Power transaction security risk management method based on big data analysis

Abstract

The invention relates to the technical field of power transaction security risk management, and discloses a power transaction security risk management method based on big data analysis. The method comprises the following steps: establishing a transaction subject and charging period set, processing abnormal and missing data, constructing a symmetric adjacency matrix, sequentially calculating a degree matrix and a normalized Laplacian matrix, forming an initial risk vector in combination with subject historical quotation information, adaptively setting a diffusion step length and a convergence threshold by using a maximum degree value, and calculating the risk of the transaction subject according to the diffusion step length and the convergence threshold. Stable risk distribution is obtained through graph structure risk diffusion iteration, and then recognition and clustering of high-risk nodes are achieved through Laplacian spectral decomposition and spectral space clustering. Through the whole process of data acquisition, network modeling, risk quantification and dynamic diffusion to high-risk clustering, the problems of risk identification fragmentation and staticization are solved, the data accuracy and the risk tracking capability are improved, accurate hierarchical management of high-risk nodes is realized, and the manual intervention and the misjudgment rate are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power transaction security risk management, in particular to a power transaction security risk management method based on big data analysis. BACKGROUND

[0002] With the continuous deepening of the power market and the continuous improvement of the transaction mechanism, the participants, business types and billing periods of power transactions are increasingly diversified, and the transaction scale and complexity are significantly improved. In the existing power transaction security risk management system, static credit rating, expert experience rule, preset threshold alarm, local statistical analysis and other means are usually used to identify and manage the subject risk. The above methods achieve preliminary risk assessment to some extent, but are difficult to adapt to the complex power market environment of large scale, multi-time sequence, multi-subject and multi-interaction.

[0003] The prior art mostly analyzes the historical default situation, credit score or individual anomaly of the transaction subject from the perspective of a single node, lacks systematic modeling of the complex interaction relationship between subjects, and ignores the risk transmission, aggregation and diffusion phenomenon formed between subjects through the transaction relationship network. This leads to the fact that in a large-scale power transaction network, local risks are easily underestimated, and high-risk subject clusters or risk channels are difficult to identify in a timely manner. Secondly, the current common method relies on artificial rules or traditional missing value filling algorithms in the data processing stage, and in the face of frequent outliers, missing values and data noise in power transaction big data, it is easy to make mistakes, omissions and other problems. At the same time, the time sequence structure and dynamics of the data are often not fully utilized, which restricts the real-time and accuracy of risk identification. In addition, the existing technology usually relies on subjective parameters, empirical rules or global uniformity assumptions in key steps such as risk threshold setting and clustering boundary division, and fails to adaptively adjust to the actual network structure, data distribution and risk heterogeneity. This not only reduces the scientificity and objectivity of the risk management results, but also makes the model lack robustness and generalization ability when facing different market scales, structures and operating states. In terms of risk aggregation identification, traditional methods based on a single risk indicator or clustering algorithm are difficult to deal with complex data structures in high-dimensional spectral space, and lack the ability to deeply mine and quantify the spatial aggregation phenomenon of high-risk subjects in the network structure. The current risk clustering and high-risk cluster identification methods are mostly based on empirical parameters or fixed cluster numbers, and cannot dynamically adjust the clustering strategy according to the actual operating state of the network, nor can they combine risk distribution and network structure information for joint clustering and filtering. The actual application of current graph structure analysis and spectral methods is often limited to simple networks or static graphs, and lacks deep integration with the high-dimensional, heterogeneous and dynamic characteristics of power transaction big data, resulting in insufficient adaptability, efficiency and interpretability of traditional spectral clustering, feature decomposition and other methods in the actual power market environment. At the same time, many technical links have problems such as non-unique variables, non-transparent calculation path, imperfect boundary condition processing, etc., which further limit the practicality and scalability of the system.

[0004] Therefore, the present application aims to provide a power transaction security risk management method based on big data analysis, which combines big data and complex network theory through data preprocessing, graph structure modeling, risk quantification, diffusion simulation, spectral analysis and intelligent clustering, and realizes the accurate screening and clustering management of potential high-risk subjects in the power market. SUMMARY

[0005] The present application provides a power transaction security risk management method based on big data analysis, which solves the problems mentioned in the background art.

