A method and related device for identifying collusive behavior in the electricity market

Abstract

This application belongs to a clustering recognition method. Aiming at the technical problems existing in the existing power market collusion behavior recognition methods, such as insufficient reflection of the characteristics of collusion behavior, lack of the ability to deeply mine complex index characteristics, insufficient training samples, and insufficient abnormal collusion sample recognition ability, a power market collusion behavior recognition method and related devices are provided. First, obtain the data corresponding to the discrimination indicators of the power market collusion behavior; the discrimination indicators include electricity price indicators, electricity quantity indicators, and quantity-price relationship indicators. The electricity price indicators include the average value of the declared price fluctuations, and the quantity-price relationship indicators include the similarity of the bid curves. Then, input the data corresponding to the discrimination indicators into the collusion recognition model to obtain the collusion behavior recognition result; the input of the collusion recognition model is the interaction feature data between two market entities. It can analyze the interaction patterns between market entities in different clusters and does not need to rely on collusion label data.

Description

Technical Field

[0001] This application belongs to a clustering recognition method, and specifically relates to a method and related device for identifying collusive behaviors in the electricity market. Background Art

[0002] The identification of collusive behaviors in the electricity market mainly includes two steps: constructing discriminant indicators and identification models. Existing technologies have constructed discriminant indicators from different perspectives, but there are problems such as difficult acquisition and insufficient reflection of collusive behavior characteristics. The identification model mainly relies on subjective or subjective-objective combined evaluation methods and lacks the ability to deeply mine the characteristics of complex indicators. With the development of artificial intelligence technology, methods based on machine learning and deep learning have gradually been applied to the research on the identification of collusive behaviors in the electricity market, but there are still problems such as lack of training samples, susceptibility to label noise, and insufficient ability to identify abnormal collusive samples. Summary of the Invention

[0003] This application aims at the technical problems existing in the existing methods for identifying collusive behaviors in the electricity market, such as insufficient reflection of collusive behavior characteristics, lack of the ability to deeply mine the characteristics of complex indicators, insufficient training samples, and insufficient ability to identify abnormal collusive samples when using artificial intelligence technology, and provides a method and related device for identifying collusive behaviors in the electricity market.

[0004] To achieve the above object, this application adopts the following technical solutions:

[0005] In the first aspect, this application proposes a method for identifying collusive behaviors in the electricity market, including:

[0006] Obtain the data corresponding to the discriminant indicators of collusive behaviors in the electricity market; among them, the discriminant indicators include electricity price indicators, electricity quantity indicators, and electricity quantity-price relationship indicators, the electricity price indicators include the average value of the declared price fluctuation, and the electricity quantity-price relationship indicators include the similarity of the bid curves;

[0007] According to the data corresponding to the discriminant indicators, calculate the interaction characteristic data between two market entities and input it into the collusion recognition model to obtain the collusion behavior recognition result; among them, the collusion recognition model is based on the point-order search clustering structure recognition algorithm and adopts a method for identifying collusive behaviors in the electricity market based on the interaction and density clustering of two market entities.

[0008] In the second aspect, this application proposes a system for identifying collusive behaviors in the electricity market, including:

[0009] A data module, configured to obtain the data corresponding to the discriminant indicators of collusive behaviors in the electricity market; among them, the discriminant indicators include electricity price indicators, electricity quantity indicators, and electricity quantity-price relationship indicators, the electricity price indicators include the average value of the declared price fluctuation, and the electricity quantity-price relationship indicators include the similarity of the bid curves;

[0010] An identification module, configured to calculate interaction feature data between two market entities according to discriminant index corresponding data, and input the data into a collusion identification model to obtain a collusion behavior identification result; wherein, the collusion identification model is based on a point-order search clustering structure identification algorithm and adopts a power market collusion behavior identification method based on the interaction and density clustering of two market entities.

[0011] In a third aspect, the present application provides an electronic device, including: a memory and one or more processors; the memory is coupled to the processor; wherein, computer program code is stored in the memory, and the computer program code includes computer instructions, when the computer instructions are executed by the processor, the electronic device executes the steps of the above-mentioned power market collusion behavior identification method.

[0012] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned power market collusion behavior identification method are implemented.

[0013] Compared with the prior art, the present application has the following beneficial effects:

[0014] The present application provides a power market collusion behavior identification method, which obtains discriminant index corresponding data of power market collusion behavior. The electricity price index further includes the average value of declared price fluctuations, and the quantity-price relationship index further includes the similarity of bid curves. Then, the discriminant index corresponding data is input into the collusion identification model to obtain a collusion behavior identification result. Among them, the input of the collusion identification model is the interaction feature data between two market entities. The collusion identification model is based on a point-order search clustering structure identification algorithm and adopts a power market collusion behavior identification method based on the interaction and density clustering of two market entities. The present application innovatively proposes two new indicators, namely the average value of declared price fluctuations and the similarity of bid curves. Considering that it is difficult to obtain collusion labels, the present application constructs a power market collusion behavior identification method based on the interaction and density clustering of two market entities. This method does not directly predict whether a single market entity has collusion suspicion, but starts from the perspective of mining the interaction behavior characteristics of power market entities. It can not only effectively identify collusion sample pairs, but also analyze the interaction patterns between market entities in different clusters. The identification method of the present application can accurately identify power market collusion behavior without relying on collusion label data, which is beneficial to ensuring the safe and stable operation of the power market.

