Electric power spot market subject collusion quotation identification method, device and equipment

Abstract

The invention provides an electricity spot market subject collusion quotation identification method, device and equipment, and relates to the technical field of electricity marketization transactions, and the method comprises the steps: obtaining the declaration data of each market subject on a target day; the target day is any day in a preset supervision period; calculating a feature vector of each declaration data; the feature vector comprises quoted prices under different percentage output levels of the corresponding unit, and each percentage output level of the unit corresponds to one capacity point; performing parameter optimization on a preset kernel function according to all the feature vectors to obtain a target kernel function; based on the target kernel function, performing density clustering on all the feature vectors to obtain a clustering result of the target day; and after a daily clustering result in a preset supervision period is obtained, performing collusion quotation behavior identification on each market subject according to all the clustering results. The method can improve the accuracy of the recognition result of the collusion quotation behavior of the market subject of the electric power spot market, and gives consideration to the recognition efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power market-oriented trading, and in particular to a method, device and equipment for identifying collusive bidding by market players in the electric power spot market. Background Art

[0002] With the advancement of the construction of the electricity spot market, some pilot areas have successively seen market players collude to bid and obtain common profits, which has seriously damaged the fairness of market competition. The concept of collusive bidding behavior is defined as the behavior of two or more market players who do not have an actual control relationship to coordinate competition through collaborative bidding to manipulate prices. Based on the definition of collusive bidding behavior, collusive bidding behavior can be divided into two categories: one is the similarity of quotations in on-site transactions; the other is tacit collusion in over-the-counter transactions. The former is a cooperative collusion, and the results of the conspiracy will be directly reflected in the declared data of the market players. It is the preferred way for market players to collude and is also the focus of supervision of collusive bidding behavior; the latter is a non-cooperative collusion, which is formed and maintained without negotiation and agreement, based only on common interests. There is no factual evidence, and it is difficult to regulate it. Therefore, at this stage, the research on the identification and supervision of collusive bidding in the electricity spot market is centered on on-site transactions.

[0003] At present, the methods for identifying the collusive bidding behavior of market players in the electricity spot market include the sorted multivariate logit model, the semi-supervised support vector machine model and the decision tree classification algorithm. The sorted multivariate logit model constructs the unit bidding regression equation and uses the likelihood ratio test to analyze the differences in the unit bidding strategies, thereby identifying the unit's collusive bidding behavior. However, the sorted multivariate logit model assumes that the coefficients of all independent variables are the same at all levels, but in fact, the importance of the indicators used to identify the collusive bidding of market players is not the same, which will lead to deviations in the classification results; the semi-supervised support vector machine model can classify unlabeled data, but the model relies on the smoothing assumption (similar points have similar labels) and the low-density separation assumption (the decision boundary is located in the low-density area). The massive declared data in reality may not necessarily meet this assumption, resulting in low accuracy of the model classification results; the decision tree classification algorithm has low requirements for data preprocessing, but for declared data with a large number of features or a large number of feature values, the computational complexity of constructing a decision tree is high, and it often takes a long time to train the model, making the recognition efficiency of collusive bidding behavior low.

[0004] Based on this, how to improve the accuracy of the identification results of collusive bidding behavior of market players in the electricity spot market while taking into account its identification efficiency has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of this, in order to solve the above-mentioned technical problems, a method, device and equipment for identifying collusive bidding by market players in the electricity spot market are provided.

[0006] The present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for identifying collusive bidding by market players in a power spot market, comprising:

[0008] Obtaining the declared data of each market player on the target day; the target day is any day within the preset regulatory period, and the declared data includes the capacity and quotation corresponding to different unit output stages;

[0009] Calculating a characteristic vector of each of the declared data; the characteristic vector includes quotations at different percentage output levels of the corresponding unit, each percentage output level of the unit corresponding to a capacity point;

[0010] According to all the eigenvectors, the preset kernel function is optimized to obtain a target kernel function;

[0011] Based on the target kernel function, density clustering is performed on all the feature vectors to obtain a clustering result of the target day;

[0012] After obtaining the daily clustering results within the preset regulatory period, the collusive bidding behavior of each market player is identified based on all the clustering results.

[0013] Optionally, the calculating of the characteristic vector of each of the declared data specifically includes:

[0014] For each of the declared data, determining each capacity point of the declared data; each capacity point is d capacity points divided evenly according to the capacity between the minimum technical output and the maximum technical output of the corresponding unit;

[0015] Determine the quotation corresponding to each capacity point according to the declared data;

[0016] Using the quotations corresponding to the capacity points, a feature vector of the declared data is obtained.

[0017] Optionally, the preset kernel function is a Gaussian kernel function;

[0018] The step of optimizing the parameters of the preset kernel function according to all the feature vectors to obtain the target kernel function specifically includes:

[0019] Based on Silverman's method, the bandwidth parameter of the Gaussian kernel function is optimized according to all the eigenvectors to obtain the target kernel function.

[0020] Optionally, the step of performing density clustering on all the feature vectors based on the target kernel function to obtain a clustering result of the target day specifically includes:

[0021] Based on the target kernel function, calculating the kernel density of each of the feature vectors;

[0022] Determining a density threshold according to the kernel density of each of the eigenvectors;

[0023] Select a first target feature vector from all currently unprocessed feature vectors;

[0024] Determine whether the kernel density of the first target feature vector is greater than or equal to the density threshold;

[0025] If the kernel density of the current first target feature vector is less than the density threshold, it is determined that the current first target feature vector is not a core feature vector, and a step of selecting a first target feature vector from all current unprocessed feature vectors is performed;

[0026] If the kernel density of the current first target feature vector is greater than or equal to the density threshold, determining the current first target feature vector as a core feature vector;

[0027] Among all the current unprocessed feature vectors, a second target feature vector of the same cluster class as the current first target feature vector is selected to obtain a cluster of the current first target feature vector; wherein the cluster contains the current first target feature vector and its second target feature vector, and the feature vectors in the cluster are all marked as processed feature vectors;

[0028] The step of selecting a first target feature vector from all currently unprocessed feature vectors is performed until clustering is completed for all the feature vectors.