[0006] The present application provides the following technical scheme: a power transaction security risk management method based on big data analysis, comprising: A set of transaction subjects and charging periods is established, all transaction volume data of all subjects in each period is collected and sorted, and a complete symmetric adjacency matrix is constructed by processing abnormal values and missing data; Based on the adjacency matrix, the degree value of each node is calculated, and the degree matrix and the normalized Laplacian matrix are constructed accordingly; Combined with the historical quotation information of the transaction subjects, the average quotation and volatility of each subject are calculated, the risk deviation of each node is obtained based on the standardization method, and the initial risk vector is formed; According to the maximum value of the degree of each node, the step size and convergence threshold of risk diffusion iteration are set; The risk diffusion iteration algorithm on the graph structure is adopted to continuously update the risk vector of each node until the preset convergence standard or the maximum iteration number is reached, and the stable risk distribution result is obtained; The risk distribution result is subjected to graph Laplacian spectral decomposition to obtain the eigenvalue and eigenvector of the normalized Laplacian matrix, and the spectral features related to high-risk aggregation are extracted; Based on the risk distribution and spectral analysis results, a risk identification threshold is set to screen high-risk nodes and form an index set of high-risk subjects; The high-risk nodes are subjected to clustering analysis in the spectral space, different risk clusters are divided, and the identification results of each high-risk cluster are output.

[0007] Optionally, the set of transaction subjects and charging periods is established, all transaction volume data of all subjects in each period is collected and sorted, and a complete symmetric adjacency matrix is constructed by processing abnormal values and missing data, specifically including: Reading power transaction records, including A set of transaction subjects and A set of charging periods, respectively constructing a set of all transaction subjects in the power market And a set of historical transaction periods : , ; wherein, is the th transaction subject; is the transaction subject index; is the period index; Reading the original transaction volume ; wherein, is the power transaction volume from subject to subject in period ; If is missing, then ; Obtaining the maximum transaction volume of all subjects in all periods : ; Abnormal truncation: if , let ; Calculate the total electricity volume of transactions from subject to subject in all time periods , ; Calculate the total transaction volume between subject and subject , ; Get the maximum value in the symmetric edge weight ; Construct a symmetric adjacency matrix , where is the set of all matrices composed of row column real numbers; is the normalized adjacency strength from node to , specifically: .

[0008] Optionally, based on the adjacency matrix, the degree value of each node is calculated, and a degree matrix and a normalized Laplace matrix are constructed accordingly, specifically including: Calculate the degree of the node in the normalized adjacency matrix ; Construct the diagonal matrix of ; the diagonal element is ; Construct the normalized Laplace matrix .

[0009] Optionally, the historical pricing information of the transaction subjects is combined to calculate the average pricing and volatility of each subject, and the risk deviation of each node is obtained based on the standardization method to form an initial risk vector, specifically including: Get the electricity price of subject in time period , denoted as ; Calculate the historical average pricing of the node ; Calculate the standard deviation of the pricing sequence of the node ; Perform deviation standardization, specifically: ​ ; wherein, is a node the standard deviation of the latest period price to the historical mean; constructing an initial risk vector: ; wherein is an initial risk vector with dimension , the th component is .

[0010] Optionally, the maximum value of the degree of each node is used to set the step size and convergence threshold of the risk diffusion iteration, specifically including: obtain the maximum value of all node degrees ; set the diffusion iteration step size ; set the convergence threshold ; wherein, is a norm operation.

[0011] Optionally, the risk diffusion iteration algorithm on the graph structure is used to continuously update the risk vector of each node until the preset convergence criterion or the maximum iteration number is reached, and a stable risk distribution result is obtained, specifically including: set the maximum iteration number to ; perform on : ; wherein, is the risk vector of the th iteration; is the iteration number; if , let and immediately stop iteration; wherein, is the number of iterations that converge or reach the upper limit; if and not converged, let and stop iteration; get the final converged risk distribution vector .

[0012] Optionally, the risk distribution result is subjected to graph Laplacian spectral decomposition to obtain the eigenvalues and eigenvectors of the normalized Laplacian matrix, and the spectral features related to high-risk aggregation are extracted, specifically including: eigenvalue solution: , get ; wherein, is an eigenvalue; is an identity matrix; is the th column of the identity matrix. Small eigenvalues; is the eigenvalue index; For each , solve the equation , and order ;in, For The corresponding unit eigenvector; Extract the vector corresponding to the second smallest eigenvalue The vector corresponding to the third smallest eigenvalue .

[0013] Optionally, the risk identification threshold is set based on the risk distribution and spectrum analysis results, high-risk nodes are screened, and an index set of high-risk entities is formed, specifically including: Calculate the average risk of all nodes ; Calculate the standard deviation of the equilibrium risk vector ; Set a high-risk identification threshold ; Build an index set of all high-risk nodes ; like , then terminate; if ,make , and terminate; wherein, For collection The number of elements in ; is the set of high-risk cluster nodes in the spectrum space.