[0015] The present application also provides a power market collusion behavior identification system, an electronic device and a computer-readable storage medium, which have all the advantages of the above-mentioned power market collusion behavior identification method. Description of the Drawings

[0016] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a schematic diagram of a method for identifying collusive behavior in the electricity market of the present application;

[0018] Figure 2 It is a schematic diagram of the ratio of the area difference of the bid curves in the embodiments of the present application;

[0019] Figure 3 It is a schematic diagram of the similarity of the bid curves in the embodiments of the present application;

[0020] Figure 4 It is a schematic diagram of an example of constructing paired features in the embodiments of the present application;

[0021] Figure 5 It is a schematic diagram of the core concept of the clustering algorithm in the embodiments of the present application;

[0022] Figure 6 Distribution diagram of the declared electricity price and declared electricity quantity of all segments of the experimental data in the embodiments of the present application;

[0023] Figure 7 It is a reachability distance graph obtained by using the clustering algorithm in the embodiments of the present application;

[0024] Figure 8 For the embodiments of the present application Schematic diagram of the clustering result at this time;

[0025] Figure 9 It is a schematic diagram of a system for identifying collusive behavior in the electricity market of the present application. Detailed implementation manners

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Usually, the components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0027] Accordingly, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but merely represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the scope of protection of the present application.

[0028] It should be noted that like reference numerals and letters denote like items in the following drawings, and thus, once an item is defined in one drawing, it does not require further definition and explanation in subsequent drawings.

[0029] The method for identifying collusive behaviors in the electricity market mainly consists of two core steps: First, construct discriminant indicators for collusive behaviors in the electricity market to quantify relevant influencing factors; Second, based on the data corresponding to these discriminant indicators, construct an identification model for collusive behaviors to reveal potential collusive behaviors.

[0030] The prior art has constructed discriminant indicators from different perspectives. For example, select market power risk warning indicators based on three dimensions of market structure, market behavior, and market performance. Or, construct an evaluation index system for identifying illegal behaviors covering the whole process before, during, and after transactions. And construct monitoring indicators from two major dimensions of the overall market situation and the behaviors of market entities. The indicator systems constructed by such research are relatively systematic and comprehensive. However, these studies mainly start from the perspective of individual market entity members and fail to fully reflect the interaction relationships among market entities. Therefore, it is difficult to capture the characteristics of collusive behaviors. In addition, due to factors such as confidentiality, it is difficult to obtain relevant indicator data (such as marginal cost, economic profit, etc.), which limits the feasibility of such indicator systems in actual application scenarios. The quotation data, as the core data for market entities to participate in electricity transactions, has relatively convenient acquisition channels. By mining the relationship of quotation data among multiple market entities, it is possible to reveal whether there are collusive behaviors during the bidding process. Therefore, based on parallel pricing behaviors, the prior art has proposed an index of similarity of declaration information or introduced an index of the ratio of the difference area of quotation curves to extract the quotation connection between any two market entities. Although certain progress has been made in the design of the indicator system based on quotation data, there are still certain deficiencies in reflecting the relationship of quotation data, such as failing to fully reflect the fluctuation relationship and the quantity-price relationship of quotations.

[0031] In view of the above-established discrimination index system, in the prior art, early warning threshold values are set for each discrimination index through the expert evaluation method, and a comprehensive index is constructed in combination with the weighted linear model to warn of market power risks. There are also methods that use the fuzzy comprehensive evaluation method to calculate the early warning index scores of market entities under different illegal trading behaviors. There are also three-stage evaluations of the market power behavior of electricity selling companies using the grey comprehensive evaluation method for pre-event prevention, in-event monitoring, and post-event analysis. The recognition models established in the above research mainly rely on subjective or subjective-objective combined evaluation methods for judgment, lacking in-depth exploration of the characteristics of complex indicators, which will affect the objectivity and accuracy of recognition.

[0032] With the continuous progress of artificial intelligence technology, compared with statistical methods, methods based on machine learning and deep learning have significant advantages in processing complex feature data and are gradually applied to the research on the identification of collusive behaviors in the electricity market. The identification of collusive behaviors essentially belongs to a binary classification problem. Some studies use supervised classification algorithms such as the Naive Bayes classifier, Isolation Forest, ensemble learning, and support vector machine to predict the probability of abnormal behaviors. However, due to the lack of sample data with collusive labels, which are difficult to obtain, and there may also be collusive behaviors that have not been identified by humans, the methods based on supervised classification algorithms lack training samples, resulting in poor training and identification effects. There are also proposed identification models based on semi-supervised support vector machines. First, the support vector machine model is trained using labeled samples, and the unlabeled samples are labeled. Then, the labeled samples are combined with the original samples and trained again to optimize the model. There are also methods that use the K-means clustering algorithm (an unsupervised learning algorithm) to perform preliminary pseudo-label assignment on unlabeled samples and further use cost-sensitive transductive support vector machines for identification. Such semi-supervised methods are easily affected by label noise, resulting in worse identification results. In addition, there is also an unsupervised variational autoencoder Gaussian mixture model proposed by combining the idea of anomaly detection, which can maximize the retention of original data information but has insufficient recognition ability for abnormal collusive samples. These methods are all difficult to reveal the interactive bidding behaviors among market entity members.

