A power system power failure detection method and device, electronic equipment, storage medium and product

Abstract

This invention discloses a power system outage detection method, device, electronic equipment, storage medium, and product, relating to the field of power system operation monitoring technology. The power system outage detection method includes: acquiring power time-series data and communication time-series data; determining the time alignment confidence level between the power time-series data and the communication time-series data using a self-attention mechanism; determining the power grid equipment and the power supply correlation between the power grid equipment; constructing a power grid topology model according to the power grid equipment and the power supply correlation using a graph neural network; determining the spatial correlation strength based on the power grid topology model; determining the power system outage probability based on the time alignment confidence level and the spatial correlation strength; and generating an outage detection result according to the outage probability, thereby achieving accurate detection and early warning of power system outages.

Description

Technical Field

[0001] This invention relates to the field of power system operation monitoring technology, and in particular to a power system outage detection method, device, electronic equipment, storage medium and product. Background Technology

[0002] In the field of power system outage detection, technological innovation continues to drive improvements in power grid operation and maintenance efficiency, but it also faces multiple challenges in complex environments. Current mainstream technologies can be categorized into three types, each exhibiting unique application value and limitations.

[0003] Traditional threshold-based detection methods identify power outages by setting predefined voltage and current thresholds. While simple to implement and computationally inefficient, this method is suitable for basic distribution network monitoring. However, it struggles to handle complex grid conditions, such as voltage dips, harmonic interference, or sudden load changes, and is prone to false alarms. Especially in modern grids with extensive distributed generation, traditional methods cannot effectively distinguish between planned and fault-based outages.

[0004] Signal processing-based detection techniques utilize methods such as Fourier transform and wavelet analysis to analyze power signal characteristics. These methods excel at identifying transient faults and disturbances, capable of capturing signal anomalies at the microsecond level. However, their algorithmic complexity increases exponentially with the number of monitoring points, posing a computational bottleneck when implementing full-network monitoring. Furthermore, signal processing methods often have detection blind spots for slow-developing latent faults such as insulation degradation.

[0005] In recent years, artificial intelligence technology has brought breakthroughs to power outage detection. Machine learning models, trained on historical data, can identify complex fault patterns, and the application of deep learning, in particular, has significantly improved detection accuracy in image recognition and time series prediction. However, these methods rely on high-quality labeled data, while the scarcity of actual power grid fault samples limits the generalization ability of the models. Therefore, how to automatically and accurately detect power outages in the power system has become an urgent problem to be solved. Summary of the Invention

[0006] This invention provides a method, apparatus, electronic device, storage medium, and product for detecting power outages in power systems, in order to solve the problem that existing technologies cannot accurately detect power outages in power systems.

[0007] According to one aspect of the present invention, a power system outage detection method is provided, wherein the method includes: Acquire power time-series data and communication time-series data, and determine the time alignment confidence between the power time-series data and the communication time-series data through a self-attention mechanism; Determine the power grid equipment and the power supply correlation between the power grid equipment, construct a power grid topology model according to the power grid equipment and the power supply correlation using a graph neural network, and determine the spatial correlation strength based on the power grid topology model; The power outage probability of the power system is determined based on the time alignment confidence and the spatial correlation strength, and the power outage detection result is generated according to the power outage probability.

[0008] According to another aspect of the present invention, a power system outage detection device is provided, wherein the device comprises: The confidence determination module is used to acquire power time-series data and communication time-series data, and determine the time alignment confidence between the power time-series data and the communication time-series data through a self-attention mechanism. The correlation strength determination module is used to determine the power grid equipment and the power supply correlation between the power grid equipment. It constructs a power grid topology model according to the power grid equipment and the power supply correlation relationship through a graph neural network, and determines the spatial correlation strength based on the power grid topology model. The detection result determination module is used to determine the power outage probability of the power system based on the time alignment confidence and the spatial correlation strength, and generate the power outage detection result according to the power outage probability.

[0009] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the power system outage detection method according to any embodiment of the present invention.

[0010] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the power system outage detection method according to any embodiment of the present invention.

[0011] According to another aspect of the present invention, embodiments of the present invention also provide a computer program product, the computer program product including a computer program, which, when executed by a processor, implements the power system outage detection method of any embodiment of the present invention.

[0012] The technical solution of this invention acquires power time-series data and communication time-series data, determines the time alignment confidence of the power time-series data and communication time-series data through a self-attention mechanism, determines the power grid equipment and the power supply correlation between the power grid equipment, constructs a power grid topology model according to the power grid equipment and power supply correlation through a graph neural network, determines the spatial correlation strength based on the power grid topology model, determines the power outage probability of the power system based on the time alignment confidence and spatial correlation strength, and generates power outage detection results according to the power outage probability, thereby achieving accurate detection and early warning of power system outages. At the same time, by capturing the time correlation between power time-series data and communication time-series data, false judgments are reduced and the user experience is improved.