[0014] Optionally, performing cluster analysis on high-risk nodes in the spectral space, dividing them into different risk clusters, and outputting identification results of each high-risk cluster specifically includes: For each , get the The coordinates of the nodes in the two-dimensional spectral embedding space : ;in, is the eigenvector No. Quantity; is the eigenvector No. Quantity; Get the geometric center of the high-risk cluster in the spectral space ; Calculate the The Euclidean distance from the high-risk node to the center ; Get the median of a distance set : ;in, Sort in ascending order elements; Median function; Construct the final identified high-risk node cluster .

[0015] The present invention has the following beneficial effects: 1. Through rigorous anomaly detection, missing value correction, and normalization, multidimensional time-series power trading data is converted into a symmetric adjacency matrix that represents the true interactive relationships between market participants. Unlike traditional one-way data recording or crude weighting based solely on total trading volume, this approach comprehensively considers the intensity of two-way trading and scientifically truncates extreme outliers, making the network model more representative and resilient to interference. This process enhances the structural accuracy of the network model and the reliability of subsequent analysis. Compared with existing technologies, it resolves practical issues such as data inconsistency and arbitrary network construction, providing a high-quality data foundation for the entire process.

[0016] 2. Drawing heavily on complex network analysis theory, this algorithm structures adjacency relationships into a degree matrix and a normalized Laplace matrix, accurately reflecting the global connectivity characteristics of each node and the overall network connectivity. This algorithm not only provides solid mathematical support for subsequent advanced analyses such as risk diffusion and spectral decomposition, but also enables risk management to move beyond traditional approaches that rely on single-point thresholds and static indicators, enabling a deeper characterization of complex dynamic relationships at the network level. Compared to existing risk management models that focus on isolated entities, this approach improves the ability to identify systemic risks and collective anomalies.

[0017] 3. This system integrates multi-dimensional indicators based on historical quote averages and volatility, unifying risk characteristics across different time periods and entities into a single evaluation system through standardized methods. This approach effectively prevents misleading risk assessments caused by single price anomalies or short-term fluctuations, ensuring that initial risk estimates consider both individual historical performance and deviations from overall market levels. Compared to traditional static or empirical threshold methods, this approach enhances the scientific nature and differentiation of risk quantification, effectively supporting subsequent dynamic risk transmission and iterative optimization.

[0018] 4. The step size and convergence threshold of the risk diffusion model are adaptively set based on structural characteristics such as the maximum node degree, ensuring both stability and efficiency in the diffusion process and avoiding misjudgments caused by subjective parameter settings. This strategy improves the algorithm's adaptability and robustness across networks of varying sizes and structures. Compared to existing fixed step sizes or manual empirical settings, it not only simplifies model parameter adjustment but also improves the consistency of results and the feasibility of practical applications.

[0019] 5. The risk diffusion iterative algorithm based on graph structure is adopted to fully simulate the real flow path of risk in the complex transaction network, realize the dynamic transmission of risk in multiple subjects and the whole chain. This method breaks through the limitations of single-point and single-period static evaluation, and can effectively capture the potential accumulation and concentration trend of risk in the whole market system. The multi-round iteration and adaptive convergence mechanism make the final result more stable and objective. Compared with the traditional static scoring or linear threshold method, the perception ability of risk aggregation and chain reaction is improved.

[0020] 6. The graph theory is introduced into the power transaction risk management, and the dimension reduction mapping of the network structure and the feature extraction of the risk group are realized through the spectral decomposition of the Laplacian operator. This technology can accurately reveal the implicit grouping structure and local aggregation effect in the network, and provides data support for the subsequent spatial positioning and clustering of high-risk nodes. Compared with the conventional plane clustering or threshold-based grouping method, the spectral decomposition can mine the implicit connection and potential risk diffusion path, and is an important tool for discovering invisible systemic risk.

[0021] 7. The adaptive threshold method is adopted to scientifically select high-risk nodes by combining the mean and standard deviation of risk distribution, spectral clustering characteristics and other multi-dimensional information, so as to prevent missed judgment and misjudgment. This algorithm not only realizes the dynamic balance of individual risk and group risk, but also effectively avoids the subjectivity caused by artificial setting standards. Compared with the existing methods of selecting only a single indicator or using a global one-size-fits-all method, the sensitivity and pertinence of risk identification are improved.