[0033] In view of the above situation, this application proposes a method and related device for identifying collusive behaviors in the electricity market. The following will make a detailed description of this application in combination with embodiments and drawings.

[0034] As Figure 1 shown, it is a schematic flowchart of a method for identifying collusive behaviors in the electricity market of this application, which may include:

[0035] S101, obtaining the corresponding data of the discrimination indexes for collusive behaviors in the electricity market; among them, the discrimination indexes include electricity price indexes, electricity quantity indexes, and electricity quantity-electricity price relationship indexes. The electricity price indexes include the average value of the declared price fluctuations, and the electricity quantity-electricity price relationship indexes include the similarity of the bidding curves.

[0036] This application constructs a discriminant index system from three dimensions: electricity price, electricity quantity, and the relationship between quantity and price, covering the direct manifestations and potential correlations of market behaviors. Among them, the electricity price index includes indicators such as the average value of the declared price fluctuation, which is used to capture the abnormal stability or coordination of the quotations of market players. The electricity quantity index can involve the deviation between the declared electricity quantity and the cleared electricity quantity, the concentration of electricity quantity allocation, etc., reflecting the abuse of market power or the manipulation of supply and demand. The quantity-price relationship index can measure the consistency of the quotation strategies of different market players through the similarity of the quotation curves, revealing potential coordination behaviors. The acquisition of the data corresponding to these discriminant indicators can be obtained through market declaration data (such as quotation curves, declared electricity quantities), transaction clearing results, historical transaction records, etc. In practical applications, the data corresponding to the obtained discriminant indicators can also be preprocessed, for example, standardized processing, missing value filling, outlier detection, etc.

[0037] S102. Calculate the interaction feature data between two market players based on the data corresponding to the discriminant indicators, and input it into the collusion recognition model to obtain the collusion behavior recognition result; among them, the collusion recognition model is based on the point sequence search clustering structure recognition algorithm and adopts a method for recognizing electricity market collusion behaviors based on the interaction and density clustering of two market players.

[0038] It should be noted that the interaction feature data between two market players can include the quotation synchronization coefficient, strategy response delay, joint market power index, etc. Input the interaction feature data into the collusion recognition model and use a method for recognizing electricity market collusion behaviors based on the interaction and density clustering of two market players for recognition. Among them, the principle of the method for recognizing electricity market collusion behaviors based on the interaction and density clustering of two market players is that in the high-dimensional feature space, a candidate clustering center sequence is generated through local density search. Points with a density higher than the neighborhood threshold can be preferentially selected as the initial centers to avoid the sensitivity of random initialization. Then, based on the density reachability of the core points, the clustering is extended to automatically discover clusters of any shape. The output decision of the collusion recognition model can be to output the confidence level of the collusion behavior (usually between 0 and 1) based on density clustering and pattern matching degree. The threshold setting can combine the regulatory tolerance and Bayesian decision theory to determine the optimal classification threshold.

[0039] The following further details the method for recognizing electricity market collusion behaviors of this application through some specific embodiments:

[0040] 1. Regarding the improved discriminant index system for collusion behaviors.

[0041] Collusive behavior in the electricity market refers to the illegal act of two or more market entity members raising prices through means such as collusive bidding, thereby maximizing the interests of the collusive market entities. When the electricity market shows a supply surplus, the collusive market entity members often obtain profits through parallel bidding. In the stage of relatively balanced supply and demand, the collusive market entity members create artificial local supply and demand tensions through abnormal or extreme bidding, inducing abnormal fluctuations in the market clearing price. This strategic bidding behavior is mainly reflected in the bidding data, so it is necessary to design a discriminant index system to reflect the interactive bidding behavior.

[0042] In the existing technology, the discriminant index system designed for collusive behavior usually has two strongly correlated indicators, namely the average value of the declared price safety degree and the average value of the declared price relative ratio, which may lead to a high degree of dependence of the recognition model on such features and affect the accuracy of the recognition method. Based on the three-stage bidding method, this application constructs a collusive discriminant index system around the bidding and quantity data, optimizes the bidding relative ratio index on the basis of the existing index system, and adds two new indicators: the average value of the declared price fluctuation and the similarity of the bidding curve.

[0043] As shown in Table 1, taking two market entities as the recognition objects, the constructed discriminant index system covers three dimensions: electricity price, electricity quantity, and the relationship between quantity and price.

[0044] Table 1 Collusive Behavior Discriminant Index System

[0045]

[0046] In some embodiments of this application, the meanings and calculation methods of each discriminant index are as follows:

[0047] Set to represent the declared price, to represent the declared electricity quantity, to represent the bidding stage, to represent the number of market entities.

[0048] (1) Declaration price consistency.

[0049] The declaration price consistency index measures the degree of correlation between the bids of two market entities. The lower the corresponding data value of this index, the more consistent the bids of the two market entities are, and the closer the adjustment ranges in each bidding stage are. Market entity and market entity 's declaration price consistency The calculation formula is:

[0050] .

[0051] Among them, is the market entity In the segment quotation, the declared price is the declared price of the market entity in the segment quotation, and is the average value of the declared prices of all market entities in the segment quotation.

[0052] The average value of the declared prices of all market entities in the segment quotation is calculated as:

[0053] .

[0054] Among them, is the segment quotation, and the market entity declares the price.

[0055] (2)Average value of declared price fluctuations.