[0013] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a power system outage detection method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a power system outage detection method provided in Embodiment 2 of the present invention; Figure 3 This is a flowchart of a power system outage detection method provided in Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of the structure of a power system outage detection system according to Embodiment 3 of the present invention; Figure 5 This is a schematic diagram of the structure of a power system outage detection device according to Embodiment 4 of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device that implements the power system outage detection method according to an embodiment of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Example 1 Figure 1 This is a flowchart of a power system outage detection method according to Embodiment 1 of the present invention. This embodiment is applicable to situations involving power system outage detection. The method can be executed by a power system outage detection device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes: S110. Acquire power time-series data and communication time-series data, and determine the time alignment confidence between power time-series data and communication time-series data through a self-attention mechanism.

[0019] Power time-series data can be understood as the power grid operating status data continuously collected over time in a power system. Generally, power time-series data may include, but is not limited to, information such as voltage, current, voltage sag events, current surge events, and equipment load. In actual operation, power time-series data can be collected through a Supervisory Control and Data Acquisition (SCADA) system. In one embodiment, SCADA can collect power time-series data at a frequency of seconds. Communication time-series data refers to the communication status data continuously collected over time in the power system's supporting communication network. Generally, communication time-series data may include, but is not limited to, the optical modem offline rate sequence, i.e., the proportion of offline optical modem devices in a certain area to the total number of optical modems in that area. Generally, communication time-series data can be collected at a frequency of minutes. The self-attention mechanism is a feature learning mechanism based on the Transformer model. Time alignment confidence can be understood as the degree of correlation and matching between power time series data and communication time series data in the time dimension. Generally speaking, the value range of time alignment confidence can be 0-1. The closer the time alignment confidence is to 1, the stronger the temporal correlation between power time series data and communication time series data, and the more likely they are to correspond to the same power grid outage event. The closer the time alignment confidence is to 0, the weaker the correlation.

[0020] In this embodiment, power grid status data can be extracted from the SCADA system as power time-series data, and optical modem offline rate can be extracted from the power communication network monitoring system as communication time-series data. The power time-series data and communication time-series data are then encoded into vector forms recognizable by the self-attention mechanism. For example, the power time-series data can be encoded as a query vector, and the communication time-series data as a key vector and a value vector. The key vector is used to calculate the similarity with the query vector, and the value vector is used for subsequent weighted summation to obtain the association features. In one embodiment, positional encoding can also be added to the power time-series data and communication time-series data to preserve the temporal order information of the data and avoid misjudgments of temporal association due to data scrambling. The temporal association strength of the power time-series data and communication time-series data is calculated using the self-attention mechanism to obtain the attention weight. In actual operation, the query vector and key vector can be multiplied to obtain an initial similarity score representing the degree of temporal matching between the two types of data. The initial similarity score is scaled according to a preset scaling factor to obtain the target similarity score, for example, by scaling according to the square root of the key vector dimension. The scaled target similarity score is then Softmax normalized to obtain the attention weight. The time-series correlation features are obtained by weighted summation of the value vectors using attention weights, and then time-aligned confidence is obtained by normalizing the time-series correlation features through a fully connected layer. In actual operation, power event intensity weights can also be determined, attention weights can be adjusted, key faults can be reinforced, and noise can be suppressed. Power event intensity weights can be determined based on voltage-related parameters. In one embodiment, after acquiring power time-series data and communication time-series data, data preprocessing such as normalization, missing value imputation, and outlier removal can be performed on the power time-series data and communication time-series data, respectively.

[0021] S120. Determine the power grid equipment and the power supply correlation between the power grid equipment. Construct a power grid topology model according to the power grid equipment and power supply correlation using a graph neural network. Determine the spatial correlation strength based on the power grid topology model.

[0022] In this context, "power grid equipment" refers to various devices that constitute the power supply links of a power system, including substations, distribution transformers, distribution lines, and primary user nodes. "Power supply correlation" refers to the hierarchical power supply relationships between power grid equipment, including the primary line connection relationships, power supply direction, power supply capacity, and link priority. A Graph Neural Network (GNN) is a deep learning model for processing graph-structured data. A power grid topology model is a graph structure model constructed with power grid equipment as nodes and the power supply correlations between equipment as directed edges. Generally, each node can store the operating status information of the corresponding equipment, such as load, service life, rated voltage, and load capacity. Each directed edge stores the corresponding power supply topology relationship, such as power supply direction and weight. Spatial correlation strength is a scalar value used to characterize the degree of fault propagation correlation and power supply dependence between power grid equipment. It reflects the degree of fault propagation correlation between power grid equipment and downstream equipment and users. Generally, the spatial correlation strength ranges from 0 to 1. The closer the spatial correlation strength is to 1, the greater the impact of an anomaly in a certain device on its downstream equipment or users, and the higher the risk of power outage; the closer the spatial correlation strength is to 0, the smaller the correlation impact.