[0022] 8. The adaptive division of different risk clusters is realized by using the geometric distribution of high-risk nodes in the spectral space. Through the combination of spatial distance and median adaptive threshold, the problems of presetting cluster number and strong subjectivity of boundary in traditional clustering are effectively solved, and the automatic identification and hierarchical management of high-risk groups are realized. Compared with the traditional method, the changes of power transaction network structure and risk pattern can be dynamically responded, the scientificity and forward-looking of risk governance are improved, and an efficient tool is provided for accurate supervision and strategy formulation. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The flowchart of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0025] Embodiment, refer to Figure 1, a power transaction security risk management method based on big data analysis, including: Establish a set of transaction entities and billing periods, collect and organize the transaction volume data of all entities in each period, and construct a complete symmetric adjacency matrix by processing outliers and missing data; Based on the adjacency matrix, the degree value of each node is calculated, and the degree matrix and normalized Laplace matrix are constructed accordingly; Combined with the historical quotation information of trading entities, the average quotation and volatility of each entity are calculated, and the risk deviation of each node is obtained based on the standardized method to form the initial risk vector; According to the maximum value of each node degree, the step size and convergence threshold of risk diffusion iteration are set; Adopting the risk diffusion iterative algorithm on the graph structure, the risk vector of each node is continuously updated iteratively until the preset convergence standard or the maximum number of iterations is reached, thus obtaining a stable risk distribution result. Perform Laplace spectral decomposition on the risk distribution results to obtain the eigenvalues ​​and eigenvectors of the normalized Laplace matrix, and extract spectral features related to high-risk clustering; Based on the risk distribution and spectrum analysis results, risk identification thresholds are set to screen high-risk nodes and form an index set of high-risk entities; Perform cluster analysis on high-risk nodes in the spectral space, divide them into different risk clusters, and output the identification results of each high-risk cluster.

[0026] By integrating the eight core links of the entire power trading process, from data collection, network construction, risk quantification, dynamic diffusion to spectral decomposition and clustering, this method solves the problems of fragmentation, staticness, and excessive manual intervention in risk identification in traditional power trading. In terms of steps, this method first establishes a complete set of trading entities and time periods and eliminates data anomalies and missing data to ensure the accuracy of basic data; then, through graph structure construction and matrix operations, trading behavior is converted into an analyzable network model; then, based on historical quotes, risks are initialized and the propagation path of risks in the network is dynamically simulated with the help of an adaptive diffusion iterative algorithm; finally, Laplace spectral decomposition and spectral space clustering are combined to achieve accurate identification and cluster management of high-risk entities. This series of steps are closely linked to form a closed-loop risk management process, which completely breaks through the bottleneck of existing technologies that have insufficient grasp of network correlation and difficulty in tracking risk propagation mechanisms. It improves the accuracy and real-time nature of risk identification, enabling regulators and operators to promptly identify potential problems and formulate targeted strategies. By using spectral clustering technology to manage high-risk nodes in layers, it improves the sophistication of risk prevention and control, reduces the workload and error rate of manual review, and provides strong decision-making support for the safe operation of the power market.

[0027] The process of establishing a set of transaction entities and billing periods, collecting and collating transaction volume data of all entities in each period, and constructing a complete symmetric adjacency matrix by processing outliers and missing data specifically includes: Read electricity transaction records, including Transaction entities and billing period, respectively construct the collection of all trading entities in the electricity market Collection with historical trading sessions : , ;in, For the a trading entity; Index of the transaction subject; Indexing time periods; clarifying the transaction entities and time ranges involved in the system, providing a basic framework for subsequent data indexing; Reading raw transaction volume ;in, For the period within, subject To the subject The actual transaction volume of each subject pair in each period is obtained as the raw data of network weighting; like If missing, then ; Ensure the matrix is ​​complete to avoid null value errors in subsequent operations; Get the maximum trading volume of all principal pairs in all periods : ;Determine the cutoff threshold to prevent extreme anomalies from interfering with the analysis; Abnormal truncation: If , then let ; Unify extreme values ​​beyond the reasonable range to the maximum observed value to reduce noise; Calculation from the main body To the subject Total transaction volume during all periods , ;Merge the time series transaction volume into a single weighted value and construct the original directed network; Computing Subject With the subject Total two-way trading volume , ;Convert the directed network into an undirected one to capture the bidirectional transaction intensity; Get the maximum value among the symmetric edge weights ; Constructing a symmetric adjacency matrix ,in, For by Row All matrix sets composed of real numbers in column; For the normalized adjacency strength of node To , specifically: ; all edge weight standards are normalized to the range of , eliminating dimensional effects, and constructing the final adjacency matrix .