[0056] In market collusion behavior, there may be a performance of first reporting a low price to ensure market share and then reporting a high price to obtain higher profits. In response to this behavior performance, this application innovatively proposes an average value index of declared price fluctuations to reflect the fluctuation situation between each segment of quotations of two market entities. The higher the corresponding data value of this discrimination index, the greater the volatility of the quotations of the two market entities, indicating a higher degree of speculation. The average value of declared price fluctuations of market entity and market entity is calculated as:

[0057] .

[0058] Among them, is the weighted average quotation of market entity , is the weighted average quotation of market entity , and the weight value is the declared electricity quantity. The weighted average quotation of market entity is calculated as:

[0059] .

[0060] Among them, is the electricity quantity declared by market entity in the segment quotation.

[0061] (3)Average value of declared price safety.

[0062] The average value of the declared price security is used to evaluate the deviation degree between the average quotes of two market entities and the historical marginal price. The higher the corresponding data value of this discriminant index, the greater the possibility that these two market entities jointly quote higher prices. The average value of the declared price security is calculated as follows:

[0063] .

[0064] Wherein, represents the expected value of the historical market marginal price.

[0065] (4)Relative comparison of declared prices.

[0066] The relative comparison of declared prices is used to measure the relative high or low of two enterprises in price declaration. The closer the average quotes of two market entities are, the closer the corresponding data of the relative comparison of declared prices is to 1, suggesting that there may be a collusive behavior of jointly pushing up prices. Market entity and market entity The relative comparison of declared prices is calculated as follows:

[0067] .

[0068] (5)Consistency of declared electricity quantities.

[0069] The consistency of declared electricity quantities is a measure of the correlation of the declared electricity quantities of two market entities. The lower the value of this index, the closer the declared electricity quantities of the two market entities are, and the more consistent the changes in their differences from the market average quote are. Market entity and market entity The consistency of declared electricity quantities is calculated as follows:

[0070] .

[0071] Wherein, is the electricity quantity declared by market entity in the th segment of quotes, is the electricity quantity declared by market entity in the th segment of quotes, is the average value of the electricity quantities declared by all market entities in the th segment of quotes.

[0072] The average value of the electricity quantities declared by all market entities in the th segment of quotes is calculated as follows:

[0073] .

[0074] Among them, is the electricity quantity declared by the market entity in the segment quotation.

[0075] (6) Ratio of the difference area of the quotation curves.

[0076] The ratio of the difference area of the quotation curves refers to the ratio of the difference area formed by the three-segment quotation curves of two market entities. As Figure 2 shown, it is a schematic diagram of the ratio of the difference area of the quotation curves. This indicator can reveal the differences in the shape and distribution of the quotation curves. The lower the ratio of the difference area of the quotation curves, the more similar the quantity-price relationship of the quotations of the two market entities, thus suggesting a greater possibility of collusion between them. The market entity and the market entity The ratio of the difference area of the quotation curves is calculated as follows:

[0077] .

[0078] Among them, is the quotation curve function of the market entity , is the quotation curve function of the market entity , is the minimum total power generation of the market entity and the market entity , is calculated as follows:

[0079] .

[0080] (7) Similarity of the quotation curves.

[0081] The calculation boundary of the ratio of the difference area of the quotation curves is set as the minimum power generation, which cannot fully reveal the differences in the quantity-price of the third segment. Considering that the quotation curve is a curve composed of three pairs of declared quantities and declared prices , among which, is the declared quantity, is the declared price. This application innovatively introduces a discriminant index for the similarity of the quotation curves. As Figure 3 shown, it is a schematic diagram of the similarity of the quotation curves. By calculating the Euclidean distance between points on the quotation curve, the position differences of points on the quotation curve can be quantified. The similarity of the quotation curves of the market entity and the market entity is calculated as follows:

[0082] .

[0083] Among them, is the electricity quantity declared by the market entity in the th paragraph of the quotation, is the electricity quantity declared by the market entity in the th paragraph of the quotation.

[0084] 2. Collusion identification algorithm (collusion identification model) based on the point sequence search clustering structure identification algorithm.

[0085] (1) Feature construction method.

[0086] Collusive behavior in the electricity market usually involves complex interactions between two or more market entities. If only the characteristics of a single market entity are concerned, it will be difficult to comprehensively reveal the intricate interaction relationships between market entities. Therefore, this application uses the interaction feature data between two market entities as the input of the collusion identification algorithm, aiming to more accurately capture and analyze the behavior patterns between market entities. Specifically, assume there are market entities, and the corresponding eigenvalue is calculated according to the seven key interaction indicators in the collusion discrimination index system constructed in this application. By calculating the corresponding data values of the discrimination indicators between every two market entities, samples can be generated. Each sample depicts the interaction characteristics between a pair of market entities, so as to more accurately reflect the behavior relationship and potential collusive behavior characteristics between market entities. As Figure 4 shown, it is a schematic diagram of the construction example of paired interaction features.