[0023] In this embodiment, all grid devices in the power system can be identified, their operating status information determined, and the current supply direction and line connection methods between them determined. Each grid device is treated as a node, and its operating status information, including operating parameters, years of operation, real-time load, and historical fault records, is bound to each node. The power supply relationships between grid devices are used as directed edges in a graph, with the direction of the directed edges consistent with the current supply direction. Each edge is bound with feature information, such as power supply capacity, link loss, and topology weight, forming a grid topology graph structure. In one embodiment, two edge types can be defined: for example, the power transmission edge from substation to transformer records the load carrying ratio, and the power supply affiliation edge from transformer to user group uses Boolean values ​​to mark direct power supply relationships. The grid topology graph structure is input into a graph neural network, and the operating status information such as load and operating delay of each grid device node is collected. Combined with the number of neighboring nodes, topology weights, and historical operating data, a risk score is calculated for each node. If the risk score exceeds a preset threshold, the node can be marked as a high-risk node, such as overloaded equipment or aging equipment. Then, through an attention mechanism, the importance of each neighboring node is dynamically assessed by comprehensively considering the real-time load of the node, its network topology location, and its historical operating status. The attention weight is increased for overloaded and high-risk devices, and decreased for normally operating devices, prioritizing the amplification of the impact of high-risk nodes. Based on the assigned attention weights, the load and offline rate of each node's neighboring nodes are weighted and aggregated. Through multi-layer feature fusion using a graph neural network, the spatial correlation strength of each power grid device node is quantified, completing the spatial modeling. After completing the spatial modeling, the graph neural network outputs the spatial correlation strength corresponding to each power grid device node.

[0024] S130. Determine the power outage probability of the power system based on the time alignment confidence and spatial correlation strength, and generate power outage detection results according to the power outage probability.

[0025] The power outage probability refers to the likelihood of a power outage event occurring in a certain area of ​​the power system. Generally, the power outage probability ranges from 0 to 1. The closer the power outage probability is to 1, the higher the probability of a power outage; the closer the power outage probability is to 0, the lower the probability of a power outage. The power outage detection result refers to specific information generated based on the power outage probability to guide power grid operation and maintenance. For example, the power outage detection result may include, but is not limited to, information such as power outage warning level and fault location results.

[0026] In this embodiment, the probability of a power outage can be determined by weighted fusion based on time-aligned confidence and spatial correlation strength. In actual operation, the weighting coefficients corresponding to time-aligned confidence and spatial correlation strength can be pre-set. For example, the weighting coefficient for time-aligned confidence can be set to 60%, and the weighting coefficient for spatial correlation strength can be set to 40%, etc. This is not limited and can be set according to business needs, as long as the sum of the two weighting coefficients is 1. Real-time meteorological data of the target area, such as rainfall intensity, wind speed, and lightning level, is then extracted from the meteorological monitoring system. This real-time meteorological data is quantified to obtain an environmental meteorological risk index; alternatively, environmental meteorological risk indices corresponding to different meteorological data can be extracted. Generally, the environmental meteorological risk index ranges from 0 to 1. A power outage threshold is determined based on the environmental meteorological risk index. Generally, the higher the meteorological risk index (e.g., during thunderstorms), the lower the power outage threshold, meaning a minor power outage triggers an alert; conversely, the lower the meteorological risk index (e.g., during clear weather), the higher the power outage threshold, which can filter out false alarms caused by normal fluctuations. In actual operation, the rules for adjusting the power outage judgment threshold can be pre-set. For example, different meteorological risk indices can be pre-determined for power outage judgment thresholds, and the corresponding power outage judgment thresholds can be matched accordingly. When the power outage probability is greater than or equal to the power outage judgment threshold, an advanced power outage warning is generated. When the power outage probability is lower than the power outage judgment threshold, a regular risk warning can be generated, prompting maintenance personnel to focus on high-risk equipment. In one embodiment, a preset low-risk threshold can also be set. When the power outage probability is lower than the preset low-risk threshold, a no-power outage risk detection result can be generated, eliminating the need to activate the warning and reducing ineffective maintenance costs. In one embodiment, while generating a fault warning, fault location results (based on spatial correlation strength to lock high-risk equipment and fault propagation paths) and fault root cause analysis results (such as overload of old equipment, voltage drop caused by lightning strikes, etc.) can also be output to guide maintenance personnel to quickly handle the situation. The power outage warning, fault location results, and fault root cause analysis results are used as the power outage detection results.

