Online data anomaly detection method for power distribution network

Abstract

The invention relates to the technical field of abnormal data detection, and discloses a power distribution network online data anomaly detection method, which comprises the following steps: deconstructing a failure concurrent event chain of a power distribution network, and generating an equipment failure topology library by establishing a causal association diagram of adjacent failure nodes in the failure concurrent event chain; s2, constructing an environment interference matrix of the power distribution network based on the equipment failure topology library and the environment monitoring parameters of the power distribution network, and performing sensitivity superposition on the environment interference matrix based on a pre-acquired strategy sensitivity coefficient to obtain a heterogeneous gradient matrix; s3, actively disturbing the heterogeneous gradient matrix to obtain an anti-fact virtual matrix, and performing attention confrontation splicing on the anti-fact virtual matrix and the physical feature vector of the power distribution network to obtain an anti-interference environment matrix; according to the invention, the power distribution network anomaly detection efficiency can be improved.

Description

Technical Field

[0001] This invention relates to the field of abnormal data detection technology, and in particular to a method for detecting online data anomalies in power distribution networks. Background Technology

[0002] Traditional methods rely on threshold alarms for isolated devices, failing to systematically deconstruct the chain-like topology of multiple device failures. This makes it difficult to quickly locate the source when sudden concurrent failures occur. Using only static threshold criteria ignores the multidimensional coupling effect of environmental interference and cannot generate an environmental sensitivity quantification matrix, causing anomaly detection to lag behind the actual speed of fault propagation.

[0003] On the other hand, existing solutions lack proactive defense capabilities for complex scenarios. Their data processing flow passively receives monitoring signals without introducing environmental disturbances, making them prone to false alarms and missed alarms under strong interference conditions. Secondly, existing methods use linear rules for anomaly propagation simulation, without establishing a counterfactual rule engine and a domino effect transmission model, making it impossible to predict the cascading failure paths of multiple devices, thus keeping the security defense strategy in a passive response state.

[0004] Improving the efficiency of anomaly detection when abnormal information appears in the power distribution network has become an urgent problem to be solved. Summary of the Invention

[0005] This invention provides a method for detecting online data anomalies in a power distribution network. Its main purpose is to solve the problem of how to improve the efficiency of anomaly detection when abnormal information occurs in the power distribution network.

[0006] To achieve the above objectives, the present invention provides a method for detecting online data anomalies in a power distribution network, characterized in that the method includes: S1: Deconstruct the failure concurrency event chain of the distribution network, and generate a device failure topology library by establishing a causal relationship graph of adjacent failure nodes in the failure concurrency event chain; S2: Construct the environmental interference matrix of the power distribution network based on the equipment failure topology library and the environmental monitoring parameters of the power distribution network. Perform sensitivity superposition on the environmental interference matrix based on the pre-acquired strategy sensitivity coefficient to obtain a heterogeneous gradient matrix. S3: Actively perturb the heterogeneous gradient matrix to obtain a counterfactual virtual matrix, and perform attention adversarial splicing on the counterfactual virtual matrix and the physical feature vector of the power distribution network to obtain an anti-interference environment matrix; S4: Perform hierarchical dynamic pruning on the anti-interference environment matrix to obtain the device influence matrix; S5: Perform scene rule fusion on the heterogeneous gradient matrix and the anti-interference environment matrix to obtain counterfactual rules, and perform domino deduction on the device influence matrix based on the counterfactual rules to obtain multi-device abnormal propagation information.

[0007] Optionally, the process of deconstructing the failure concurrency event chain of the distribution network and generating a device failure topology library by establishing a causal relationship graph of adjacent failure nodes in the failure concurrency event chain includes: The historical failure data stream of the distribution network is acquired and its spatiotemporal features are extracted to obtain the failure concurrent event chain; Physical dependency analysis is performed on the nodes and adjacent nodes in the failure concurrency event chain to obtain a failure causal relationship graph. Based on a predefined equipment failure mode library, the failure causal relationship graph is topologically mapped to obtain a distribution network equipment failure topology library.

[0008] Optionally, constructing the environmental interference matrix of the power distribution network based on the equipment failure topology library and the environmental monitoring parameters of the power distribution network includes: Extract the device association relationships from the device failure topology library and analyze the association weights of the device association relationships; Risk analysis is performed on the environmental monitoring parameters based on the aforementioned correlation weights to obtain environmental risk factors; The environmental disturbance matrix of the power distribution network is constructed based on the correlation weights and environmental risk factors.

[0009] Optionally, the formula for calculating the strategy sensitivity coefficient is: in: The sensitivity coefficient of the stated strategy. The probability distribution entropy corresponding to the associated weights. This is the distance matrix corresponding to the environmental risk factors. It is a nonlinear activation operator. Here is the numerical stability constant. for The covariance matrix corresponding to the time penalty coefficient at time step [time]. For the device association matrix, The covariance eigenvalues ​​of the device correlation matrix are... for The dynamic environmental coupling factor at any given moment. For a moment, is the time-varying conduction attenuation coefficient.