[0029] Optionally, determining a density threshold according to the kernel density of each of the feature vectors specifically includes:

[0030] The median of the kernel density of each of the feature vectors is determined as the density threshold.

[0031] Optionally, selecting, from all current unprocessed feature vectors, a second target feature vector that is of the same cluster type as the current first target feature vector to obtain a cluster of the current first target feature vector specifically includes:

[0032] Select a third target feature vector from all currently unprocessed feature vectors;

[0033] Calculating the Euclidean distance between the third target feature vector and the first target feature vector;

[0034] According to the current Euclidean distance, determining whether the current third target feature vector and the current first target feature vector are directly density-reachable, density-reachable, and density-connected;

[0035] If the third target feature vector is directly density-reachable, density-reachable, and density-connected to the first target feature vector, then determining that the third target feature vector is a second target feature vector of the first target feature vector;

[0036] Classifying the currently determined second target feature vector into the cluster of the currently determined first target feature vector;

[0037] Marking the feature vectors of the cluster of the first target feature vector at present with the goal of marking each feature vector in the cluster of the first target feature vector at present as a processed feature vector;

[0038] Determining whether the search for the second target feature vector for the current first target feature vector is completed;

[0039] If the search for the second target feature vector for the current first target feature vector has been completed, then the second target feature vector search process is terminated;

[0040] If the search for the second target feature vector for the current first target feature vector has not been completed, a step of selecting a third target feature vector from all current unprocessed feature vectors is performed.

[0041] Optionally, the step of identifying the bidding collusion behavior of each of the market players based on all the clustering results specifically includes:

[0042] For each of the clustering results, a target cluster is screened for the clustering results; all market entities corresponding to the target cluster have different equity relationships, and the excess returns of all market entities exceed a preset return threshold;

[0043] When the target clusters are screened out, for each target cluster, all target Euclidean distances corresponding to the target cluster are normalized, and the normalized value is determined as the single-day quotation similarity between the corresponding two feature vectors; wherein all target Euclidean distances corresponding to the target cluster include the Euclidean distance between every two feature vectors of the target cluster;

[0044] For each target feature vector pair, the average value of all the single-day quotation similarities of the target feature vector pair is calculated to obtain the multi-day quotation similarity of the target feature vector pair; the target feature vector pair is the feature vector pair corresponding to the single-day quotation similarity;

[0045] Determine whether the multi-day quotation similarity is greater than a target similarity; the target similarity is the average value of all the single-day quotation similarities;

[0046] If the multi-day quotation similarity is greater than the target similarity, then the two market entities corresponding to the target feature vector pair are determined to be highly similar market entities.

[0047] Optionally, before calculating the characteristic vector of each declared data, the method further includes:

[0048] The declared data is subjected to denoising processing.

[0049] In a second aspect, the present invention provides a device for identifying collusive bidding by market players in a power spot market, comprising:

[0050] An acquisition module is used to acquire the declared data of each market player on a target day; the target day is any day within a preset regulatory period, and the declared data includes the capacity and quotation corresponding to different unit output stages;

[0051] A calculation module, used for calculating a characteristic vector of each of the declared data; the characteristic vector includes quotations at different percentage output levels of the corresponding unit, and each percentage output level of the unit corresponds to a capacity point;

[0052] An optimization module, used to optimize the parameters of a preset kernel function according to all the feature vectors to obtain a target kernel function;

[0053] A clustering module, used for performing density clustering on all the feature vectors based on the target kernel function to obtain a clustering result of the target day;

[0054] The identification module is used to identify the collusive bidding behavior of each market entity based on all the clustering results after obtaining the daily clustering results within the preset supervision period.

[0055] In a third aspect, the present invention provides a computer device, comprising:

[0056] at least one processor; and,

[0057] a memory communicatively connected to the at least one processor; wherein,

[0058] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to implement the method for identifying collusive bidding by market entities in the electricity spot market as described above.

[0059] The present invention adopts the above technical scheme, a method for identifying collusive bidding of market players in the electricity spot market, including: obtaining the declared data of each market player on the target day; the target day is any day within the preset regulatory period, and the declared data includes the capacity and quotations corresponding to different unit output sections; calculating the characteristic vector of each declared data; the characteristic vector includes the quotations at different percentage output levels of the corresponding unit, and each percentage output level of the unit corresponds to a capacity point; according to all characteristic vectors, the preset kernel function is optimized to obtain the target kernel function; based on the target kernel function, all characteristic vectors are density clustered to obtain the clustering result of the target day; after obtaining the daily clustering results within the preset regulatory period, according to all clustering results, the collusive bidding behavior of each market player is identified.

[0060] Based on this, by uniformly preprocessing each declared data into a feature vector containing quotations at different percentage output levels of the corresponding unit, it is possible to ensure that the declared capacity and quotations of market entities at different capacity levels are at the same observable and comparable level, so as to improve the accuracy of quotation similarity identification. In addition, the use of an improved density clustering recognition model based on the kernel method can reduce the dependence on density clustering parameters to a certain extent and improve the accuracy of similarity identification of market entity declared data. In addition, the present invention does not need to spend a long time to train the model. Therefore, the present invention can improve the accuracy of the identification results of collusion in quotation behavior of market entities in the electricity spot market, while taking into account its recognition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0062] Figure 1 It is a flow chart of a method for identifying collusive bidding by market players in the electricity spot market provided by an embodiment of the present invention;

[0063] Figure 2 It is a structural schematic diagram of a device for identifying collusive bidding by market entities in a power spot market provided by an embodiment of the present invention;

[0064] Figure 3 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0065] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0066] Example

[0067] Figure 1 1 is a flow chart of a method for identifying collusive bidding by market players in the electricity spot market provided by an embodiment of the present invention. Figure 1 As shown in the figure, the method for identifying collusive bidding by market players in the electricity spot market includes:

[0068] Step 101: Obtain the declared data of each market player on the target day; the target day is any day within the preset regulatory period, and the declared data includes the capacity and quotation corresponding to different unit output sections.