[0027] This invention, through acquiring power time-series data and communication time-series data, determines the time alignment confidence of the power time-series data and communication time-series data using a self-attention mechanism, identifies power grid equipment and the power supply correlation between power grid equipment, constructs a power grid topology model based on the power grid equipment and power supply correlation using a graph neural network, determines the spatial correlation strength based on the power grid topology model, determines the power outage probability of the power system based on the time alignment confidence and spatial correlation strength, and generates outage detection results according to the outage probability, thereby achieving accurate detection and early warning of power system outages. Simultaneously, by capturing the time correlation between power time-series data and communication time-series data, false alarms are reduced, improving the user experience.

[0028] Example 2 Figure 2 This is a flowchart of a power system outage detection method according to Embodiment 2 of the present invention. This embodiment is a further optimization and extension based on the above embodiments, and can be combined with various optional technical solutions in the above embodiments. Figure 2 As shown, the method includes: S210. The power time series in the monitoring and data acquisition system is used as power time series data, and the offline rate series is used as communication time series data.

[0029] Among them, the monitoring and data acquisition system is the core system in the power system used for real-time monitoring and acquisition of the power grid's operating status. It can achieve second-level acquisition, transmission, and storage of operating parameters of power grid equipment, and is the main data source for obtaining power time-series data, accurately capturing real-time changes in the power grid's operation. Power time series refers to a set of power grid operating data arranged in chronological order, which can include various dynamic parameters during the power grid's operation. Offline rate series refers to a sequence of continuously statistically analyzed data over time, showing the proportion of offline communication terminals (such as optical modems) to the total number of terminals within a certain area.

[0030] In one embodiment, a monitoring and data acquisition system can be activated, and a data acquisition frequency can be set to collect the power time series of power grid equipment within the target area in real time. In one embodiment, parameters such as voltage and current can be collected, and voltage drop events and current surge events can be identified. The time of occurrence of these events is recorded, and the collected power time series is used as power time series data. Through the power grid-connected communication network monitoring system, the number of offline and total communication terminals (optical modems) within the target area can be counted in real time, and the offline rate at each time point can be calculated to form an offline rate sequence as communication time series data.

[0031] S220. Use the power time series data as a query vector and map the communication time series data into a key vector and a value vector.

[0032] Among them, the query vector (Q) refers to the vector form obtained by encoding power time series data; the key vector (K) is the vector form obtained by mapping communication time series data; and the value vector (V) is also a vector form obtained by mapping communication time series data.

[0033] In this embodiment, power time-series data can be converted into query vectors, and communication time-series data can be converted into key vectors and value vectors. At the same time, positional encoding (PE) can be added to preserve the temporal order of the data and avoid the loss of time information.

[0034] S230. Based on the self-attention mechanism, the time alignment confidence between power time series data and communication time series data is determined by query vector, key vector and value vector.

[0035] In this embodiment, the query vector, key vector, and value vector can be input into a self-attention mechanism to calculate the time alignment confidence between power time series data and communication time series data.

[0036] In one embodiment, the time alignment confidence between power time-series data and communication time-series data is determined based on a self-attention mechanism using query vectors, key vectors, and value vectors, including: The dot product of the query vector and the key vector is determined as the initial similarity score. The initial similarity score is then scaled according to a preset scaling factor to obtain the target similarity score. The target similarity scores are normalized to obtain the attention weights. The value vectors are then weighted and summed according to the attention weights to obtain the time alignment confidence between the power time series data and the communication time series data.

[0037] In this embodiment, the query vector and key vector can be multiplied by a dot product to obtain the matching degree between the power time-series data and the communication time-series data at each time point, which serves as the initial similarity score. A larger dot product indicates a stronger correlation between the power data and the communication data at that time point. The initial similarity score is then scaled using a preset scaling factor to obtain the target similarity score. In actual operation, the square root of the dimension of the key vector can be used as the preset scaling factor, i.e. The scaled target similarity scores are then normalized to obtain attention weights. After normalization, the attention weights range from 0 to 1, and the sum of all attention weights is 1. Each attention weight represents the importance of the communication data at the corresponding time point to the current power anomaly event. Based on the obtained attention weights, the value vectors are weighted and summed to obtain the time alignment confidence between the power time series data and the communication time series data.

[0038] In one embodiment, after normalizing the target similarity score to obtain the attention weight, the method further includes: The absolute value of voltage change is determined based on power time-series data, and the ratio of the absolute value of voltage change to the rated voltage is determined as the voltage correlation factor. The product of the voltage correlation factor and the preset time location encoding scaling factor is determined as the target value, and the sine value of the target value is determined as the power event intensity weight. When the ratio of the absolute value of the voltage change to the rated voltage is greater than a preset threshold, the attention weight is updated according to the power event intensity weight.