[0028] Through systematic processing of original power transaction records, including key steps such as subject and time period set construction, missing value filling, outlier truncation, bidirectional transaction volume aggregation and normalization processing, the problems of excessive noise, data missing and extreme value interference in large-scale power transaction data are solved. In practice, in view of the transaction volume difference generated by different subjects and different billing periods, first, the original transaction volume data is completed by the regularization process, and the abnormal values exceeding the reasonable range are uniformly truncated to avoid the distortion of the network model by extreme data; then the directed transaction is aggregated into undirected bidirectional total volume, and normalized based on the maximum value of the whole network, so as to form a symmetric adjacency matrix that truly reflects the transaction strength and eliminates the dimensional effect. The reliability and stability of network construction in the subsequent analysis link are improved, so that the risk diffusion simulation and spectral clustering analysis based on the matrix can be carried out under the guarantee of high-quality data; at the same time, through unified standardization processing, the comparability of data of different scales and different time periods is ensured, which provides a solid and reproducible data foundation for the whole risk management system, and fundamentally reduces the possibility of misjudgment, overestimation or omission.

[0029] Based on the adjacency matrix, the degree value of each node is calculated, and the degree matrix and the normalized Laplacian matrix are constructed accordingly, specifically including: The degree of the node in the normalized adjacency matrix is calculated ; measure the overall connection strength of node in the normalized network, used for Laplace construction; Construct the diagonal matrix of ; the diagonal element is ; store the degrees of all nodes in the diagonal form to facilitate matrix operation; Construct the normalized Laplacian matrix ; generate the Laplace operator describing the network structure, which is used for simulating risk diffusion in the future.

[0030] The normalized adjacency matrix is calculated by the node degree value, and the diagonal matrix and the normalized Laplace matrix are constructed based on the degree value, which solves the problem of insufficient description of the overall structure characteristics of the power transaction network. In the implementation process, the overall connection strength of each node in the standardized network is quantified as the degree value, which is used to generate the diagonal matrix, and the normalized Laplace matrix is constructed in combination with the adjacency information, providing a mathematical operator for network analysis that can reflect both local connection and global structure. The introduction of the Laplace matrix not only accurately describes the connection relationship between the power transaction subjects, but also provides a natural propagation operator for the subsequent risk diffusion algorithm, ensuring that the risk transmission in the network conforms to the real transaction relationship; in addition, this construction method can reveal the potential hierarchical structure and community characteristics of the network through spectral decomposition, making the risk management more systematic and interpretable, making up for the group effect that cannot be captured by simple analysis of degree value or edge weight, and helping to achieve more scientific overall risk insight.

[0031] The historical quotation information of the combined transaction subject is calculated, the average quotation and volatility of each subject are calculated, the risk deviation of each node is obtained based on the standardization method, and an initial risk vector is formed, specifically including: Obtaining the subject In the time period The electricity price is denoted as ; the quotation of each subject in each time period is obtained, which lays the foundation for risk measurement; Calculate the historical average quotation of the th node; Calculate the standard deviation of the quotation sequence of the th node; Quantify the central tendency and volatility of each node price, which is used for deviation standardization; Perform deviation standardization, specifically: ; wherein is the standardized deviation of the latest time period price of node from the historical mean; the deviation degree of the latest time period price is converted into a dimensionless risk indicator to ensure stability; Construct the initial risk vector: ; wherein is the initial risk vector, the dimension is , and the th component is ; integrate the initial risk values of all nodes to provide a starting point for subsequent diffusion iteration.

[0032] By deep mining of the historical quotation information of the transaction subject, including the quantification of historical average quotation and price volatility, and generating a dimensionless risk deviation index based on the standardization method, the problems of traditional single index being difficult to balance mean and volatility, and initial risk value being difficult to compare are solved. First, the price data of each subject in different billing periods is collected, the historical quotation center trend and volatility degree are calculated, and then the deviation of the latest period price from the historical mean is standardized to form a unified risk deviation value, and finally the initial risk vector is integrated to provide an accurate risk starting point for each subject. The horizontal comparison of different transaction subjects in different historical performance dimensions is realized, avoiding the risk assessment distortion caused by different data magnitudes; at the same time, combined with the price volatility information, the abnormal behavior can be captured earlier, and the sensitivity of initial risk calibration is improved; and through standardization, the risk values of each subject are comparable, providing a fair and reliable baseline for subsequent risk diffusion and dynamic updating, laying a solid foundation for overall risk early warning and control.

[0033] The maximum value of the degree of each node is used to set the step size and convergence threshold of risk diffusion iteration, specifically including: Obtain the maximum value of all node degrees Obtain the maximum value of all node degrees Set the diffusion iteration step size Determine the proportion of risk diffusion in each iteration to ensure numerical stable convergence Set the convergence threshold Where, is the norm operation; set the convergence criterion, when the iteration change is lower than the threshold, it is considered to be balanced.