[0069] Specifically, the data tags included in the declared data may include date, unit ID, start output (MW), end output (MW), quotation and declared segment number. Among them, each market entity corresponds to a unit ID, that is, the market entity and the unit ID correspond one to one.

[0070] In a specific example, the following Table 1 shows the declaration data of market subject A (market subject A is any one of the market subjects) on the target day:

[0071] Table 1

[0072]

[0073] As can be seen from Table 1 above, market player A adopts a four-stage declaration mode, corresponding to declaration stage numbers 1-4, and the corresponding date, unit ID, starting output (MW), ending output (MW), and quotation of each declaration stage are shown in Table 1. It is generally believed that the starting output of the first stage (i.e. 100MW) is the minimum technical output, and the ending output of the last stage (i.e. 600MW) is the maximum technical output.

[0074] Step 102: Calculate the characteristic vector of each declared data; the characteristic vector includes the quotations at different percentage output levels of the corresponding unit, and each percentage output level of the unit corresponds to a capacity point.

[0075] Specifically, the market entities in the electricity spot market use an N-stage step-by-step incremental declaration curve for declaration. The unit divides the available capacity from the minimum technical output to the maximum technical output into N stages (as shown in Table 1, N=4, corresponding to declaration stage numbers 1-4 respectively). The N-stage declared capacity and the corresponding quotations constitute the declared quantity-price curve of the unit. In order to characterize the original declaration data of the unit to the greatest extent, the unit declaration curve can be regarded as a collection of the starting output point and its corresponding quotation and the ending output point and its corresponding quotation of each declaration stage. However, it is necessary to take into account the different capacities of the units, and it is impossible to directly compare each capacity point of different units. Therefore, it is necessary to pre-process the unit capacity points to ensure that the declared capacity and quotations of market entities of different capacity levels are at the same observable and comparable level.

[0076] In the embodiment of the present invention, calculating the characteristic vector of each declared data may specifically include:

[0077] (1) For each declared data, determine each capacity point of the declared data; each capacity point is d capacity points that are evenly divided according to the capacity between the minimum technical output and the maximum technical output of the corresponding unit.

[0078] Here, d can be determined by the maximum number of declaration segments specified in the spot market operating rules of each region or the maximum number of declaration segments generally adopted by market players.

[0079] d capacity points can be expressed as: 0% capacity point, Capacity point, Capacity point, ..., 100% capacity point.

[0080] (2) Determine the quotation corresponding to each capacity point based on the declared data.

[0081] (3) Using the quotations corresponding to each capacity point, the characteristic vector of the declared data is obtained.

[0082] Specifically, the expression of the eigenvector X is as follows:

[0083]

[0084] Each element in X is a quote, for example, P 0％ The quotation corresponds to the 0% capacity point.

[0085] The following example illustrates the specific implementation process of calculating the characteristic vectors of each reported data.

[0086] Continuing with the above example, for market player A, assuming that the value of d is 6, then the capacity between the minimum technical output (i.e. 100MW) and the maximum technical output (i.e. 600MW) is evenly divided into 6 capacity points, namely: 0% capacity point, 20% capacity point, 40% capacity point, 60% capacity point, 80% capacity point, 100% capacity point. From Table 1, it can be seen that the technical outputs corresponding to the 6 capacity points are: 100MW, 200MW, 300MW, 400MW, 500MW, 600MW.

[0087] By querying Table 1 again, we can see that the corresponding quotations for 100MW, 200MW, 300MW, 400MW, 500MW, and 600MW are: 0, 0, 300, 400, 400, and 550. Therefore, the characteristic vector of the declared data of market entity A is: X = [0, 0, 300, 400, 400, 550].

[0088] Step 103: Optimize the parameters of the preset kernel function according to all the feature vectors to obtain the target kernel function.

[0089] In the embodiment of the present invention, the preset kernel function may be a Gaussian kernel function.

[0090] The formula of Gaussian kernel function is:

[0091]

[0092] Wherein, x and y are two data points, and in the present invention, x and y are two feature vectors respectively. K(x, y) represents the similarity between x and y; h is a bandwidth parameter, which is used to control the smoothness of the kernel function.

[0093] The selection of bandwidth parameter h can be optimized by cross-validation method to ensure the accuracy of kernel density estimation. Based on this, according to all feature vectors, the preset kernel function is optimized to obtain the target kernel function, which can specifically include:

[0094] Based on Silverman's method, the bandwidth parameter of the Gaussian kernel function is optimized according to all eigenvectors to obtain the target kernel function.

[0095] Specifically, it can be seen from the above content that the present invention divides the unit declaration data into d-dimensional feature vectors. Therefore, when using Silverman's method to optimize the bandwidth parameter h, the feature vector X corresponding to the declaration data of each market entity on the target day is firstly subjected to dimensionality reduction processing, and then the optimization range of the bandwidth parameter h is set. The specific steps for optimizing the bandwidth parameter h of the Gaussian kernel function are as follows:

[0096] (1) Data standardization.