[0039] In this embodiment, voltage change information can be extracted from power time-series data to determine the absolute value of the voltage change. That is, the absolute value of the difference between the actual voltage value and the rated voltage; calculate the absolute value of the voltage change and the rated voltage. The ratio of the voltage change to the rated voltage is used as the voltage correlation factor. A preset time position encoding scaling factor is determined, and the product of the voltage correlation factor and the preset time position encoding scaling factor is calculated as the target value. Then, a sine value is taken from the target value as the power event intensity weight. When the ratio of the absolute value of the voltage change to the rated voltage is greater than a preset threshold, it can be identified as a high-amplitude event. The attention weight and the power event intensity weight can be multiplied element-wise to obtain the updated attention weight; alternatively, the power event intensity weight can be used as an additive bias and incorporated into the attention weight; or, the power event intensity weight can be used as a multiplicative factor and multiplied with the attention weight. In actual operation, the preset time position encoding scaling factor can be... , where t is the time variable. This is the scaling factor, which varies with i and d. If d is a fixed parameter (such as a dimension), then 2i / d determines the scaling exponent. Generally, i is the dimension index, ranging from 0 to d / 2-1; d is the total dimension, usually the same as the dimension of the key vector.

[0040] In actual operation, preset thresholds may include, but are not limited to, 14%, 15%, and 16%. When the ratio of the absolute value of the voltage change to the rated voltage is greater than the preset threshold, it can be identified as a high-amplitude event (i.e., a precursor to a major power grid anomaly, such as a severe voltage drop); when the ratio is less than or equal to the preset threshold, it is identified as a small voltage fluctuation (noise). Using updated attention weights to perform weighted summation on the value vector can improve the accuracy of confidence and reduce false positives.

[0041] S240. Extract the power grid equipment and the power supply relationships between the power grid equipment. Construct a power grid topology graph structure with each power grid equipment as a node and the power supply relationships as directed edges.

[0042] Each node stores its corresponding operating status information, and each directed edge stores the power supply topology relationship.

[0043] The power grid topology graph refers to a graph structure model constructed with power grid equipment as nodes and power supply relationships as directed edges. It is a digital representation of the physical topology of the power grid, which can intuitively reflect the distribution and connection relationships of power grid equipment and the direction of power supply. The power supply topology relationship refers to the detailed information representing the power supply relationship between devices stored on each directed edge. For example, the power supply topology relationship can include information such as power supply direction, power supply capacity, link loss, and topology weight.

[0044] In this embodiment, all power grid devices can be sorted out and the power supply relationships between them can be determined. Each power grid device is treated as a node, a unique identifier is assigned to each node, the operating status information of the device is stored in the corresponding node, and a directed edge is established between each pair of device nodes with a power supply relationship according to the power supply relationship. The direction of the directed edge is consistent with the direction of the power supply current. The power supply topology relationship is stored in the corresponding directed edge to obtain the power grid topology graph structure.

[0045] S250. Spatial modeling of the power grid topology is performed using a graph neural network. The operating status information of each node's neighboring nodes is aggregated along the power supply association relationship to obtain the spatial association strength of each node.

[0046] In this context, a neighbor node refers to a node in the power grid topology that is directly connected to a certain node through a directed edge. That is, a device that directly supplies power to or receives power from a certain node. For example, the neighbor nodes of a distribution transformer include the upstream substation and the downstream user node.

[0047] In this embodiment, the power grid topology can be input into a graph neural network. The graph neural network collects the operational status information of each node's neighboring nodes along the power supply relationships (directed edge directions). Combining the number of neighboring nodes, topology weights, and historical operational data, a preliminary risk score is calculated for each node. At this point, overloaded and abnormal high-risk neighboring nodes can be identified. Through the attention mechanism of the graph neural network, the real-time load, network topology location (e.g., core hub equipment has higher weight), and historical operational status of each node can be comprehensively considered to dynamically assess the importance of each neighboring node. Attention weights are increased for abnormal high-risk neighboring nodes and decreased for normally operating neighboring nodes, prioritizing the amplification of the impact of high-risk nodes. In actual operation, the methods for increasing and decreasing attention weights can be pre-set. For example, attention weights for abnormal high-risk neighboring nodes and normally operating neighboring nodes can be pre-set. According to the assigned attention weights, the characteristics of each node and the operational status information of its neighboring nodes are weighted and aggregated. Through multi-layer iterative learning of the graph neural network, the characteristics of the node itself and its neighborhood are fused to uncover fault propagation patterns and power supply dependencies between devices, completing spatial modeling. After completing spatial modeling, the graph neural network outputs the spatial correlation strength corresponding to each power grid device.