[0010] Optionally, the step of performing sensitivity superposition on the environmental disturbance matrix based on the pre-acquired policy sensitivity coefficients to obtain a heterogeneous gradient matrix includes: The sensitivity coefficients of the aforementioned strategies are eliminated based on the 3σ principle to obtain standard sensitivity coefficients; The environmental disturbance matrix is ​​weighted by the standard sensitivity coefficient to obtain a heterogeneous gradient matrix.

[0011] Optionally, the active perturbation of the heterogeneous gradient matrix to obtain the counterfactual virtual matrix includes: The environmental interference factors of the power distribution network are quantified and organized to obtain the interference impact value; The heterogeneous gradient matrix is ​​weighted and scaled based on the interference impact value to obtain a counterfactual virtual matrix.

[0012] Optionally, the attention adversarial concatenation of the counterfactual virtual matrix and the physical feature vectors of the distribution network to obtain the anti-interference environment matrix includes: Based on the physical attributes of the equipment in the power distribution network and the counterfactual virtual matrix, the core physical vectors in the physical feature vectors are selected. Based on preset abnormal path countermeasure rules, the counterfactual virtual matrix and the core physical vector are linearly mapped to obtain the anti-interference environment matrix.

[0013] Optionally, the hierarchical dynamic pruning of the anti-interference environment matrix to obtain the device influence matrix includes: Extract the influence weights of the anti-interference environment matrix, and perform priority pruning on the equipment node vectors of the power distribution network based on the influence weights to obtain hierarchical influence vectors; The hierarchical influence vector and the equipment matrix of the distribution network are subjected to feature interaction to obtain the equipment influence matrix of the distribution network.

[0014] Optionally, the step of fusing scene rules between the heterogeneous gradient matrix and the anti-interference environment matrix to obtain counterfactual rules includes: Based on the high-risk paths of the heterogeneous gradient matrix, the anti-interference environment matrix is ​​mapped to obtain the triggering conditions. The triggering conditions are subjected to counterfactual verification to obtain counterfactual rules.

[0015] Optionally, the step of performing a domino effect deduction on the device influence matrix based on the counterfactual rule to obtain multi-device abnormal propagation information includes: Based on the counterfactual rules, the topological connection nodes of the device influence matrix are triggered to fail step by step to obtain the failure propagation path chain; Based on the failure propagation path chain, the topology connection nodes are traversed to obtain cascading anomaly information, the cascading anomaly information of the distribution network is recorded, and multi-device anomaly propagation information is generated.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. By deconstructing the chain of concurrent failure events and establishing a causal relationship graph of adjacent failure nodes, this method can accurately identify the failure propagation path between devices, thereby generating a high-precision device failure topology library, avoiding blind spots and omissions in traditional methods. Combined with the policy sensitivity coefficient, sensitivity is superimposed to generate a heterogeneous gradient matrix, which enables the system to automatically adjust the sensitivity threshold under dynamic environmental parameters, enhancing the adaptability and stability of detection, effectively resisting noise interference, and improving the signal-to-noise ratio of anomaly detection.

[0017] 2. The active perturbation and attention adversarial splicing mechanism enhances the anti-interference capability of data by simulating counterfactual scenarios, ensuring the rapid capture of core abnormal patterns in complex environments, avoiding false alarms and missed alarms, thereby improving the overall detection accuracy and providing a reliable basis for real-time decision-making.

[0018] 3. Through a hierarchical dynamic pruning mechanism, the system significantly reduces redundant computation while retaining information about the impact of critical equipment, thereby improving data processing speed and resource utilization, making it particularly suitable for the online monitoring needs of large-scale distribution networks. Furthermore, the combination of scenario rule fusion and domino-effect simulation with heterogeneous gradient matrices and anti-interference environment matrices not only allows for the counterfactual derivation of anomaly propagation chains between devices but also enables the prediction and prevention of potential future faults. This not only reduces the operational risks of the distribution network but also reduces the cost of manual intervention through automation, providing cost-effective data support for the construction and maintenance of smart grids. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an online data anomaly detection method for a power distribution network according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments belong to some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0022] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0023] This application provides a method for detecting online data anomalies in a power distribution network. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for detecting online data anomalies in a power distribution network can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0024] like Figure 1 The diagram shown is a flowchart illustrating a method for detecting online data anomalies in a power distribution network according to the present invention. In this embodiment, the method for detecting online data anomalies in a power distribution network includes: S1: Deconstruct the failure concurrency event chain of the distribution network, and generate a device failure topology library by establishing a causal relationship graph of adjacent failure nodes in the failure concurrency event chain.