[0034] By setting the risk diffusion iteration step size and convergence threshold based on the maximum degree value of the nodes in the network, the problem of fixed parameters being difficult to balance multiple network structures and scales, and the iteration process being unstable or excessively oscillating is solved. First, the maximum connection strength of the whole network is extracted from the degree matrix, and the appropriate proportion of each risk diffusion iteration is dynamically calculated according to the maximum value, and further the convergence threshold is set when the iteration change amplitude is lower than a certain standard, so as to avoid the slow convergence caused by setting too small step size, and avoid the oscillation and non-convergence caused by setting too large step size. The adaptive ability of the algorithm is improved, which can realize fast and stable risk diffusion simulation in different scale power transaction networks without manual repeated parameter adjustment; at the same time, through the automatic convergence judgment mechanism, the calculation resources are saved, and the analysis time is shortened, which is helpful to provide real-time or near real-time risk assessment results in actual operation, and ensures the efficient response of the risk early warning system.

[0035] The risk diffusion iterative algorithm based on the graph structure is used to continuously iterate and update the risk vector of each node until a preset convergence standard or a maximum number of iterations is reached to obtain a stable risk distribution result. Specifically, the algorithm includes: Let the maximum number of iterations be ; Prevent infinite loop in the case of extreme non-convergence; if If the value is too small, the algorithm may be forced to end before it actually converges, resulting in large deviations in the results. If the value is too large, computing resources may be wasted, and meaningless iterations may be performed even after convergence. and computing resource settings, general experience As an initial attempt. Combined with the convergence threshold , if in front Reached multiple times , can be terminated early. For dozens to hundreds of nodes, it is advisable to ; Thousands of nodes can be adjusted to the resource .

[0036] right implement: ;in, For the Risk vector for round iteration; Number the iterations; based on the network structure and current risk distribution Conduct the next round of risk transfer; like , then let And stop the iteration immediately; is the number of iterations required to converge or reach the upper limit; like If it does not converge, let and stop iterating; Decide when to stop the simulation based on a threshold or upper iteration limit; The final converged risk distribution vector is ; Output the final risk distribution for high-risk node screening.

[0037] The simulation process solves the problem that static assessment cannot reflect the risk propagation path and cumulative effect by updating the risk vector on the graph structure network for multiple rounds of iteration until adaptive convergence or reaching the maximum number of iterations. Relying on the constructed network structure and the initial risk vector, the simulation is performed according to the established diffusion ratio for multiple iterations, and the overall risk change amplitude is evaluated after each iteration to determine whether to continue iteration or terminate early, ensuring that the simulation can fully capture the layer-by-layer transmission of risk in the network and will not fall into an infinite loop. The simulation results can dynamically reflect the risk transmission and reinforcement effect caused by the interaction between different subjects, providing a more realistic and complete risk distribution map for the identification of high-risk subjects. At the same time, the iteration termination judgment mechanism effectively balances the simulation depth and computational efficiency, making the risk distribution results accurate and timely, which helps regulatory agencies and operators respond to potential risks and adjust prevention and control strategies more quickly.

[0038] The risk distribution results are subjected to graph Laplacian spectral decomposition to obtain the eigenvalues and eigenvectors of the normalized Laplacian matrix, and the spectral features related to high-risk aggregation are extracted, specifically including: Eigenvalue solving: Solving the spectral properties of the network Laplacian reveals the network structure pattern; obtaining Arranging the eigenvalues in order of size facilitates the selection of corresponding eigenvectors; wherein, is the eigenvalue; is the identity matrix; is the small eigenvalue; is the eigenvalue index; For each , solve the equation and let ; wherein, is the unit eigenvector corresponding to ; obtain the unique unit eigenvector corresponding to each eigenvalue; Extract the second smallest eigenvalue corresponding vector and the third smallest eigenvalue corresponding vector ; use the low-order eigenvectors to realize the two-dimensional dimension reduction representation of the nodes.

[0039] By spectral decomposition of the normalized Laplacian matrix and extracting low-order eigenvectors, the problem that traditional grouping methods based on plane or threshold are difficult to reveal the hidden community structure of the network is solved. On the basis of risk distribution results, the eigenvalues and corresponding spectral vectors of the network operator are calculated, and the vector space corresponding to the lowest few eigenvalues is particularly concerned, so as to realize the dimensionality reduction mapping of the original high-dimensional network and provide a natural coordinate system for subsequent clustering analysis. The potential "community" or "group" relationship between power trading subjects is accurately mined, so that the spatial distribution and internal association of high-risk groups are more clear; at the same time, the vector representation obtained by spectral decomposition can fully reflect the structural characteristics of the local and global network, and provide a more explanatory feature basis for group risk identification, making up for the defects of simple risk value sorting or threshold screening in capturing group effect.