[0097] The data of each dimension is standardized so that the mean of all feature vectors of the target day is 0 and the standard deviation is 1. The calculation formula used is as follows:

[0098]

[0099] here,

[0100] Among them, x ij is the value of the i-th eigenvector X in the j-th dimension; μ j is the mean of all eigenvectors in the jth dimension; σ j is the standard deviation of all eigenvectors in the jth dimension; n is the total number of eigenvectors X; the value range of j is 0 to d.

[0101] (2) Calculate the covariance matrix.

[0102] Calculate the standardized data matrix Z (i.e. Z ij The covariance matrix Σ of the matrix (composed of ) is calculated as follows:

[0103]

[0104] Where Z is the data matrix obtained after standardization in step (1) above.

[0105] (3) Eigenvalue decomposition.

[0106] Perform eigenvalue decomposition on the covariance matrix Σ to obtain eigenvalues ​​and eigenvectors:

[0107] Σ=VΛV T ......(5)

[0108] Among them, V is the eigenvector matrix, each column is an eigenvector; Λ is a diagonal matrix, and the elements on the diagonal are eigenvalues.

[0109] (4) Select the main ingredients.

[0110] Select the first two principal components (i.e. the eigenvectors corresponding to the two largest eigenvalues) according to the size of the eigenvalue:

[0111] Var(PC i )=λ i ......(6)

[0112] Among them, λ i is the eigenvalue of the i-th principal component; Var(PC i ) is the variance of the ith principal component.

[0113] (5) Calculate the principal component variance.

[0114] Select the variances of the first k principal components and calculate the average of these variances as the representative variance:

[0115]

[0116] Where k is the number of principal components selected. In this application, k = 2; λ i is the eigenvalue of the i-th principal component.

[0117] (6) Calculate the bandwidth parameter h.

[0118] According to Silverman's method, the bandwidth parameter is set and the calculation formula is:

[0119] h=0.9·σ PCA ·n -1 / 5 ......(8)

[0120] (7) Determine the candidate range of bandwidth parameters.

[0121] In order to optimize the bandwidth parameter, a candidate range of the bandwidth parameter is usually determined based on the bandwidth parameter calculated by Silverman's method. The upper and lower limits of the bandwidth parameter are set as:

[0122] h min =α·h......(9)

[0123] h max =β·h......(10)

[0124] Among them, α and β are adjustment parameters, and the value range is 0.5 to 2.0. For example, α=0.5 and β=2.0 can be set.

[0125] (8) Determine the number of candidate bandwidth parameter values

[0126] Use the rule of thumb based on a single day's reported data to determine the number of candidate bandwidth parameter values:

[0127]

[0128] Where m is the number of bandwidth parameters h; n is the total number of feature vectors X.

[0129] (9) Based on the cross-validation method, the optimal value of the bandwidth parameter h is determined according to the bandwidth parameter candidate range determined in step (7) and the number of bandwidth parameter candidate values ​​determined in step (8).

[0130] Specifically, the data set is divided into Q parts (when 0 < n < 100, Q is taken as 10; when n ≥ 100, Q is taken as 5). Among them, Q - 1 parts are used to train the model, and the remaining 1 part is used to verify the model. This process is repeated Q times, with different subsets used as the validation set each time, and the remaining subsets used as the training set. For each candidate value h, the mean squared error of Q - fold cross - validation is calculated. The h value with the minimum mean squared error is the optimal value.

[0131] Step 104: Based on the target kernel function, perform density clustering on all feature vectors to obtain the clustering result of the target day.

[0132] In the embodiment of the present invention, based on the target kernel function, performing density clustering on all feature vectors to obtain the clustering result of the target day may specifically include:

[0133] (1) Based on the target kernel function, calculate the kernel density of each feature vector.

[0134] Specifically, the calculation formula of the kernel density is as follows:

[0135]

[0136] Among them, is the estimated density value at point x.

[0137] (2) Determine the density threshold according to the kernel density of each feature vector.

[0138] Here, determining the density threshold according to the kernel density of each feature vector may specifically include:

[0139] Determine the median of the kernel densities of each feature vector as the density threshold ρ min . Among them, the calculation formula of the median of the kernel densities of each feature vector is as follows:

[0140]

[0141] Among them, represents the median of the kernel densities of each feature vector.

[0142] (3) Select a first target feature vector from all the currently unprocessed feature vectors.

[0143] (4) Determine whether the kernel density of the current first target feature vector is greater than or equal to the density threshold.

[0144] (5) If the kernel density of the current first target feature vector is less than the density threshold, determine that the current first target feature vector is not a core feature vector, and execute the step of "(3) Select a first target feature vector from all the currently unprocessed feature vectors".

[0145] (6) If the kernel density of the current first target feature vector is greater than or equal to the density threshold, the current first target feature vector is determined to be a core feature vector.

[0146] (7) From all the current unprocessed feature vectors, a second target feature vector that is in the same cluster as the current first target feature vector is selected to obtain a cluster of the current first target feature vector; wherein the cluster contains the current first target feature vector and its second target feature vector, and the feature vectors in the cluster are all marked as processed feature vectors.

[0147] Here, selecting a second target feature vector of the same cluster class as the current first target feature vector from among all current unprocessed feature vectors to obtain the cluster of the current first target feature vector may specifically include:

[0148] (7.1) Select a third target feature vector from all current unprocessed feature vectors.

[0149] (7.2) Calculate the Euclidean distance between the current third target feature vector and the current first target feature vector.

[0150] Specifically, any two eigenvectors x j and x i The Euclidean distance D(x i ,x j ) is calculated as follows:

[0151]

[0152] (7.3) According to the current Euclidean distance, determine whether the current third target feature vector and the current first target feature vector are directly density reachable, density reachable, and density connected.