[0025] In this embodiment of the invention, the step of deconstructing the failure concurrency event chain of the distribution network and generating a device failure topology library by establishing a causal relationship graph of adjacent failure nodes in the failure concurrency event chain includes: The historical failure data stream of the distribution network is acquired and its spatiotemporal features are extracted to obtain the failure concurrent event chain; Physical dependency analysis is performed on the nodes and adjacent nodes in the failure concurrency event chain to obtain a failure causal relationship graph. Based on a predefined equipment failure mode library, the failure causal relationship graph is topologically mapped to obtain a distribution network equipment failure topology library.

[0026] Specifically, the historical failure data stream consists of fault records generated during the operation of various devices in the distribution network, such as transformers, circuit breakers, and lines. It includes information such as device ID, fault type, occurrence time, and geographical location.

[0027] Spatiotemporal characteristics include temporal and spatial characteristics. Temporal characteristics include the timestamp of the fault occurrence, duration, and periodicity, such as the fact that high temperatures in summer can easily cause overload faults. Spatial characteristics include the physical location of the equipment, such as densely packed sections of lines in a certain area and network connection relationships, such as feeder topology.

[0028] Specifically, physical dependency analysis involves analyzing the current and voltage transmission relationships between devices based on the electrical wiring diagram of the distribution network, and determining the physical path of fault propagation, such as how a short circuit in a certain line can cause the upstream circuit breaker to trip.

[0029] It also includes quantified dependency weights, which calculate the probability of adjacent devices failing simultaneously or sequentially through historical data statistics, such as the conditional probability of device B failing after device A fails.

[0030] Specifically, the failure causal relationship graph is as follows: nodes represent devices, edges represent causal relationships, such as A→B means that a failure of A may lead to a failure of B, and the weight of the edge represents the strength of the dependency.

[0031] Specifically, the topology logic mapping involves matching a predefined library of equipment failure modes, such as transformer overheating and line grounding, to convert the causal relationship graph into a standardized topology structure, forming an "equipment failure topology library".

[0032] S2: Construct the environmental interference matrix of the power distribution network based on the equipment failure topology library and the environmental monitoring parameters of the power distribution network. Perform sensitivity superposition on the environmental interference matrix based on the pre-acquired policy sensitivity coefficients to obtain a heterogeneous gradient matrix.

[0033] In this embodiment of the invention, constructing the environmental interference matrix of the power distribution network based on the equipment failure topology library and the environmental monitoring parameters of the power distribution network includes: Extract the device association relationships from the device failure topology library and analyze the association weights of the device association relationships; Risk analysis is performed on the environmental monitoring parameters based on the aforementioned correlation weights to obtain environmental risk factors; The environmental disturbance matrix of the power distribution network is constructed based on the correlation weights and environmental risk factors.

[0034] Specifically, the equipment failure topology library is a data set that records the interrelationships between various types of equipment in the distribution network, such as transformers, circuit breakers, transmission lines, and switchgear, when they fail due to faults. It includes information such as fault propagation paths between equipment and historical common-cause failure records. For example, a short-circuit fault on a transmission line may cause a connected circuit breaker to trip; this relationship between the line and the circuit breaker is stored in the topology library.

[0035] Specifically, the equipment relationships are those formed between distribution network equipment due to electrical connections, functional dependencies, etc. For example, transformers and low-voltage side distribution cabinets are related due to power transmission, and protection devices and protected equipment are related due to fault protection.

[0036] The correlation weight is a quantitative value that measures the degree of correlation between equipment. It is influenced by the physical environment, such as the location of the equipment, electrical distance, and frequency of historical failures caused by the same cause. For example, in areas prone to lightning, transmission lines and surge arresters on the same path have a high probability of failing together due to lightning strikes, resulting in a high correlation weight. In urban power distribution networks, adjacent transformers with closely related loads also have a high correlation weight due to mutual influences such as load transfer.

[0037] Specifically, the environmental monitoring parameters include various monitoring data covering the physical environment in which the power distribution network equipment is located, such as temperature (affecting equipment heat dissipation and insulation performance), humidity (causing equipment to become damp and insulation to age), wind speed (affecting line galloping and tower stress), lightning strike frequency (threatening line and equipment insulation), and pollution level (causing insulator flashover).

[0038] Environmental risk factors are quantitative indicators that measure the impact of environmental monitoring parameters on the failure risk of power distribution network equipment. They reflect the potential risk of equipment failure caused by different environmental parameters under specific equipment correlation.

[0039] Specifically, based on the equipment correlation and correlation weight, the effects of different environmental monitoring parameters on the associated equipment are determined. For example, high-temperature environmental parameters, in the correlation between transformers and cooling systems, will affect the transformer's heat dissipation, and thus affect the cooling system load and the transformer's own failure risk through the correlation.

[0040] A risk analysis model is constructed, incorporating correlation weights as coefficients to calculate the impact of environmental monitoring parameters on equipment failure probability. For example, fault tree analysis can be used, treating environmental parameters as base events and combining them with logic gates representing equipment relationships to calculate top events, such as the probability of equipment failure, thus obtaining environmental risk factors. Alternatively, a machine learning model can be used, employing historical environmental parameters and equipment failure data as training sets to learn the mapping between environmental parameters and failure risk, outputting risk factors.