[0087] (2) Principle of the point sequence search clustering structure identification algorithm

[0088] Clustering algorithms belong to unsupervised machine learning algorithms and are mainly divided into types such as partitioning clustering, hierarchical clustering, and density clustering. In the identification of collusive behaviors in the electricity market, the use of the clustering structure identification algorithm based on point sequence search has advantages compared to other clustering methods. First, the characteristics of collusive behaviors in the electricity market are complex and changeable, and the clusters formed in the feature space often exhibit irregular shapes. The density-based clustering method can break through the limitations of traditional partitioning clustering algorithms. By deeply analyzing the density connectivity of data points, it can accurately identify clusters of any shape, thus more effectively capturing the potential patterns and feature associations of collusive behaviors. Second, collusive behaviors in the electricity market are essentially abnormal behaviors, and the density-based clustering method has a strong ability to identify outliers. It can accurately identify outliers as low-density regions and exclude them from normal clusters, thereby ensuring the reliability of the clustering results. In addition, other clustering methods may have problems such as low computational efficiency and insufficient result stability when processing data. In contrast, the density-based clustering method is relatively more optimal in terms of computational efficiency and is not affected by the order of data input, and can provide more stable and reliable clustering results. This application uses an algorithm based on the interaction and density clustering of two market entities, which is improved on the basis of the traditional density clustering algorithm, reducing the impact of the selection of input parameters on the clustering results.

[0089] Let the sample set be , and each sample contains seven discriminant index features in the collusive discrimination index system:

[0090] .

[0091] Among them, is the th sample, and to are the first to seventh discriminant index features in the discriminant index system of the th sample.

[0092] Given the neighborhood radius and the minimum number of data points for clustering, the samples in the sample set are arranged in an orderly manner so that similar samples are adjacent in the orderly arrangement. A decision graph can be generated according to the orderly arrangement, and then the clustering results based on any neighborhood radius can be obtained from the decision graph.

[0093] The above algorithm involves three core concepts, as Figure 5 shown, which is a schematic diagram of the above core concepts:

[0094] 1) Core object: For a certain sample , if the circle within its neighborhood radius contains at least If there are

[0095] samples, then the sample is defined as a core object. The mathematical expression of the core object is:

[0096] where is the sample within the number of samples contained within the circle with

[0097] 2) Core distance: It is the minimum neighborhood radius required for the sample to become a core object. Let be the sample the -th nearest neighbor node within the neighborhood radius of

[0098] .

[0099] where is the core distance of the sample , is undefined, is otherwise, represents and the Euclidean distance between.

[0100] 3) Reachability distance: It represents the distance from the sample to the sample that is a core object, reflecting the clustering of samples. The mathematical formula of the reachability distance is expressed as:

[0101] .

[0102] where is and the reachability distance of, is the core distance of, is and the Euclidean distance between.

[0103] It should be noted that Figure 5 in

[0104] rd represents the reachability distance, co represents the core object, and cd represents the core distance.

[0105] 1) Initialization.

[0106] Initialize three empty queues: the core object queue and the queue of unprocessed neighbor points and an ordered queue (i.e., the final output queue) .

[0107] 2) Identify core objects.

[0108] Traverse all samples in the sample set. For each sample, calculate the number of samples within its neighborhood radius . Mark all core objects that meet the requirements as unprocessed and put them into the core object queue , and calculate the core distance.

[0109] 3) Process core objects.

[0110] Check whether the core object queue is empty. If it is empty, the algorithm ends. Otherwise, take an unprocessed core object from , put it into the ordered queue , and mark it as processed. Calculate the reachability distance between all unprocessed samples within the neighborhood of the core object and the core object, and put them into the unprocessed neighbor point queue in ascending order of the reachability distance.

[0111] 4) Process neighbor points.

[0112] Take an unprocessed sample with the smallest reachability distance from the unprocessed neighbor point queue , put into the ordered queue , and mark it as processed. If is a core object, then add all unprocessed samples within the neighborhood of to the unprocessed neighbor point queue , and calculate the reachability distance of each sample in the unprocessed neighbor point queue compared to . If there is a smaller reachability distance, update it. Repeat step 4) until the unprocessed neighbor point queue is empty, then return to step 3).

[0113] 3. Collusion behavior identification.

[0114] Based on the improved collusion discrimination index system of this application and the above algorithm, this application constructs a method for identifying power market collusion behavior based on the interaction and density clustering of two home field entities. The specific steps are as follows:

[0115] (1) Set the input data as the three-section bid volume and price quotation data sets of market entities, where​ , , is the first-stage quotation for the market entity . is the first-stage quantity declaration for the market entity . is the second-stage quotation for the market entity . is the second-stage quantity declaration for the market entity . is the third-stage quotation for the market entity . is the third-stage quantity declaration for the market entity .

[0116] (2) Calculate the interaction feature data between any two market entities to obtain a sample set , where the number of samples .

[0117] (3) Normalize the interaction feature data to convert it into a standard normal distribution with zero mean and unit variance. Use the Z-score normalization method, that is, normalize by subtracting the mean of each interaction feature data and dividing by its standard deviation.

[0118] (4) Analyze the normal distribution of the interaction feature data, including its central tendency, dispersion degree, and skewness.

[0119] (5) Given , , use a clustering algorithm based on the point-order search clustering structure recognition algorithm to cluster the sample set to generate a reachable distance graph.

[0120] (6) According to the reachable distance graph, select an appropriate , and analyze the clustering results. Determine the pairs of members within the cluster identified as the abnormal class as suspected of collusion, and export the pairs of market entity members involved to form a list of suspected collusion.