[0040] Based on the risk distribution and spectral analysis results, the risk identification threshold is set, the high-risk nodes are screened, and the index set of high-risk subjects is formed, specifically including: Calculate the average value of the balance risk of all nodes ; Calculate the standard deviation of the balance risk vector ; Respectively measure the center and dispersion degree of the risk distribution to provide a basis for threshold setting; Set the high-risk identification threshold ; Determine which nodes have significantly higher risk than the average level; Construct the index set of all high-risk nodes ; Screen out the node index of all risks exceeding the threshold; If , terminate; if , let , and terminate; wherein, The number of elements in the set ; The high-risk cluster node set in the spectral space; handle the extreme number of cases to avoid clustering errors.

[0041] By dynamically setting a high-risk identification threshold based on the mean and dispersion of the risk distribution, and screening out entities that exceed the average level, the problem of missed or misjudgment of high-risk entities caused by fixed thresholds or manual divisions is solved. According to statistical principles, the central trend and dispersion of the overall risk level are first calculated, and then the high-risk judgment criteria are adaptively determined. Finally, an index set of high-risk entities is generated to ensure that only entities with truly abnormal risk performance are included in the scope of attention. The objectivity and robustness of risk screening are improved, and the subjective bias caused by artificially set standards is avoided. At the same time, through dynamic threshold adjustment, the screening mechanism can respond quickly and maintain accuracy when the market environment or trading pattern changes. The screening results are more targeted, providing reliable target objects for subsequent in-depth analysis and precise supervision.

[0042] The cluster analysis of high-risk nodes in the spectral space is performed to divide different risk clusters, and the identification results of each high-risk cluster are output, which specifically includes: For each , get the The coordinates of the nodes in the two-dimensional spectral embedding space : ;in, is the eigenvector No. Quantity; is the eigenvector No. Component; Map high-risk nodes to two-dimensional space to facilitate spatial clustering; Get the geometric center of the high-risk cluster in the spectral space ; Obtain the spatial average position of high-risk nodes as the clustering benchmark; Calculate the The Euclidean distance from the high-risk node to the center ; Quantify the distance between each node and the center for subsequent threshold screening; Get the median of a distance set : ;in, Sort in ascending order elements; Median function; adaptively determine cluster radius without presetting the number of clusters or distance threshold; Construct the final identified high-risk node cluster ; Filter out the high-risk node group closest to the center to complete the risk cluster division.

[0043] By mapping the screened high-risk entities to the spectral space and adaptively clustering them based on spatial geometric features, the problem of traditional clustering requiring a preset number of clusters or distance thresholds and highly subjective boundaries is solved. First, the low-dimensional representation obtained by spectral decomposition is used to obtain the coordinates of each high-risk entity in the two-dimensional space. Then, the spatial distance between the point and the cluster center is calculated and the cluster radius is adaptively determined in combination with the median, and finally the core group of the high-risk cluster is automatically identified. Accurate and efficient risk group division can be completed without manually estimating the number of clusters or adjusting the distance threshold; at the same time, the spatial clustering results are easy to visualize, which provides convenience for regulators to intuitively grasp the distribution of risk groups; in addition, this clustering method can be dynamically adjusted as data is updated, and has good adaptability to emergency events or market mutations, providing a solid basis for subsequent precise risk management and control and strategy formulation.

[0044] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0045] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for managing power transaction security risks based on big data analysis, characterized in that: include: Establish a set of transaction entities and billing periods, collect and organize the transaction volume data of all entities in each period, and construct a complete symmetric adjacency matrix by processing outliers and missing data; Based on the adjacency matrix, the degree value of each node is calculated, and the degree matrix and normalized Laplace matrix are constructed accordingly; Combined with the historical quotation information of trading entities, the average quotation and volatility of each entity are calculated, and the risk deviation of each node is obtained based on the standardized method to form the initial risk vector; According to the maximum value of each node degree, the step size and convergence threshold of risk diffusion iteration are set; Adopting the risk diffusion iterative algorithm on the graph structure, the risk vector of each node is continuously updated iteratively until the preset convergence standard or the maximum number of iterations is reached, thus obtaining a stable risk distribution result. Perform Laplace spectral decomposition on the risk distribution results to obtain the eigenvalues ​​and eigenvectors of the normalized Laplace matrix, and extract spectral features related to high-risk clustering; Based on the risk distribution and spectrum analysis results, risk identification thresholds are set to screen high-risk nodes and form an index set of high-risk entities; Perform cluster analysis on high-risk nodes in the spectral space, divide them into different risk clusters, and output the identification results of each high-risk cluster.