[0041] By substituting real-time or historical environmental monitoring data, environmental risk factors are calculated through modeling. For example, in a severely polluted area, by combining the correlation weights between insulators and lines and based on real-time pollution monitoring parameters, the environmental risk factors of insulator flashover caused by pollution, which in turn lead to line faults, can be calculated.

[0042] Specifically, the environmental interference matrix presents the combined impact of the interference of environmental risk factors and the associated weights under different equipment relationships in the power distribution network in matrix form. The rows and columns can correspond to different equipment or equipment combinations, and the matrix elements are the quantitative values ​​of the degree of environmental interference.

[0043] Specifically, the rows and columns of the matrix are determined based on the number of distribution network equipment and the complexity of their relationships. Typically, the rows and columns correspond to the equipment or equipment groups involved in the relationships, covering key equipment such as transformers, lines, and circuit breakers.

[0044] The association weights and environmental risk factors are combined and calculated to serve as matrix element values. For example, for the association between equipment A and equipment B, the association weight is WAB, and the environmental risk factor is RAB. The corresponding element in the matrix can be obtained by multiplying WAB by RAB. Alternatively, multiple environmental risk factors can be considered, weighted, summed, and then combined with the association weights to calculate the value, reflecting the degree of interference to the association relationship of this set of equipment under this combination of environmental parameters.

[0045] Fill the matrix elements according to the above rules to form an environmental interference matrix. Then, use historical fault cases and simulated fault scenarios to verify and correct the matrix to ensure that it can accurately reflect the interference of the physical environment on the correlation of distribution network equipment. This provides a basis for analyzing equipment anomalies caused by environmental factors in subsequent online data anomaly detection.

[0046] In this embodiment of the invention, the formula for calculating the strategy sensitivity coefficient is as follows: in: The sensitivity coefficient of the stated strategy. The probability distribution entropy corresponding to the associated weights. This is the distance matrix corresponding to the environmental risk factors. It is a nonlinear activation operator. Here is the numerical stability constant. for The covariance matrix corresponding to the time penalty coefficient at time step [time]. For the device association matrix, The covariance eigenvalues ​​of the device correlation matrix are... for The dynamic environmental coupling factor at any given moment. For a moment, is the time-varying conduction attenuation coefficient.

[0047] Specifically, in the scenario of online data anomaly detection in distribution networks, the policy sensitivity coefficient This value is used to measure the sensitivity of a distribution network fault detection or anomaly identification strategy to environmental changes and equipment failure correlations. The higher the value, the more sensitive the strategy is to disturbances in the distribution network's physical environment, such as lightning strikes or temperature changes triggering a chain reaction of equipment failures, thus helping to determine the effectiveness and adaptability of the strategy in complex environments.

[0048] The probability distribution entropy corresponds to the associated weights. In a power distribution network, the causal relationship of equipment failure, such as the causal chain of line short circuit → circuit breaker tripping, has different probabilities of occurrence for causal pairs with different basic weights. The probability distribution entropy measures the uncertainty of this weight. The larger the entropy, the more random and difficult to predict the causal relationship of failure. For example, in a power distribution network in a remote mountainous area, equipment is greatly affected by extreme weather, and the causal entropy of failure may be even higher.

[0049] Specifically, The matrix is ​​composed of environmental risk factors, such as the influence coefficients of temperature and humidity on equipment insulation. The distance matrix reflects the differences between different combinations of environmental factors. For example, the difference in correction factors between high temperature and high humidity environments and normal temperature and dry environments is quantified by the distance between matrix elements, reflecting the spatial-type span of environmental risks.

[0050] Specifically, Similar to neural network activation functions, nonlinear transformations are performed on the fused high-dimensional data to highlight key coupling features, such as filtering out minor environmental interferences and amplifying the impact of strongly correlated environmental-failure causality, such as lightning strikes and icing, thus adapting to the nonlinear and strongly coupled physical characteristics of power distribution networks.

[0051] Specifically, Due to the large volume of data in online monitoring of power distribution networks and the susceptibility of numerical oscillations in calculations, such as matrix singularities caused by sudden environmental changes, ϵ is usually taken as a small positive number, such as... This is used to avoid meaningless cases such as log(0) in logarithmic operations and to ensure computational stability.

[0052] Specifically, Distribution network faults have time characteristics, such as concentrated faults during peak load periods and lightning strikes. The time penalty coefficient reflects the strategy's attention to anomalies at different times. Does the covariance matrix characterize the fluctuation correlation of these coefficients? For example, the covariance of the penalty coefficients during the evening peak and early morning periods reflects the time-varying impact of load fluctuations on the strategy.

[0053] It measures the degree of divergence of the time penalty coefficient covariance matrix. The larger the determinant, the more drastic the fluctuation of the strategy weights in the time dimension. For example, in summer, extreme weather occurs frequently, and the environmental interference varies greatly at different times, so the determinant may be higher.