[0121] (7) Start the collusion behavior recognition program to monitor each pair of members in the list of suspected collusion. The monitoring content includes: detecting whether two market entities without an actual control relationship have the same trading declaration MAC (Media Access Control) address and IP (Internet Protocol) address; detecting whether two market entities without an actual control relationship use the same trading account for declaration; detecting whether there is data interaction on the information-based trading platforms of two market entities without an actual control relationship. According to the further monitoring results, execute the corresponding market control procedures.

[0122] To verify the effectiveness of the method proposed in this application, the data in the monthly centralized bidding transaction data table of a certain province is used as experimental data, which includes three-section bidding data of 25 electricity selling companies (market entities). Some of the original data is shown in Table 2. Among them, the collusion possibilities of the 1st and 10th electricity selling companies, and the 2nd, 3rd, and 5th electricity selling companies are the highest, and they are regarded as collusion pairs and are abnormal samples. It should be noted that in Table 2, the "id" column represents the serial number of the market entity.

[0123] Table 2 Partial transaction data table

[0124]

[0125] As Figure 6 shown, it is the distribution diagram of the declared prices and declared volumes of all segments of the experimental data. It can be seen that the declared prices are mainly distributed around 400 yuan / (MW·h) -1 , while the declared volumes are concentrated around 5000 MW. There are individual declared prices lower than 390 yuan / (MW·h) -1 , and individual declared volumes are higher than 150000 MW.

[0126] Based on the improved collusion discrimination index system, the characteristics of the collusion behavior discrimination indexes of 300 samples are calculated. Table 3 shows the characteristics of the collusion behavior discrimination indexes of some samples. Among them, the first four samples belong to the collusion sample pairs. By comparing and analyzing each discrimination index, it can be found that the values of the first four samples on discrimination index 1 (declared price consistency) and index 4 (relative ratio of quotes) are significantly lower than those of other samples, indicating that the collusion sample pairs have discrimination index characteristics different from those of normal samples and may have a significant distance from normal samples in the sample space. Therefore, through the method of this application, these collusion sample pairs can be accurately identified as outliers, thus effectively distinguishing their differences from normal samples. It should be noted that index 1 in Table 3 is the declared price consistency, index 2 is the average value of the declared price fluctuations, index 3 is the average value of the declared price safety degree, index 4 is the relative ratio of the declared prices, index 5 is the declared electricity volume consistency, index 6 is the ratio of the difference area of the quote curves, and index 7 is the similarity of the quote curves.

[0127] Table 3 Characteristics of some discrimination samples

[0128]

[0129] The reachable distance graph obtained using the point-order search clustering structure recognition algorithm is as Figure 7 shown, and the reachable distance graph records the reachable distances of each sample.

[0130] The silhouette coefficient and the DB (Davies - Bouldin) index are used to evaluate the clustering structure recognition algorithm based on point - order search in this application. Among them, the silhouette coefficient is used to quantify the closeness of a sample to other samples within the same cluster and the separation from the nearest neighbor cluster. Its value range is [-1, 1]. The larger the value, the better the clustering effect, indicating that the samples within the cluster are close and the separation between clusters is obvious. The DB index measures the clustering quality by calculating the ratio of the within - cluster compactness to the between - cluster separation. The smaller the value of this index, the more ideal the clustering result, indicating that the data points within the cluster are closely aggregated and the differentiation between clusters is high.

[0131] To verify the rationality of the selected , the clustering performance under different settings was tested. As shown in Table 4, with the increase of , the silhouette coefficient gradually increases, while the DB index generally shows a trend of first decreasing and then increasing. Therefore, combining the peak - valley changes of the reachability graph and the performance of the evaluation indicators, this application selects as the parameter input for the final clustering.

[0132] Table 4 Clustering algorithm performance under different settings

[0133]

[0134] For the convenience of visualization, this application can also use the principal component analysis dimensionality reduction method to process the samples for dimensionality reduction. As Figure 8 shown, it presents the clustering results when . In the figure, cluster - 1 (purple points) represents outliers, and clusters 0 (blue), 1 (green), and 2 (yellow points) represent normal clusters.

[0135] The number of samples contained in each cluster and the average volume - price similarity within the cluster are shown in Table 5. Through analysis, it can be obtained that in cluster - 1 where the outliers are located, the quote similarity of paired market - entity members is at the lowest level, and for cluster 0 containing most samples, its similarity is the second. This shows that although different market entities may show a certain degree of similarity in their quotes due to key factors such as costs that affect volume and price, such similarity is not sufficient to classify them as abnormal clusters. The similarity value of cluster 1 is relatively large, indicating that there are obvious differences in the quote behaviors of paired entity members within the cluster. The volume - quote similarity value of cluster 2 is relatively large, indicating that there are relatively large differences in the volume quotes of paired entity members within the cluster. Further analysis shows that the collusive sample pairs (1, 10), (2, 5), (3, 5) are correctly assigned to cluster - 1, and the collusive sample pair (2, 3) is misassigned to cluster 0. Although there are 4 non - collusive samples in cluster - 1, this mis - clustering phenomenon can be corrected through subsequent monitoring procedures.

[0136] The above experimental analysis shows that the present application can identify most of the collusive samples. At the same time, through the method proposed in the present application, the interaction behavior characteristics between market entities in different clusters can be deeply explored, so as to discover the illegal interaction transactions existing among potential electricity market members.