2. The method for managing power transaction security risks based on big data analysis according to claim 1 is characterized in that: The process of establishing a set of transaction entities and billing periods, collecting and collating transaction volume data of all entities in each period, and constructing a complete symmetric adjacency matrix by processing outliers and missing data specifically includes: Read electricity transaction records, including Transaction entities and billing period, respectively construct the collection of all trading entities in the electricity market Collection with historical trading sessions : , ;in, For the a trading entity; Index of the transaction subject; is the time period index; Reading raw transaction volume ;in, For the period within, subject To the subject electricity trading volume; like If missing, then ; Get the maximum trading volume of all principal pairs in all periods : ； Abnormal truncation: If , then let ; Calculation from the main body To the subject Total transaction volume during all periods , ; Computing Subject With the subject Total two-way trading volume , ; Get the maximum value among the symmetric edge weights ; Constructing a symmetric adjacency matrix ,in, for the reason OK The set of all matrices whose columns are real numbers; For nodes arrive The normalized adjacency strength is: 。 3. The method for managing power transaction security risks based on big data analysis according to claim 2 is characterized in that: The process of calculating the degree of each node based on the adjacency matrix and constructing the degree matrix and normalized Laplace matrix accordingly includes: Calculate the normalized adjacency matrix Node degree ; structure The diagonal matrix ;No. The diagonal element is ; Construct the normalized Laplace matrix .

4. The method for managing power transaction security risks based on big data analysis according to claim 3 is characterized in that: The above mentioned process combines the historical quotation information of the trading entities, calculates the average quotation and volatility of each entity, obtains the risk deviation of each node based on the standardized method, and forms the initial risk vector, specifically including: Get the subject In the period The electricity price is ; Calculate the The node's historical average quote ; Calculate the Standard deviation of the node's quote sequence ; Perform deviation normalization, specifically: ;in, For nodes Normalized deviation of quotes from the historical mean in the latest period; Construct the initial risk vector: ;in is the initial risk vector, and its dimension is , No. The component is .

5. The method for managing power transaction security risks based on big data analysis according to claim 4 is characterized in that: The step size and convergence threshold of the risk diffusion iteration are set according to the maximum value of each node degree, specifically including: Get the maximum degree of all nodes ; Set the diffusion iteration step size ; Set the convergence threshold ;in, is the norm operation.

6. A method for managing power transaction security risks based on big data analysis according to claim 5, characterized in that: The risk diffusion iterative algorithm based on the graph structure is used to continuously iterate and update the risk vector of each node until a preset convergence standard or a maximum number of iterations is reached to obtain a stable risk distribution result. Specifically, the algorithm includes: Let the maximum number of iterations be ; right implement: ;in, For the Risk vector for round iteration; Number the iterations; like , then let And stop the iteration immediately; is the number of iterations required to converge or reach the upper limit; like If it does not converge, let and stop iterating; The final converged risk distribution vector is .

7. The method for managing power transaction security risks based on big data analysis according to claim 6 is characterized in that: The risk distribution results are subjected to graph Laplace spectral decomposition to obtain the eigenvalues ​​and eigenvectors of the normalized Laplace matrix, and the spectral features associated with high-risk aggregation are extracted, specifically including: Eigenvalue solution: ,have to ;in, is the characteristic value; for Identity matrix; For the Small eigenvalues; is the eigenvalue index; For each , solve the equation , and order ;in, For The corresponding unit eigenvector; Extract the vector corresponding to the second smallest eigenvalue The vector corresponding to the third smallest eigenvalue .

8. The method for managing power transaction security risks based on big data analysis according to claim 7 is characterized in that: Based on the risk distribution and spectrum analysis results, the risk identification threshold is set, high-risk nodes are screened, and an index set of high-risk entities is formed, which specifically includes: Calculate the average risk of all nodes ; Calculate the standard deviation of the equilibrium risk vector ; Set a high-risk identification threshold ; Build an index set of all high-risk nodes ; like , then terminate; if ,make , and terminate; wherein, For collection The number of elements in ; is the set of high-risk cluster nodes in the spectrum space.

9. The method for managing power transaction security risks based on big data analysis according to claim 8 is characterized in that: The cluster analysis of high-risk nodes in the spectral space is performed to divide different risk clusters, and the identification results of each high-risk cluster are output, which specifically includes: For each , get the The coordinates of the nodes in the two-dimensional spectral embedding space : ;in, is the eigenvector No. Quantity; is the eigenvector No. Quantity; Get the geometric center of the high-risk cluster in the spectral space ; Calculate the The Euclidean distance from the high-risk node to the center ; Get the median of a distance set : ;in, Sort in ascending order elements; Median function; Construct the final identified high-risk node cluster .

Images