[0054] It compresses and maps the complexity of fluctuations in the time dimension, transforming large-span determinant values ​​into relatively flat logarithmic scales, which facilitates the calculation of parameters for other dimensions of the environment and equipment. At the same time, it highlights the logarithmic sensitivity of time fluctuations, being insensitive to small fluctuations and responding quickly to large fluctuations.

[0055] Specifically, It is a matrix that describes the electrical and physical connections between equipment in a power distribution network, such as the topological connection weights of transformer-line-circuit breaker. The larger the value of the matrix element, the closer the connection between the equipment. For example, the connection matrix value between a city ring main unit and its surrounding lines is higher than that of remote and isolated equipment.

[0056] This involves performing covariance decomposition on the equipment correlation matrix, with eigenvalues ​​reflecting the intensity of the main fluctuation direction of the correlation matrix. Larger values ​​indicate that a certain type of equipment correlation pattern, such as series correlation in radial lines or parallel correlation in ring networks, is more prominent in the distribution network and has a more significant impact on strategy.

[0057] It is a dynamic binding coefficient between real-time environmental parameters, such as wind speed and lightning intensity, and the equipment, reflecting how the environment changes the equipment's correlation strength in real time. For example, in windy weather, the swaying of transmission lines may enhance the correlation coupling between adjacent towers.

[0058] Specifically, By using the eigenvalue logarithm operation of the time-dimensional covariance matrix, compared with the traditional linear weighting of time series, it is better able to characterize the decay and abrupt changes of features at multiple time scales during fault propagation.

[0059] Specifically, The critical temperature for coupling between the distribution network environment and equipment can be considered. For example, high temperatures can cause a decrease in equipment insulation, making the equipment more sensitive to environmental factors. This sensitivity can be calibrated; the smaller the value, the easier it is for environmental coupling to trigger a policy response.

[0060] in The calculation formula is: in: The time-varying conduction attenuation coefficient, For dynamic time coupling factor, For robust compression operator coefficients, As a dynamic environment coupling factor, For the device association matrix, The covariance eigenvalues ​​of the device correlation matrix are... For predefined constants, Let the time parameter be the tensor norm. Here is the numerical stability constant. The largest eigenvalue of the time covariance matrix. for The covariance matrix corresponding to the time penalty coefficient at time step [time]. For a moment, The Hadamard product of environmental factors and equipment spectral energy. It is the sum of the diagonal elements. for The transpose of the matrix.

[0061] Specifically, The sum of the diagonal elements is used to quantify the overall coupling strength between environmental factors and equipment.

[0062] , where is the largest eigenvalue of the time covariance matrix, and measures the principal component strength of failure risk in the time dimension.

[0063] Specifically, The coupling strength of the environment-equipment association is nonlinearly compressed, and the output value is in the range of (-1,1) to simulate the saturation effect of the environmental interference-equipment association in the power distribution network. For example, under extreme conditions, the equipment association is limited by physical limits, and the coupling strength no longer grows indefinitely.

[0064] Specifically, To link equipment failure logic with environmental risks, then use Activate key features to capture how the environment alters the causal probability of failures.

[0065] In this embodiment of the invention, the step of performing sensitivity superposition on the environmental interference matrix based on the pre-acquired policy sensitivity coefficients to obtain a heterogeneous gradient matrix includes: The sensitivity coefficients of the aforementioned strategies are eliminated based on the 3σ principle to obtain standard sensitivity coefficients; The environmental disturbance matrix is ​​weighted by the standard sensitivity coefficient to obtain a heterogeneous gradient matrix.

[0066] Specifically, the 3σ principle, also known as the 3-standard-deviation principle, is a statistical method based on the normal distribution. In the distribution network scenario, the values ​​of the policy sensitivity coefficients theoretically follow a normal distribution around the mean, and σ measures the dispersion of these coefficients. The 3σ principle states that values ​​deviating from the mean by more than three standard deviations can be considered outliers.

[0067] Specifically, the standard sensitivity coefficient is a coefficient that has been screened using the 3σ principle and can reasonably reflect the relationship between the physical environment of the distribution network and the policy sensitivity. Different standard sensitivity coefficients correspond to different equipment associations and different environmental regions, reflecting the degree of attention the policy pays to the environment-equipment association scenario.

[0068] Specifically, the heterogeneous gradient matrix is ​​the matrix obtained by weighting the environmental interference matrix with standard sensitivity coefficients. Heterogeneity reflects the gradient differences of environmental interference under different physical environment areas and different equipment connections in the distribution network. The gradient reflects the changing trend of the degree of environmental interference, and helps to identify the distribution and propagation of environmental interference anomalies in the power distribution network.

[0069] S3: Actively perturb the heterogeneous gradient matrix to obtain a counterfactual virtual matrix. Perform attention adversarial splicing on the counterfactual virtual matrix and the physical feature vector of the power distribution network to obtain an anti-interference environment matrix.