[0137] Table 5 Sample situation within each cluster

[0138]

[0139] To verify the effectiveness of the method proposed in the present application, the performance of various clustering methods was compared, including the KMeans method (K-means method) based on partition clustering, the method of first performing PCA (Principal Component Analysis) dimensionality reduction and then KMeans clustering, and the method of first performing PCA dimensionality reduction and then conventional clustering. The silhouette coefficient and the DB index were used to evaluate different methods, and the results are shown in Table 6. Since the KMeans clustering method requires the number of clusters to be specified manually, for consistency, the number of clusters was set to 4. In addition, since the reachability distance graph changes after PCA dimensionality reduction, the method adopted is indicated in parentheses after the conventional clustering method. The experimental results show that the method adopted in the present application has a higher silhouette coefficient and a lower DB index compared with other methods, that is, the data points within the clusters are closer and the separation degree of the data points between the clusters is larger, and the clustering effect is the best.

[0140] Table 6 Comparison of clustering effects of different methods

[0141]

[0142] Compared with the supervised algorithms mentioned in the prior art, the method proposed in the present application does not need to rely on labeled data and can directly automatically mine potential interaction patterns and characteristics from the unlabeled market entity quotation data. This method does not rely on a clear objective function to guide the learning process, thus having stronger flexibility and adaptability. It enables it to closely combine with the operating characteristics of the electricity market and has the potential to identify collusive patterns that have not been clearly defined.

[0143] The clustering method based on the characteristics of independent entities mainly focuses on the clustering analysis of the quotation characteristics of a single entity. However, in the electricity market, different enterprises often generate similar quotations due to cost similarity, which may lead to misclassifying a single entity involved in collusion as a normal sample, thus reducing the accuracy of identifying collusion behavior. In contrast, the clustering method based on the interaction characteristics of entities starts from the perspective of the interaction behavior characteristics between entities and conducts clustering analysis on the interaction behavior of entity pairs. This method can capture the potential associations and abnormal interaction patterns between entities more comprehensively and deeply, thus more effectively achieving the accurate identification of collusion behavior.

[0144] Aiming at the problems that the existing methods do not fully explore the characteristics of the quotation data of market entity interactions and the scarcity of collusion samples in the electricity market, this application improves the existing discrimination index system for electricity market collusion behavior and introduces three new key indicators around the quotation data, namely, the relative quotation ratio, the average volatility of declared prices, and the similarity of quotation curves. Based on the constructed discrimination index system, this application constructs a new collusion identification model, which can effectively identify abnormal collusion behavior clusters by dividing regions with different densities in the sample space into different clusters. The resulting list of suspected collusion can be provided to electricity market monitoring personnel for the next step of program identification, which helps to prevent and resolve market collusion risks.

[0145] As Figure 9 shown, a schematic diagram of an electricity market collusion behavior identification system may include:

[0146] A data module for obtaining the data corresponding to the discrimination indicators of electricity market collusion behavior; among them, the discrimination indicators include electricity price indicators, electricity quantity indicators, and quantity-price relationship indicators. The electricity price indicators include the average volatility of declared prices, and the quantity-price relationship indicators include the similarity of quotation curves;

[0147] An identification module for calculating the interaction characteristic data between two market entities based on the data corresponding to the discrimination indicators and inputting it into the collusion identification model to obtain the collusion behavior identification result; among them, the collusion identification model is based on the point-order search clustering structure recognition algorithm and adopts a method for identifying electricity market collusion behavior based on the interaction and density clustering of two market entities.

[0148] It should be noted that in the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each module is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The modules described as separate components may or may not be physically separated. The components shown as modules can be one physical unit or multiple physical units, that is, they can be located in one place or distributed to multiple different places. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0149] In addition, in each embodiment of the present invention, the modules can be integrated in a processing unit, or each module can exist physically alone, or two or more modules can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0150] The embodiments of this application also provide an electronic device, which may include one or more processors, a memory, and a communication interface.

[0151] Among them, the memory and the communication interface are coupled to the processor. For example, the memory and the communication interface can be coupled together through a bus.

[0152] Among them, the communication interface is used for data transmission with other devices. The memory stores computer program code. The computer program code includes computer instructions. When the computer instructions are executed by the processor, the electronic device executes the steps of the above method for identifying collusive behavior in the electricity market.

[0153] Among them, the processor can be a processor or a controller. For example, it can be a Central Processing Unit (CPU), a general-purpose processor, a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the present disclosure. The processor can also be a combination that realizes computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and so on. The processor can be used to support the electronic device in executing the method steps provided in the above embodiments.

[0154] Among them, the bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The above bus can be divided into an address bus, a data bus, a control bus, etc.

[0155] A computer-readable storage medium provided by an embodiment of the present application stores a computer program, and when the computer program is executed by a processor, the steps of the above method for identifying collusive behavior in the electricity market are realized.

[0156] The computer-readable storage medium involved in the present application includes a Random Access Memory (RAM), an internal memory, a Read-Only Memory (ROM), an Electrically Programmable ROM, an Electrically Erasable Programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium well-known in the technical field.