[0048] S260. The power outage probability of the power system is obtained by weighted fusion of time alignment confidence and spatial correlation strength.

[0049] In this embodiment, the weighting coefficients corresponding to the time alignment confidence and spatial correlation strength can be determined, and the power outage probability of the power system can be calculated using a weighted summation formula.

[0050] S270. Obtain the environmental meteorological risk index and determine the power outage judgment threshold based on the environmental meteorological risk index.

[0051] The environmental meteorological risk index refers to the data used to dynamically adjust the power outage judgment threshold. It can be quantified based on real-time meteorological data (such as thunderstorms, strong winds, and high temperatures); alternatively, corresponding environmental meteorological risk indices can be preset based on different real-time meteorological data. The higher the meteorological risk, the more lenient the threshold, avoiding the omission of sudden faults; the lower the meteorological risk, the more stringent the threshold, reducing false alarms. The power outage judgment threshold refers to the critical value used to determine whether the probability of a power outage reaches the warning standard, and it is dynamically determined by the environmental meteorological risk index, rather than a fixed value.

[0052] In this embodiment, meteorological data of the target area can be acquired in real time, including key meteorological parameters such as rainfall intensity, wind speed, lightning level, and temperature. A preset quantification formula is used to convert the real-time meteorological data into an environmental meteorological risk index. For example, when the lightning level is high and the wind speed is high, the environmental meteorological risk index value is close to 1; when the weather is clear, the environmental meteorological risk index value is close to 0. By pre-setting a mapping rule between the environmental meteorological risk index and the power outage judgment threshold, the power outage judgment threshold corresponding to different environmental meteorological risk indices is determined. For example, when the environmental meteorological risk index ≥ 0.7 (severe weather), the power outage judgment threshold is set to 0.5; when the environmental meteorological risk index ≤ 0.3 (clear weather), the power outage judgment threshold is set to 0.8; intermediate indices correspond to intermediate thresholds, and so on. In one embodiment, the power outage judgment threshold can also be two values, namely a high-risk power outage judgment threshold and a low-risk power outage judgment threshold.

[0053] S280. Determine the comparison result between the power outage probability and the power outage judgment threshold, and determine the power outage detection result according to the comparison result.

[0054] In this embodiment, the probability of a power outage can be compared with a power outage determination threshold to determine the risk level to which the probability of the power outage belongs. When there is one high-risk power outage determination threshold and one low-risk power outage determination threshold, it can be determined that a power outage probability ≥ high-risk power outage determination threshold is high-risk; a power outage determination threshold > power outage probability ≥ low-risk power outage determination threshold is medium-risk; and a power outage probability < low-risk power outage determination threshold is medium-risk. When there is only one power outage determination threshold, the comparison result can be divided into two risk levels. Different power outage detection results can be given according to different risk levels. For example, when it is determined to be high-risk, advanced power outage early warning information can be generated, the warning area can be clearly defined, and the high-risk equipment and fault propagation path can be locked by combining spatial correlation strength, and the fault location result can be output; combined with time alignment confidence and abnormal power events (including voltage drop events and current surge events, etc.), the root cause of the fault can be analyzed, such as overload of old equipment or voltage drop caused by lightning strikes. When a medium-risk condition is identified, a routine risk warning can be generated, prompting maintenance personnel to focus on high-risk equipment, such as overloaded and aging equipment, to strengthen real-time monitoring and prevent the fault from escalating into a power outage. When a low-risk condition is identified, a no-outage-risk detection result is generated, indicating that the current power grid is operating normally and the probability of an outage is low, possibly due to normal voltage fluctuations or occasional communication downtime, thus eliminating the need to activate a warning and reducing unnecessary maintenance costs. In one embodiment, the generated outage detection result can also be transmitted to the power grid operation and maintenance monitoring platform and simultaneously pushed to the maintenance personnel's terminals, ensuring that maintenance personnel receive information promptly and respond quickly.

[0055] This invention, in its embodiments, collects power time series data from a monitoring and data acquisition system as power time series data and offline rate data as communication time series data. Power time series data is used as a query vector, and communication time series data is mapped as key and value vectors. Based on a self-attention mechanism, the time alignment confidence of power and communication time series data is determined using the query vector, key vector, and value vector. Power supply correlation relationships between power grid devices are extracted. A power grid topology is constructed with each device as a node and the power supply correlation relationships as directed edges. A graph neural network is used to spatially model the power grid topology, aggregating the operating status information of each node's neighboring nodes along the power supply correlation relationships to obtain the spatial correlation strength of each node. The time alignment confidence and spatial correlation strength are weighted and fused to obtain the power outage probability of the power system. An environmental meteorological risk index is obtained, and a power outage judgment threshold is determined based on the environmental meteorological risk index. The comparison result between the power outage probability and the power outage judgment threshold is determined, and the power outage detection result is determined according to the comparison result. This achieves power outage detection by combining communication time series data and power time series data, resulting in a more comprehensive power outage probability calculation. Simultaneously, the use of dynamic meteorological thresholds makes the early warning strategy more intelligent.