[0070] In this embodiment of the invention, the active perturbation of the heterogeneous gradient matrix to obtain the counterfactual virtual matrix includes: The environmental interference factors of the power distribution network are quantified and organized to obtain the interference impact value; The heterogeneous gradient matrix is ​​weighted and scaled based on the interference impact value to obtain a counterfactual virtual matrix.

[0071] Specifically, the interference impact value is a quantitative result obtained by quantifying and normalizing environmental interference factors, reflecting the degree of threat posed by the interference to the distribution network. It needs to be calibrated in conjunction with the physical characteristics of the distribution network. For example, when the lightning strike intensity is >15kA, the threat to line insulation increases sharply, and the quantitative value in this range has a higher weight.

[0072] Specifically, weighted scaling uses the impact value of disturbance as weight to amplify / scale the elements of the heterogeneous gradient matrix, simulating how the gradient matrix changes when environmental disturbances are enhanced / weakened.

[0073] In this embodiment of the invention, the step of performing attention adversarial concatenation on the counterfactual virtual matrix and the physical feature vector of the power distribution network to obtain an anti-interference environment matrix includes: The core physical vectors in the physical feature vectors are selected based on the physical attributes of the equipment in the power distribution network and the counterfactual virtual matrix.

[0074] Based on preset abnormal path countermeasure rules, the counterfactual virtual matrix and the core physical vector are linearly mapped to obtain the anti-interference environment matrix.

[0075] Specifically, the physical attributes of the equipment are the inherent physical parameters of the power distribution network equipment, which determine the abnormal response of the equipment under environmental interference.

[0076] The physical feature vector is a high-dimensional vector that describes the physical state of the distribution network. It includes equipment physical attributes, environmental parameters such as temperature and humidity, and operating status such as load current and voltage.

[0077] The core physical vector, selected from the physical feature vectors, is the key dimension that best distinguishes between environmental interference and equipment anomalies, and is a condensed expression of the physical essence of the distribution network.

[0078] By setting a correlation threshold, such as a correlation coefficient > 0.7, physical feature dimensions that are strongly correlated with the counterfactual virtual matrix are selected. For example, in a counterfactual scenario of lightning interference, dimensions such as line insulation thickness and surge arrester operating voltage are prioritized for selection.

[0079] Based on the physical mechanism of the power distribution network, key dimensions are manually verified. For example, when the transformer oil temperature exceeds 80°C, the oil temperature plus the load current must be taken as the core vector, because this is the key cause of insulation aging.

[0080] Specifically, the abnormal path countermeasure rules simulate the physical path of power distribution network environmental interference → equipment abnormality → fault propagation, with preset interference-defense countermeasure logic. For example: Rule 1: Lightning interference → line insulator flashover → circuit breaker tripping, the countermeasure logic is "strengthen insulator insulation status monitoring and weaken interference response of non-associated equipment"; Rule 2: High temperature environment → transformer overload → protection device action, countermeasure logic is "focus on oil temperature and load current correlation, suppress weak correlation interference such as ambient humidity".

[0081] The anti-interference environment matrix is ​​a matrix that integrates physical core features, counterfactual interference scenarios, and adversarial rules. It is used for online anomaly detection to highlight abnormal signals in the real physical environment and filter environmental noise.

[0082] S4: Perform hierarchical dynamic pruning on the anti-interference environment matrix to obtain the device influence matrix.

[0083] In this embodiment of the invention, the hierarchical dynamic pruning of the anti-interference environment matrix to obtain the device influence matrix includes: Extract the influence weights of the anti-interference environment matrix, and perform priority pruning on the equipment node vectors of the power distribution network based on the influence weights to obtain hierarchical influence vectors; The hierarchical influence vector and the equipment matrix of the distribution network are subjected to feature interaction to obtain the equipment influence matrix of the distribution network.

[0084] Specifically, the influence weight is a quantified value of the contribution of different dimensions in the anti-interference environment matrix, such as equipment physical attributes and environmental interference types, to the impact of equipment anomalies. For example, transformer oil temperature has a high influence weight on overload anomalies, while humidity has a low influence weight on the anomaly.

[0085] A device node vector is a vector that describes the physical state of a single device. It includes device attributes such as transformer capacity, operating status such as load current, environmental factors such as oil temperature, and other dimensions. It is an atomic-level expression of the physical characteristics of the distribution network.

[0086] Priority pruning is based on the influence weight, retaining the high-influence dimensions in the device node vector and pruning the low-influence / redundant dimensions, highlighting the most critical physical features for anomaly detection.

[0087] The hierarchical influence vector is a pruned vector organized according to the physical hierarchy of the distribution network, such as equipment → feeder → substation. It retains the key influence dimensions of each level and reflects the hierarchical propagation characteristics of environmental interference → equipment anomaly.