[0153] Here, we first select the optimal density clustering parameter (the minimum radius ε of the neighborhood range of the feature vector). The minimum number of feature vectors MinPts of the feature vector is the density threshold ρ mentioned above. min , the minimum number of eigenvectors MinPts defines the minimum number of neighborhood eigenvectors of the core eigenvector, that is, a eigenvector needs to have at least MinPts eigenvectors (including itself) in its neighborhood to be considered a core eigenvector. Since the minimum number of eigenvectors MinPts has been determined, there is no need to optimize it, only ε needs to be optimized.

[0154] The process of determining the minimum radius ε of the neighborhood range of the feature vector is:

[0155] (a) Define the minimum radius ε.

[0156] The minimum radius ε defines the neighborhood of the eigenvector. i , define a neighborhood with radius ε:

[0157] N ε (x i )={x j ∈x|D(x i ,x j )≤ε}......(15)

[0158] (b) Select the optimal value of the minimum radius.

[0159] 1) Determine the minimum radius candidate value range.

[0160] Use the mean of the covariance eigenvalues ​​as a preliminary estimate of the minimum radius:

[0161]

[0162] 50% to 150% of the estimated value (i.e., the value calculated using Formula 16) are selected as candidate values ​​of the minimum radius, and the step size is set to 0.1 to form a candidate value set {ε1,ε2,…,ε z}.

[0163] 2) Traverse the candidate values ​​and evaluate the parameter effects.

[0164] The step size is set to 0.1, and all candidate parameter values ​​ε are traversed i , where ε i ∈{ε1,ε2,…,ε z}, use the Silhouette coefficient to evaluate the clustering effect of the parameters:

[0165]

[0166] Among them, a(i) is the average distance from feature vector i to other feature vectors in the same cluster, b(i) is the average distance from feature vector i to the nearest cluster, and C k is a cluster other than eigenvector i, |C k | is cluster C k The number of samples, d(i,j) is the feature vector i and cluster C k The distance between the feature vectors j in .

[0167] 3) Determine the optimal value of the minimum radius.

[0168] Select the ε with the highest Silhouette coefficient i As the best ε for this application.

[0169] In the embodiment of the present invention, for any two feature vectors x j and xi , if x j For x i is directly density-reachable, then both satisfy the following conditions:

[0170] ①x j In x i within the ε neighborhood of .

[0171] ②x i is a core eigenvector, that is, x i The ε neighborhood of contains at least MinPts feature vectors, including x i The formula can be expressed as:

[0172] {x i ∈ core point}∧{D(x i ,x j )≤ε}......(20)

[0173] The conditions for determining whether the density is reachable are:

[0174] If x i With x j Direct density can be reached, x j With x m Direct density is up to x m Not in x i In this case, x i and x m It is not directly density-reachable, but can be reached through x in its ε neighborhood. j Click to x m The point is directly density-reachable, then x i With x m The density can be reached.

[0175] The judgment conditions for density connection are:

[0176] If x m Cannot pass through core point x i Directly density-reachable and density-reachable, but within the ε neighborhood of its density-reachable point, then x i With x m Connected by density.

[0177] (7.4) If the current third target feature vector is directly density-reachable, density-reachable, and density-connected to the current first target feature vector, then the current third target feature vector is determined to be a second target feature vector of the current first target feature vector.

[0178] Next, the clustering label of the current third target feature vector can be set. Specifically, if the current third target feature vector is a second target feature vector of the current first target feature vector, the clustering label of the current third target feature vector can be set to the cluster number of the current first target feature vector (for example, 1); if the current third target feature vector is not a second target feature vector of the current first target feature vector, the clustering label of the current third target feature vector can be set to -1, indicating that it is a noise point that is not assigned to any cluster.

[0179] (7.5) The currently determined second target feature vector is classified into the cluster of the current first target feature vector.

[0180] (7.6) With the goal of marking each feature vector in the cluster of the current first target feature vector as a processed feature vector, feature vector marking is performed on the cluster of the current first target feature vector.

[0181] (7.7) Determine whether the search for the second target feature vector for the current first target feature vector is completed.

[0182] (7.8) If the search for the second target feature vector for the current first target feature vector has been completed, the second target feature vector search process is terminated.

[0183] (7.9) If the search for the second target feature vector for the current first target feature vector has not been completed, execute the aforementioned step of "(7.1) selecting a third target feature vector from all current unprocessed feature vectors".

[0184] (8) Execute the step of selecting a first target feature vector from all currently unprocessed feature vectors until all feature vectors are clustered.

[0185] In a specific example, assuming that all feature vectors are: X1, X2, X3, X4 and X5, the final clustering result may be that X1, X2 and X3 are classified into the same cluster, and X4 and X5 are classified into the same cluster.

[0186] Step 105: After obtaining the daily clustering results within the preset regulatory period, the collusive bidding behavior of each market player is identified based on all clustering results.

[0187] Specifically, at present, the power market in my country declares on a daily basis, and the units declare daily quantity and price curves. For each day in the preset supervision cycle, the method described in steps 101 to 104 above is used to obtain the clustering results of the day.

[0188] In the embodiment of the present invention, based on all clustering results, the collusive bidding behavior of each market player is identified, which may specifically include:

[0189] (1) For each clustering result, a target cluster is screened; all market entities corresponding to the target cluster have different equity relationships, and the excess returns of all market entities exceed the preset return threshold.

[0190] Here, the excess return of market players can be described by the clearing Lerner index L of market players. The clearing Lerner index is used to measure the clearing quotes P of each market player. 出清 Its marginal cost P 核定成本 The degree of deviation is calculated as follows:

[0191]

[0192] (2) When selecting the target cluster, for each target cluster, all target Euclidean distances corresponding to the target cluster are normalized, and the normalized value is determined as the single-day quotation similarity between the corresponding two feature vectors; wherein, all target Euclidean distances corresponding to the target cluster include the Euclidean distance between every two feature vectors of the target cluster.