[0056] Example 3 Figure 3 This is a flowchart of a power system outage detection method according to Embodiment 3 of the present invention. In this embodiment, power SCADA data is used as power time-series data; the power data at the second level is the power SCADA data. Taking the offline rate of the communication optical modem as communication time-series data as an example, the communication data at the minute level is the communication optical modem offline rate. This provides a further explanation of a power system outage detection method. Figure 3 As shown, the method includes: This embodiment focuses on the accurate detection and early warning of power outage events in power systems. By integrating multi-dimensional data such as power SCADA data, optical modem offline rate (note: the optical modem offline rate refers to the ratio of optical network terminal equipment in a specific area that loses network connection within a given time period), and power grid topology, a highly efficient and intelligent power outage analysis system is constructed. Currently, optical modems on the operator's side have already developed second-level real-time alarm capabilities. When an optical network terminal equipment experiences a power outage, the optical modem fault system will generate a second-level alarm. By analyzing the signal characteristics of batch optical modem faults, it can significantly assist in the determination of power system outages, enabling accurate power outage alarms at the building and user levels. First, based on the Transformer self-attention mechanism, dynamic time alignment is achieved between second-level power events (power SCADA data) and minute-level communication data (optical modem offline rate). This automatically uncovers the temporal correlation between voltage drops and sudden increases in optical modem offline rates, providing direct evidence for fault determination. Subsequently, a Graph Neural Network (GNN) is introduced to spatially model the physical connections of the power grid. Through topological analysis of node (equipment / user group) attributes and edges (power supply relationships), the cascading propagation path of faults is simulated, accurately locating high-risk equipment (such as high-load transformers). Furthermore, a weighted fusion decision model is constructed by integrating multi-dimensional features such as time-aligned confidence, spatial correlation strength, and meteorological risk index (determined through environmental factors). An innovative dynamic threshold mechanism is adopted to adaptively adjust the judgment criteria under both normal scenarios and extreme weather conditions. Finally, the probability of power outages is quantified using the Sigmoid function for fault determination, outputting highly reliable fault warnings and root cause analysis results, providing intelligent decision support for proactive power grid defense.

[0057] In one embodiment, Figure 4 This is a schematic diagram of a power system outage detection system according to Embodiment 3 of the present invention. Figure 4As shown, the power system outage detection system includes a Transformer time alignment module, a Graph Neural Network (GNN) spatial modeling module, and a dynamic decision-making module. The Transformer time alignment module, based on a Transformer-GNN fusion architecture, first uses a self-attention mechanism to dynamically align second-level power data with minute-level communication data, automatically capturing the temporal correlation between voltage drops and optical modem offline rates. Then, the GNN spatial modeling module uses GNN to model the power grid topology, simulate fault spatial propagation paths, and accurately locate key risk points such as high-load transformers. Finally, the dynamic decision-making module integrates multi-dimensional features such as temporal confidence, spatial correlation strength, and meteorological factors, and uses a dynamic threshold mechanism to adaptively generate outage probability predictions, providing intelligent decision support for proactive power grid defense. In one embodiment, the specific process of the power system outage detection system executing the power system outage detection method is shown below.

[0058] The first step is the Transformer time alignment module.

[0059] The core challenge in data fusion between power systems and communication networks lies in achieving accurate correlation analysis of multi-source heterogeneous time-series data. The main difficulties are threefold: First, significant timescale differences exist at the data acquisition level. Power SCADA systems provide second-level precision power grid status monitoring data, such as voltage drops and current surges, while communication optical modems can only output minute-level offline rate statistics. This mismatch in time-series resolution hinders direct data correlation. Second, at the analysis method level, traditional alignment mechanisms based on fixed time windows, such as 5 minutes before and after an event, are ill-suited for complex fault scenarios, especially for sudden faults like lightning strikes, where the resulting voltage disturbances and communication interruptions have uncertain delay effects. Finally, at the application scenario level, the increasing penetration of distributed power sources in modern power grids makes outage events more complex and variable, necessitating the establishment of dynamic correlation models that can adapt to different fault types, such as transient and progressive faults. The Transformer time alignment module achieves dynamic time-series alignment of power and communication data through a self-attention mechanism, breaking through the limitations of fixed time windows and accurately capturing cross-modal correlation features.