[0088] S5: Perform scene rule fusion on the heterogeneous gradient matrix and the anti-interference environment matrix to obtain counterfactual rules, and perform domino deduction on the device influence matrix based on the counterfactual rules to obtain multi-device abnormal propagation information.

[0089] In this embodiment of the invention, the step of fusing scene rules between the heterogeneous gradient matrix and the anti-interference environment matrix to obtain counterfactual rules includes: Based on the high-risk paths of the heterogeneous gradient matrix, the anti-interference environment matrix is ​​mapped to obtain the triggering conditions. The triggering conditions are subjected to counterfactual verification to obtain counterfactual rules.

[0090] Specifically, high-risk paths are those elements in the matrix whose gradient values ​​are significantly higher than the threshold. These correspond to propagation paths in the distribution network physical environment where environmental interference → equipment anomaly risk is high, such as the gradient path of lightning strike → line insulator → circuit breaker tripping.

[0091] Specifically, for heterogeneous gradient matrices, a gradient threshold is set (e.g., gradient value > 0.8, combined with distribution network equipment fault probability calibration) to extract high-risk paths. For example, if the gradient value of feeder A → transformer B → circuit breaker C in the matrix is ​​continuously higher than the threshold, the corresponding physical path feeder A is affected by environmental interference → transformer B is overloaded → circuit breaker C trips.

[0092] Based on the physical topology of the distribution network, verify the rationality of the path, such as whether the path conforms to the electrical connection relationship and the direction of fault propagation.

[0093] In this embodiment of the invention, the step of performing a domino effect deduction on the device influence matrix based on the counterfactual rule to obtain multi-device abnormal propagation information includes: Based on the counterfactual rules, the topological connection nodes of the device influence matrix are triggered to fail step by step to obtain the failure propagation path chain; Based on the failure propagation path chain, the topology connection nodes are traversed to obtain cascading anomaly information, the cascading anomaly information of the distribution network is recorded, and multi-device anomaly propagation information is generated.

[0094] Specifically, topology connection nodes are equipment nodes in the physical topology of a distribution network, such as transformers, circuit breakers, and lines, as well as the connection relationships between nodes, such as the connection between a line and a transformer, or the connection between a circuit breaker and a busbar.

[0095] Failure triggered in stages: This simulates the physical process by which a device fails due to environmental interference, triggering the sequential failure of its associated devices, such as line insulator flashover → line tripping → transformer overload → low-voltage side circuit breaker tripping.

[0096] Failure propagation path chain: A chain structure that records the order in which device anomalies propagate, such as device A → device B → device C, reflecting the propagation path of the anomaly in the distribution network topology.

[0097] Specifically, traversing the topology connection nodes includes: traversing each topology connection node along the failure propagation path chain, extracting the equipment anomaly type, such as flashover, tripping, overload, and trigger time; calculating the affected physical area based on protection logic and propagation delay, such as a feeder or a transformer area.

[0098] For example: a path chain, in line A → circuit breaker B → transformer C; Line A: Lightning flashover t=0s, affecting 10kV feeders in mountainous areas; Circuit breaker B: Overcurrent trip t=0.5s, affecting the 35kV busbar of the substation; Transformer C: Overload alarm t=1s, affecting 10kV distribution area). Combine the physical environment parameters of the power distribution network, such as the area where lightning strikes occurred and temperature distribution, to supplement information on the environmental causes of the anomalies.

[0099] Specifically, chain anomaly information is integrated according to the propagation time sequence and physical area hierarchy, such as substation → feeder → transformer area, to form structured data.

[0100] For example: Use a tree structure to record substation A anomaly → feeder 1 anomaly → transformer area 1 anomaly.

[0101] Calculate the impact weight of anomalies, such as the number of users involved and the amount of load loss, to quantify the severity of cascading anomalies.

[0102] Specifically, multi-device anomaly propagation information is the process of converting the integrated chain of anomaly information into a propagation map plus feature vectors.

[0103] The propagation map visualizes the abnormal path, such as plotting the fault point and propagation range on a map; the feature vector contains dimensions such as propagation length, environmental causes, and impact load, and is used as input for the online anomaly detection model.

[0104] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0105] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0107] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0108] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for detecting online data anomalies in a power distribution network, characterized in that, The method includes: S1: Deconstruct the failure concurrency event chain of the distribution network, and generate a device failure topology library by establishing a causal relationship graph of adjacent failure nodes in the failure concurrency event chain; S2: Construct the environmental interference matrix of the power distribution network based on the equipment failure topology library and the environmental monitoring parameters of the power distribution network. Perform sensitivity superposition on the environmental interference matrix based on the pre-acquired strategy sensitivity coefficient to obtain a heterogeneous gradient matrix. S3: Actively perturb the heterogeneous gradient matrix to obtain a counterfactual virtual matrix, and perform attention adversarial splicing on the counterfactual virtual matrix and the physical feature vector of the power distribution network to obtain an anti-interference environment matrix; S4: Perform hierarchical dynamic pruning on the anti-interference environment matrix to obtain the device influence matrix; S5: Perform scene rule fusion on the heterogeneous gradient matrix and the anti-interference environment matrix to obtain counterfactual rules, and perform domino deduction on the device influence matrix based on the counterfactual rules to obtain multi-device abnormal propagation information.