[0193] (3) For each target feature vector pair, the average value of all single-day quotation similarities of the target feature vector pair is calculated to obtain the multi-day quotation similarity of the target feature vector pair; the target feature vector pair is the feature vector pair corresponding to the single-day quotation similarity.

[0194] (4) Determine whether the multi-day quotation similarity is greater than the target similarity; the target similarity is the average value of all single-day quotation similarities.

[0195] (5) If the multi-day quotation similarity is greater than the target similarity, the two market entities corresponding to the target feature vector pair are determined to be highly similar market entities.

[0196] The highly similar market entities determined by the present invention can be used to assist in the identification of collusion in bidding, that is, after obtaining highly similar market entities, first interview the entities with similar bids. Specifically, if it is determined that the bids are similar, the power market operator can ask the parties or related parties to explain by telephone and written inquiry. Then, indirect evidence is transferred. Specifically, if the power market operator determines that the market entity has similar bids and cannot give a reasonable explanation, the operator will transfer the clues of illegal and irregular activities reflected in the data reported in the on-site transaction to the energy regulatory agency for investigation and processing. Then, direct evidence investigation is carried out. Specifically, after the energy regulatory agency receives the clues of suspected abnormal trading behavior transferred by the power market operator, it shall conduct administrative investigations in accordance with the procedures prescribed by laws, regulations and rules, comprehensively, objectively and impartially review the evidence, and conduct a comprehensive and detailed investigation of the communication between similar entities in various information media such as telephone, text messages, letters, and Internet platforms.

[0197] In an embodiment of the present invention, in order to intuitively display the deviation level of quotations of each capacity point by different market players, after density clustering is performed on all feature vectors based on the target kernel function to obtain the clustering result of the target day, the Euclidean distance between every two feature vectors in each cluster can be further processed using the idea of ​​square average to obtain the quotation deviation value of each capacity point of the cluster.

[0198] In a specific example, for a cluster consisting of X1, X2, and X3, the bid deviation value d at the 0% capacity point is:

[0199]

[0200] Among them, D1 is the Euclidean distance between the quotation corresponding to the 0% capacity point in X1 and the quotation corresponding to the 0% capacity point in X2, D2 is the Euclidean distance between the quotation corresponding to the 0% capacity point in X1 and the quotation corresponding to the 0% capacity point in X3, D3 is the Euclidean distance between the quotation corresponding to the 0% capacity point in X2 and the quotation corresponding to the 0% capacity point in X3, and N is the number of Euclidean distances, where N=3.

[0201] It should be noted that the calculation process of the quotation deviation value of other capacity points of the cluster is similar and will not be repeated here.

[0202] In the embodiment of the present invention, before calculating the characteristic vector of each declared data, the method for identifying collusive bidding by market players in the electricity spot market of the present invention may further include:

[0203] The declared data is denoised to eliminate erroneous data that does not meet the unit's incremental declaration and data label requirements, thereby ensuring the quality of the data.

[0204] The present invention adopts the above technical solution to uniformly pre-process the extracted declaration data of the power spot market market players, and uniformly process the different declaration sections of different units into declaration feature vectors containing 0% to 100% output levels and quotations. This original data processing method can ensure that the declared capacity and quotations of market players of different capacity levels are at the same observable and comparable level, and improve the accuracy of quotation similarity recognition.

[0205] In addition, the density clustering identification model improved based on the kernel method can reduce the dependence on the density clustering parameters to a certain extent, and improve the accuracy and stability of the identification of the similarity of the data reported by market players. When identifying the similarity of the quotations of market players in the electricity spot market, this method innovatively uses the density clustering improved by the kernel method for analysis. First, the characteristic data of the quotations of market players are smoothed using kernel density estimation to capture the nonlinear distribution characteristics of the data. Then, by introducing a suitable kernel function, the quotation data is mapped to a high-dimensional space, thereby optimizing the clustering parameters and enhancing the processing capabilities of complex data structures. The optimal solution search mechanism of the kernel method is used to expand the data search range, prevent the clustering process from falling into the local optimum, and then construct an optimized kernel density clustering model to achieve the goal of more accurately identifying the similarity of market player quotations.

[0206] Based on a general inventive concept, the present invention also provides a device for identifying collusive bidding among market players in the electricity spot market.

[0207] Figure 2 This is a schematic diagram of the structure of a device for identifying collusive bidding by market players in the electricity spot market provided by an embodiment of the present invention. Figure 2 As shown, the device for identifying collusion in bidding by market players in the electricity spot market includes:

[0208] The acquisition module 21 is used to obtain the declared data of each market player on the target day; the target day is any day within the preset supervision period, and the declared data includes the capacity and quotation corresponding to different unit output stages.

[0209] The calculation module 22 is used to calculate the characteristic vector of each declared data; the characteristic vector includes the quotations at different percentage output levels of the corresponding unit, and each percentage output level of the unit corresponds to a capacity point.

[0210] The optimization module 23 is used to optimize the parameters of the preset kernel function according to all the feature vectors to obtain the target kernel function.

[0211] The clustering module 24 is used to perform density clustering on all feature vectors based on the target kernel function to obtain the clustering result of the target day.

[0212] The identification module 25 is used to identify the collusive bidding behavior of each market player based on all clustering results after obtaining the daily clustering results within a preset supervision period.

[0213] Optionally, the calculation module 22 is specifically used for:

[0214] (1) For each of the declared data, determine each capacity point of the declared data; the each capacity point is d capacity points that are evenly divided according to the capacity between the minimum technical output and the maximum technical output of the corresponding unit.