[0157] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A method for identifying collusive behavior in the electricity market, characterized in that, Including: Obtain the corresponding data of the discrimination indicators for collusive behaviors in the electricity market; among them, the discrimination indicators include electricity price indicators, electricity quantity indicators, and electricity quantity - price relationship indicators. The electricity price indicators include the average value of the declared price fluctuation, and the electricity quantity - price relationship indicators include the similarity of the bid curves. The calculation formula for the average value of the declared price fluctuation is: Among them, is the average price fluctuation of the declared prices of market entities and market entities ; is the price declared by the market entity in the th paragraph of the quotation, is the weighted average quotation of the market entity ; is the price declared by the market entity in the th paragraph of the quotation, is the weighted average quotation of the market entity ; The calculation formula for the similarity of the bid curves is: Among them, is the similarity of the quotation curves of market entities and market entities ; is the electricity quantity declared by market entities in the section of quotations, is the electricity quantity declared by market entities in the section of quotations; Based on the corresponding data of the discrimination indicators, calculate the interaction characteristic data between two market entities, input it into the collusion recognition model, and obtain the collusion behavior recognition result; among them, the collusion recognition model is based on the point - order search clustering structure recognition algorithm, and adopts a method for recognizing collusive behaviors in the electricity market based on the interaction and density clustering of two market entities: Normalize the interaction characteristic data between two market entities respectively, and convert it into a standard normal distribution with zero mean and unit variance; analyze the standard normal distribution of the interaction characteristic data between two market entities one by one to obtain the normal distribution analysis result; combine the normal distribution analysis result, give the neighborhood of the core object in the point - order search clustering structure recognition algorithm, and the number of samples included in the circle within the neighborhood radius. Cluster the sample set including the interaction characteristic data between multiple pairs of market entities through the point - order search clustering structure recognition algorithm to generate a reachable distance graph; according to the reachable distance graph, select the neighborhood of the core object and analyze the clustering result to obtain the recognition result.

2. The method for identifying collusive behavior in the electricity market according to claim 1, wherein The electricity price indicators also include the declared price consistency, the average value of the declared price security, and the declared price relative ratio. The electricity quantity indicators include the declared electricity quantity consistency. The electricity quantity - price relationship indicators also include the ratio of the difference area of the bid curves.

3. The method for identifying collusive behavior in the electricity market according to claim 2, wherein The calculation formula for the declared price consistency is: Among them, is the consistency of the declared prices of market entities and market entities ; it is the average value of the declared prices of all market entities in the -th paragraph of quotations, where is the price declared by market entity in the -th paragraph of quotations, and is the price declared by market entity in the The calculation formula for the average value of the declared price security is: Among them, is the average safety degree of the declared prices of market entities and market entities is the average value of the declared price safety degree of market entities, is the expected value of the historical market marginal price, is the weighted average quotation of market entities is the weighted average quotation of market entities, is the weighted average quotation of market entities is the weighted average quotation; The calculation formula for the declared price relative ratio is: Among them, is compared with the declared prices of market entities and market entities ; The calculation formula for the declared electricity quantity consistency includes: Among them, is the consistency of the declared electricity quantities of market entities and market entities ; is the average value of the declared electricity quantities of all market entities in the quotation in the th segment, is the electricity quantity declared by market entity in the quotation in the th segment, is the electricity quantity declared by market entity in the quotation in the th segment.

4. The method for identifying collusive behavior in the electricity market according to claim 2, wherein The calculation formula for the ratio of the difference area of the bid curves includes: Among them, is the ratio of the difference area between the and quotation curves of market entities, is the minimum total power generation of the and market entities, is the quotation curve function of the market entity, is the quotation curve function of the market entity, is the power generation.

5. A collusion behavior recognition system for the electricity market, characterized in that, Including: A data module for obtaining the corresponding data of the discrimination indicators for collusive behaviors in the electricity market; among them, the discrimination indicators include electricity price indicators, electricity quantity indicators, and electricity quantity - price relationship indicators. The electricity price indicators include the average value of the declared price fluctuation, and the electricity quantity - price relationship indicators include the similarity of the bid curves. The calculation formula for the average value of the declared price fluctuation is: Among them, is the average price fluctuation of the market entities and the market entities ; is the price declared by the market entity in the section bid; is the weighted average bid of the market entity ; is the price declared by the market entity in the section bid; is the weighted average bid of the market entity ; The calculation formula for the similarity of the bid curves is: Among them, is the similarity of the quotation curves of market entities and market entities ; is the electricity quantity declared by market entity in the section of quotations, and is the electricity quantity declared by market entity in the section of quotations. An identification module for calculating the interaction characteristic data between two market entities based on the corresponding data of the discrimination indicators, inputting it into the collusion recognition model, and obtaining the collusion behavior recognition result; among them, the collusion recognition model is based on the point - order search clustering structure recognition algorithm, and adopts a method for recognizing collusive behaviors in the electricity market based on the interaction and density clustering of two market entities: Normalize the interaction feature data between two market entities respectively and convert it into a standard normal distribution with zero mean and unit variance; analyze the standard normal distribution of the interaction feature data between two market entities one by one to obtain the normal distribution analysis result; combine the normal distribution analysis result, give the core object neighborhood in the point-order search clustering structure recognition algorithm, and the number of samples included in the circle within the neighborhood radius, and cluster the sample set including the interaction feature data between multiple two market entities through the point-order search clustering structure recognition algorithm to generate a reachable distance map; according to the reachable distance map, select the core object neighborhood and analyze the clustering result to obtain the recognition result.

6. An electronic device, characterized in that, Including: A memory, one or more processors; the memory is coupled to the processor; wherein, computer program code is stored in the memory, the computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the steps of the power market collusion behavior recognition method according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the power market collusion behavior recognition method according to any one of claims 1-4 are implemented.

Images