in time period t is less than the set high bid threshold, then... ; The obtained index values Perform min-max normalization to obtain the normalized index value. ; s is the index number, and s takes a value from 1 to 5; The overall consistency assessment value between entity i and entity j in the electricity market is calculated using the following formula. : In the formula Let be the weight value of the s-th indicator, and ; Set a significance threshold for consistency Construct an adjacency matrix for the transaction declaration behavior association graph. , The element in the i-th row and j-th column The rules for determining the value are as follows: , For indicator functions, If established , If not true ; Construct a relational graph corresponding to electricity market transaction declaration behavior. for ;in For point set, , This corresponds to the Nth entity in the electricity market; This is an edge set, corresponding to the collaborative reporting relationships between entities in the electricity market, and ;side The corresponding correlation strength value is ; Clearing Result Relationship Module Based on the data input to the model, calculate the common impact indicators between entity i and entity j in the electricity market; the common impact indicators include the marginal clearing contribution rate indicator, the joint winning bid rate indicator in the high price range indicator, and the bid premium rate indicator in the key period. The marginal clearing contribution rate is calculated using the following formula: In the formula The marginal clearing contribution rate indicator between subject i and subject j; As an indicator variable for whether subject i is in a marginally cleared state in time period t, and , Let t be the market clearing price for time period t. The set marginal clearing price tolerance, The amount of electricity won by subject i in time period t; The joint winning bid rate for the high-price range is calculated using the following formula: In the formula The joint success rate indicator for the high-price range between subject i and subject j; The state variable is whether time period t belongs to the high price range. When the market clearing price of time period t is greater than or equal to the set high price range threshold, ... When the market clearing price in time period t is less than the set high price range threshold ; Let be the state variable indicating whether subject i wins the bid in time period t. If the amount of electricity won by subject i in time period t is greater than 0, then... If the winning bid amount of subject i in time period t is equal to or less than 0, then ; The premium rate for key periods is calculated using the following formula: In the formula The premium rate indicator for key time periods between subject i and subject j; The state variable is whether time period t belongs to the critical time period. If time period t belongs to the critical time period, then... If time period t is not a critical time period, then ; Let i be the premium rate of the bid of subject i in time period t, and ; The obtained index values Perform min-max normalization to obtain the normalized index value. ; s1 is the index number, and the value of s1 ranges from 1 to 3; The comprehensive assessment value of the combined impact between subject i and subject j in the electricity market is calculated using the following formula. : In the formula Let be the weight value of the s1th indicator, and ; Set a threshold for the significance of common effects Construct the adjacency matrix of the clearing result association graph. , The element in the i-th row and j-th column The rules for determining the value are as follows: ; Construct a correlation diagram corresponding to the results of the electricity market clearing. for ;in This is an edge set, corresponding to the abnormal relationships between entities in the electricity market at the level of transaction outcomes, and ;side The corresponding correlation strength value is ; The processing procedure of the union graph module includes the following steps: Based on the adjacency matrix of the obtained transaction declaration behavior association graph Adjacency matrix of the graph relating to clearing results Construct a joint correlation strength matrix ; The element in the i-th row and j-th column This represents the strength of the joint association between entity i and entity j in the electricity market, and ; right For each element greater than 0 in the vector, construct the corresponding edge feature vector. Used to represent the participants in the electricity market System characteristics at both the behavioral and outcome levels ; Constructing a joint graph for ;in, Let be the set of edges, and ; The node feature matrix, The i-th row Represented as , The rated installed capacity of the unit. This refers to the unit's fuel loss rate. For theoretical marginal cost, The code corresponding to the unit type; For edge feature vectors, ; The graph neural network module includes a linear transformation layer, a multi-head attention hidden layer, and an output embedding layer; Will , and Dimension alignment is performed through a linear transformation layer, feature extraction is performed through a multi-head attention hidden layer, and finally, the feature extraction vector is output through an output embedding layer; the feature extraction vector is used to represent the individual characteristics of the power generation side in the electricity market. The output embedding layer is a fully connected mapping layer used to map the node features output from the last layer of the graph neural network to individual feature vectors of the power generation side entities in the electricity market; the processing of the output embedding layer is represented as follows: In the formula Let i be the individual feature vector of the power generation side entity i in the electricity market; For activation functions; To output the weight matrix of the embedding layer; This is the bias vector for the output embedding layer; The node features of subject i are the output of the last layer of the graph neural network; The processing steps of the multi-head attention hidden layer include the following: A graph neural network is used to construct a multi-head attention hidden layer; Let the current layer be layer l, and the characteristics of the center node i in the current layer be: The characteristics of node i's neighbor node j in the current layer are: The static edge feature vector between the center node i and its neighbor node j is: ; The unnormalized influence weight of neighbor node j on center node i is calculated using the following formula. : In the formula To define a function, and the expression is , The range is set as Parameters; This is the attention weight vector; The weight matrix of node features to be learned in the current layer; This is the weight matrix of the edge features to be learned in the current layer; This is a vector concatenation operation; Using the softmax function Perform a normalization operation to obtain the attention coefficient. ; The features at layer l+1 are calculated using the following formula. : In the formula It is the sigmoid function; This is the edge feature transformation matrix, used to map edge features to the same dimensional space as node features; Subject i in the joint graph The set of neighboring nodes in; The graph neural network used has L layers; the output features of the last layer are... After processing by the output embedding layer, individual characteristics of the power generation side in the electricity market are obtained. .