2. The method for detecting online data anomalies in a power distribution network as described in claim 1, characterized in that, The process involves deconstructing the concurrent failure event chain of the distribution network, establishing a causal relationship graph of adjacent failure nodes in the concurrent failure event chain, and generating a device failure topology library, including: The historical failure data stream of the distribution network is acquired and its spatiotemporal features are extracted to obtain the failure concurrent event chain; Physical dependency analysis is performed on the nodes and adjacent nodes in the failure concurrency event chain to obtain a failure causal relationship graph. Based on a predefined equipment failure mode library, the failure causal relationship graph is topologically mapped to obtain a distribution network equipment failure topology library.

3. The method for detecting online data anomalies in a power distribution network as described in claim 1, characterized in that, The construction of the environmental interference matrix of the power distribution network based on the equipment failure topology library and the environmental monitoring parameters of the power distribution network includes: Extract the device association relationships from the device failure topology library and analyze the association weights of the device association relationships; Risk analysis is performed on the environmental monitoring parameters based on the aforementioned correlation weights to obtain environmental risk factors; The environmental disturbance matrix of the power distribution network is constructed based on the correlation weights and environmental risk factors.

4. The method for detecting online data anomalies in a power distribution network as described in claim 3, characterized in that, The formula for calculating the strategy sensitivity coefficient is as follows: in: The sensitivity coefficient of the stated strategy. The probability distribution entropy corresponding to the associated weights. This is the distance matrix corresponding to the environmental risk factors. It is a nonlinear activation operator. Here is the numerical stability constant. for The covariance matrix corresponding to the time penalty coefficient at time step [time]. For the device association matrix, The covariance eigenvalues ​​of the device correlation matrix are... for The dynamic environmental coupling factor at any given moment. For a moment, is the time-varying conduction attenuation coefficient.

5. The method for detecting online data anomalies in a power distribution network as described in claim 4, characterized in that, The process of performing sensitivity superposition on the environmental disturbance matrix based on the pre-acquired policy sensitivity coefficients to obtain a heterogeneous gradient matrix includes: The sensitivity coefficients of the aforementioned strategies are eliminated based on the 3σ principle to obtain standard sensitivity coefficients; The environmental disturbance matrix is ​​weighted by the standard sensitivity coefficient to obtain a heterogeneous gradient matrix.

6. The method for detecting online data anomalies in a power distribution network as described in claim 1, characterized in that, The active perturbation of the heterogeneous gradient matrix yields a counterfactual virtual matrix, including: The environmental interference factors of the power distribution network are quantified and organized to obtain the interference impact value; The heterogeneous gradient matrix is ​​weighted and scaled based on the interference impact value to obtain a counterfactual virtual matrix.

7. The method for detecting online data anomalies in a power distribution network as described in claim 1, characterized in that, The step of performing attention adversarial concatenation on the counterfactual virtual matrix and the physical feature vectors of the distribution network to obtain an anti-interference environment matrix includes: Based on the physical attributes of the equipment in the power distribution network and the counterfactual virtual matrix, the core physical vectors in the physical feature vectors are selected. Based on preset abnormal path countermeasure rules, the counterfactual virtual matrix and the core physical vector are linearly mapped to obtain the anti-interference environment matrix.

8. The method for detecting online data anomalies in a power distribution network as described in claim 1, characterized in that, The hierarchical dynamic pruning of the anti-interference environment matrix to obtain the device influence matrix includes: Extract the influence weights of the anti-interference environment matrix, and perform priority pruning on the equipment node vectors of the power distribution network based on the influence weights to obtain hierarchical influence vectors; The hierarchical influence vector and the equipment matrix of the distribution network are subjected to feature interaction to obtain the equipment influence matrix of the distribution network.

9. The method for detecting online data anomalies in a power distribution network as described in claim 1, characterized in that, The process of fusing scene rules between the heterogeneous gradient matrix and the anti-interference environment matrix to obtain counterfactual rules includes: Based on the high-risk paths of the heterogeneous gradient matrix, the anti-interference environment matrix is ​​mapped to obtain the triggering conditions. The triggering conditions are subjected to counterfactual verification to obtain counterfactual rules.

10. The method for detecting online data anomalies in a power distribution network as described in claim 1, characterized in that, The step of performing a domino effect deduction on the device influence matrix based on the counterfactual rule to obtain multi-device abnormal propagation information includes: Based on the counterfactual rules, the topological connection nodes of the device influence matrix are triggered to fail step by step to obtain the failure propagation path chain; Based on the failure propagation path chain, the topology connection nodes are traversed to obtain cascading anomaly information, the cascading anomaly information of the distribution network is recorded, and multi-device anomaly propagation information is generated.

Images