[0215] (2) Determine the quotation corresponding to each capacity point based on the declared data.

[0216] (3) Using the quotations corresponding to the capacity points, a feature vector of the declared data is obtained.

[0217] Optionally, the preset kernel function is a Gaussian kernel function.

[0218] The optimization module 23 is specifically used for:

[0219] Based on Silverman's method, the bandwidth parameter of the Gaussian kernel function is optimized according to all the eigenvectors to obtain the target kernel function.

[0220] Optionally, the clustering module 24 may specifically include:

[0221] A calculation unit is used to calculate the kernel density of each of the feature vectors based on the target kernel function.

[0222] The first determining unit is used to determine a density threshold according to the kernel density of each of the feature vectors.

[0223] The first selection unit is used to select a first target feature vector from all currently unprocessed feature vectors.

[0224] A judging unit is used to judge whether the kernel density of the current first target feature vector is greater than or equal to the density threshold.

[0225] The second determining unit is configured to determine that the current first target feature vector is not a core feature vector if the kernel density of the current first target feature vector is less than the density threshold.

[0226] The third determining unit is configured to determine that the current first target feature vector is a core feature vector if the kernel density of the current first target feature vector is greater than or equal to the density threshold.

[0227] The second selection unit is used to select a second target feature vector that is of the same cluster type as the current first target feature vector from all current unprocessed feature vectors, to obtain a cluster of the current first target feature vector; wherein the cluster contains the current first target feature vector and its second target feature vector, and the feature vectors in the cluster are all marked as processed feature vectors.

[0228] Optionally, the first determining unit is specifically configured to:

[0229] The median of the kernel density of each of the feature vectors is determined as the density threshold.

[0230] Optionally, the second selection unit is specifically used to:

[0231] (1) Select a third target feature vector from all currently unprocessed feature vectors.

[0232] (2) Calculating the Euclidean distance between the third target feature vector and the first target feature vector.

[0233] (3) According to the current Euclidean distance, determine whether the current third target feature vector is directly density-reachable, density-reachable, or density-connected to the current first target feature vector.

[0234] (4) If the third target feature vector is directly density-reachable, density-reachable, or density-connected to the first target feature vector, the third target feature vector is determined to be a second target feature vector of the first target feature vector.

[0235] (5) The currently determined second target feature vector is classified into the cluster of the currently determined first target feature vector.

[0236] (6) With the goal of marking each feature vector in the cluster of the current first target feature vector as a processed feature vector, feature vector marking is performed on the cluster of the current first target feature vector.

[0237] (7) Determine whether the search for the second target feature vector for the current first target feature vector is completed.

[0238] (8) If the search for the second target feature vector for the current first target feature vector has been completed, then the second target feature vector search process is terminated.

[0239] (9) If the search for the second target feature vector for the current first target feature vector has not been completed, a step of selecting a third target feature vector from all current unprocessed feature vectors is performed.

[0240] Optionally, the identification module 25 is specifically used for:

[0241] (1) For each of the clustering results, a target cluster is screened for the clustering results; all market entities corresponding to the target cluster have different equity relationships, and the excess returns of all market entities exceed a preset return threshold.

[0242] (2) When the target clusters are screened out, for each target cluster, all target Euclidean distances corresponding to the target cluster are normalized, and the normalized value is determined as the single-day quotation similarity between the corresponding two feature vectors; wherein all target Euclidean distances corresponding to the target cluster include the Euclidean distance between every two feature vectors of the target cluster.

[0243] (3) For each target feature vector pair, the average value of all the single-day quotation similarities of the target feature vector pair is calculated to obtain the multi-day quotation similarity of the target feature vector pair; the target feature vector pair is the feature vector pair corresponding to the single-day quotation similarity.

[0244] (4) Determine whether the multi-day quotation similarity is greater than a target similarity; the target similarity is the average value of all the single-day quotation similarities.

[0245] (5) If the multi-day quotation similarity is greater than the target similarity, the two market entities corresponding to the target feature vector pair are determined to be highly similar market entities.

[0246] Optionally, the device for identifying collusive bidding by market players in the electricity spot market of the present invention further includes:

[0247] The denoising module is used to denoise the declared data.

[0248] Based on a general inventive concept, the present invention also provides a computer device.

[0249] Figure 3 Schematic diagram of the structure of a computer device provided by an embodiment of the present invention. Figure 3 As shown, the computer device 300 includes:

[0250] at least one processor 310; and,

[0251] A memory 330 is communicatively connected to at least one processor 310; wherein,

[0252] The memory 330 stores instructions 320 that can be executed by at least one processor 310. The instructions 320 are executed by at least one processor 310 so that the at least one processor 310 can implement the method for identifying collusive bidding by market entities in the electricity spot market as described in the above embodiment.

[0253] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0254] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0255] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present invention belong.

[0256] 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.

[0257] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0258] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0259] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0260] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0261] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for identifying collusive bidding by market players in the electricity spot market. It is characterized in that include: Obtaining the declared data of each market player on the target day; the target day is any day within the preset regulatory period, and the declared data includes the capacity and quotation corresponding to different unit output stages; Calculating a characteristic vector of each of the declared data; the characteristic vector includes quotations at different percentage output levels of the corresponding unit, each percentage output level of the unit corresponding to a capacity point; According to all the eigenvectors, the preset kernel function is optimized to obtain a target kernel function; Based on the target kernel function, density clustering is performed on all the feature vectors to obtain a clustering result of the target day; After obtaining the daily clustering results within the preset regulatory period, the collusive bidding behavior of each market player is identified based on all the clustering results.