6. The method for identifying collusion among power generation groups in the electricity market according to claim 5, characterized in that... The training described in step S4 specifically includes the following steps: The following formula is used as the loss function for the training process: In the formula The value of the loss function; The set of positive samples, specifically The set of edges corresponding to the non-zero elements in the set; For the negative sample set, having in Randomly generated in the middle does not exist The set of edges in; During training, the goal is to minimize the loss function value. The target is to be trained using the backpropagation algorithm.

7. The method for identifying collusion among power generation groups in the electricity market according to claim 6, characterized in that... Step S5 involves using the power market generator-side individual feature extraction model obtained in step S4 to extract features of generator-side individuals in the target power market. Specifically, this includes the following steps: Acquire and preprocess data from the target electricity market; The preprocessed data is input into the power market generator-side individual feature extraction model obtained in step S4 to obtain the features of the generator-side individuals in the target power market. and the corresponding intermediate parameters; The intermediate parameters include the joint correlation strength between entity i and entity j on the generation side of the target electricity market. .

8. The method for identifying collusion among power generation groups in the electricity market according to claim 7, characterized in that... Step S6, which involves identifying collusion among power generation groups in the target electricity market based on the feature data obtained in step S5 and using a clustering algorithm, specifically includes the following steps: The characteristics of the generation-side individuals in the target electricity market obtained in step S5 Density clustering algorithm is used for clustering: during clustering, each feature is calculated. The distance between them, and the distance is less than the set radius. Features whose density meets the set minimum number of points are grouped into the same cluster; Treat each cluster as a group; For the kth group The number of market entities it includes is The interconnected edges between various market entities are set as follows: ; Calculate the group Connection density for ; Represents the set of interconnected edges The total number of edges in; The group is calculated using the following formula. Conspiracy Risk Index : In the formula This is the first weighting coefficient set. This is the second weighting coefficient that is set; , and ; Set a low threshold for the risk of collusion. High threshold for risk of collusion And make a judgment: like Then the group is judged. The risk of collusion among the entities within the organization is high. like Then the group is judged. The risk of collusion among the entities within the organization is medium. like Then the group is judged. The risk of collusion among the entities involved is low.

9. A system for implementing the electricity market generation-side group collusion identification method according to any one of claims 1 to 8, characterized in that... It includes a data acquisition module, a data processing module, a model building module, a model training module, a feature extraction module, and a collusion detection module; the data acquisition module, data processing module, model building module, model training module, feature extraction module, and collusion detection module are connected in series; the data acquisition module is used to acquire historical data information of the electricity market and upload the data information to the data processing module; The data processing module is used to preprocess the acquired data information based on the received data information in order to build a training dataset and upload the data information to the model building module; The model building module is used to construct an initial model for extracting individual features of the power generation side of the electricity market based on the received data information, an undirected weighted graph construction scheme, a logical gating mechanism, and a graph neural network model, and upload the data information to the model training module. The model training module is used to train the initial model for extracting individual features on the power generation side of the electricity market based on the received data information and the obtained training dataset, thereby obtaining the individual feature extraction model for the power generation side of the electricity market, and uploading the data information to the feature extraction module; The feature extraction module is used to extract the features of the generator-side individuals in the target electricity market based on the received data information and the obtained electricity market generator-side individual feature extraction model, and then upload the data information to the collusion identification module. The collusion identification module is used to identify collusion among power generation groups in the target electricity market based on the received data and the obtained feature data, using a clustering algorithm.

Images