2. The method for identifying collusive bidding by market players in the electricity spot market according to claim 1, It is characterized in that The calculating of the characteristic vector of each of the declared data specifically includes: For each of the declared data, determining each capacity point of the declared data; each capacity point is d capacity points divided evenly according to the capacity between the minimum technical output and the maximum technical output of the corresponding unit; Determine the quotation corresponding to each capacity point according to the declared data; Using the quotations corresponding to the capacity points, a feature vector of the declared data is obtained.

3. The method for identifying collusive bidding by market players in the electricity spot market according to claim 1, It is characterized in that The preset kernel function is a Gaussian kernel function; The step of optimizing the parameters of the preset kernel function according to all the feature vectors to obtain the target kernel function specifically includes: Based on Silverman's method, the bandwidth parameter of the Gaussian kernel function is optimized according to all the eigenvectors to obtain the target kernel function.

4. The method for identifying collusive bidding by market players in the electricity spot market according to claim 1, It is characterized in that Based on the target kernel function, density clustering is performed on all the feature vectors to obtain the clustering result of the target day, which specifically includes: Based on the target kernel function, calculating the kernel density of each of the feature vectors; Determining a density threshold according to the kernel density of each of the eigenvectors; Select a first target feature vector from all currently unprocessed feature vectors; Determine whether the kernel density of the first target feature vector is greater than or equal to the density threshold; If the kernel density of the current first target feature vector is less than the density threshold, it is determined that the current first target feature vector is not a core feature vector, and a step of selecting a first target feature vector from all current unprocessed feature vectors is performed; If the kernel density of the current first target feature vector is greater than or equal to the density threshold, determining the current first target feature vector as a core feature vector; Among all the current unprocessed feature vectors, a second target feature vector of the same cluster class as the current first target feature vector is selected to obtain a cluster of the current first target feature vector; wherein the cluster contains the current first target feature vector and its second target feature vector, and the feature vectors in the cluster are all marked as processed feature vectors; The step of selecting a first target feature vector from all currently unprocessed feature vectors is performed until clustering is completed for all the feature vectors.

5. The method for identifying collusive bidding by market players in the electricity spot market according to claim 4, It is characterized in that Determining the density threshold according to the kernel density of each of the feature vectors specifically includes: The median of the kernel density of each of the feature vectors is determined as the density threshold.

6. The method for identifying collusive bidding by market players in the electricity spot market according to claim 4, It is characterized in that The selecting, from all current unprocessed feature vectors, a second target feature vector of the same cluster type as the current first target feature vector to obtain a cluster of the current first target feature vector specifically includes: Select a third target feature vector from all currently unprocessed feature vectors; Calculating the Euclidean distance between the third target feature vector and the first target feature vector; According to the current Euclidean distance, determining whether the current third target feature vector and the current first target feature vector are directly density-reachable, density-reachable, and density-connected; If the third target feature vector is directly density-reachable, density-reachable, and density-connected to the first target feature vector, then determining that the third target feature vector is a second target feature vector of the first target feature vector; Classifying the currently determined second target feature vector into the cluster of the currently determined first target feature vector; Marking the feature vectors of the cluster of the first target feature vector at present with the goal of marking each feature vector in the cluster of the first target feature vector at present as a processed feature vector; Determining whether the search for the second target feature vector for the current first target feature vector is completed; If the search for the second target feature vector for the current first target feature vector has been completed, then the second target feature vector search process is terminated; If the search for the second target feature vector for the current first target feature vector has not been completed, a step of selecting a third target feature vector from all current unprocessed feature vectors is performed.

7. The method for identifying collusive bidding by market players in the electricity spot market according to claim 6, It is characterized in that According to all the clustering results, the collusion bidding behavior of each market player is identified, specifically including: For each of the clustering results, a target cluster is screened for the clustering results; all market entities corresponding to the target cluster have different equity relationships, and the excess returns of all market entities exceed a preset return threshold; When the target clusters are screened out, for each target cluster, all target Euclidean distances corresponding to the target cluster are normalized, and the normalized value is determined as the single-day quotation similarity between the corresponding two feature vectors; wherein all target Euclidean distances corresponding to the target cluster include the Euclidean distance between every two feature vectors of the target cluster; For each target feature vector pair, the average value of all the single-day quotation similarities of the target feature vector pair is calculated to obtain the multi-day quotation similarity of the target feature vector pair; the target feature vector pair is the feature vector pair corresponding to the single-day quotation similarity; Determine whether the multi-day quotation similarity is greater than a target similarity; the target similarity is the average value of all the single-day quotation similarities; If the multi-day quotation similarity is greater than the target similarity, then the two market entities corresponding to the target feature vector pair are determined to be highly similar market entities.

8. The method for identifying collusive bidding by market players in the electricity spot market according to claim 1, It is characterized in that Before calculating the characteristic vector of each of the declared data, the method further includes: The declared data is subjected to denoising processing.

9. A device for identifying collusive bidding by market players in the electricity spot market. It is characterized in that include: An acquisition module is used to acquire the declared data of each market player on a target day; the target day is any day within a preset regulatory period, and the declared data includes the capacity and quotation corresponding to different unit output stages; A calculation module, used for calculating a characteristic vector of each of the declared data; the characteristic vector includes quotations at different percentage output levels of the corresponding unit, and each percentage output level of the unit corresponds to a capacity point; An optimization module, used to optimize the parameters of a preset kernel function according to all the feature vectors to obtain a target kernel function; A clustering module, used for performing density clustering on all the feature vectors based on the target kernel function to obtain a clustering result of the target day; The identification module is used to identify the collusive bidding behavior of each market entity based on all the clustering results after obtaining the daily clustering results within the preset supervision period.

10. A computer device, It is characterized in that include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to implement the method for identifying collusive bidding by market entities in the electricity spot market as described in any one of claims 1 to 8.

Images