[0060] (a) The core role of the Transformer self-attention mechanism.

[0061] This embodiment innovatively utilizes the self-attention mechanism of Transformer to achieve dynamic temporal alignment. Through a unique encoding and association learning approach, this method supports deep analysis of cross-modal data, enabling the capture of the intrinsic connections and dynamic temporal changes between different modalities.

[0062] (1) Data input: In terms of data input, it covers time series data in two dimensions: power and communication. The power time series data records the power grid status information with second-level precision, including key indicators such as voltage drop and current change. The communication time series data records the optical modem offline rate data with minute-level statistical period.

[0063] (2) Cross-modal temporal coding and adaptive association learning: In the cross-modal temporal coding stage, this embodiment encodes the power event sequence (e.g., the moment of voltage drop) into a query vector (Q), and simultaneously maps the minute-level offline rate sequence of the optical modem into a key vector (K, V). Based on this, the model enters the adaptive association learning stage, automatically mining the delay pattern between power events and communication indicators by calculating the attention weight between the query vector Q and the key vector K. This process does not require a pre-set fixed window, and the model can autonomously discover hidden temporal association patterns such as "the offline rate trend of the optical modem in the following 3 minutes should be given special attention after the voltage drop event".

[0064] Dynamic alignment mechanism, query-key-value mapping: ; .in, It is a linear transformation layer (fully connected layer) that maps the combined features into a query vector of fixed dimensions; The location is encoded as the i-th time point in the power time series data. Add location information, retaining the chronological order; For feature embedding, the first power time series data Original features at each time point Embedding vectors (e.g., voltage, current, voltage sag amplitude). Self-attention weights are calculated as follows: .

[0065] The input vector includes: Q, the Query vector, representing the information currently being analyzed; K, the Key vector, representing reference information used for matching; and V, the Value vector, representing the actual information content. It is the dimension of the key vector, used to scale the dot product result. This is an optional event bias term used to adjust attention weights. Dot product similarity calculation includes: It calculates the dot product of the query vector Q and the key vector K to obtain the original similarity score. The dot product result reflects the degree of matching between Q and K.

[0066] Scaling: That is, divide the dot product result by The purpose is to prevent the dot product result from becoming too large, which could lead to gradient vanishing or exploding. Scaling factor It balances the impact of dimension on the dot product result.

[0067] Event Bias: Optional bias item This can introduce additional prior information or event-related adjustments. For example, in time series data, It can represent a timestamp or an event type bias.

[0068] Softmax normalization transforms the scaled similarity scores into a probability distribution by applying the softmax() function. Attention weights represent the importance of each V vector to the current query Q.

[0069] Weighted summation: The attention weights output by the softmax function are multiplied by the value vector V to obtain the final attention output. The result is a weighted sum of V, with the weights determined by the similarity between Q and K.

[0070] (3) Multi-type fault identification.

[0071] The model proposed in this invention possesses powerful multi-type fault identification capabilities. For different types of faults, the model exhibits different attention weight distribution characteristics. For example, when a sudden fault (such as a lightning strike) occurs, the model captures the phenomenon of a sharp increase in the offline rate of the optical modem within one minute after a voltage drop; while a gradual fault (such as equipment aging) manifests as a gradual increase in the offline rate within 30 minutes during a slow voltage drop. In traditional methods, small voltage fluctuations and major voltage drops are treated equally, leading to insufficient model response to critical events. Therefore, this invention introduces power event intensity weights to enhance the time-point weights of high-amplitude events (such as voltage drops >15%), thereby strengthening the model's sensitivity to critical events.

[0072] (ii) Design of power event intensity weights.

[0073] Traditional methods for processing power data suffer from insensitivity to event magnitude, leading to an inability of the model to accurately capture key information and affecting fault identification. Introducing power event intensity weights aims to enable the model to more accurately focus on critical time periods that are of significant value to fault analysis by appropriately adjusting these weights.

[0074] (1) Key event enhancement: High-amplitude events (such as voltage drops >15%) are given higher time point weights. The attention score is adjusted using a formula to focus the model on strongly correlated key time periods. The formula is as follows: .

[0075] 1) Function definition: It is a function of time t and the complex exponent 2i, which may represent some kind of positional encoding or signal modulation. i usually represents the imaginary unit, but here 2i may be an index or identifier of some parameter (this needs to be confirmed in the context).

[0076] 2) Sin() is a sine function. The input is an angle (in radians), and the output is the sine value of that angle.

[0077] 3) Preset time position encoding scaling factor In this context, t represents the time variable. 2i / d is the scaling factor, which varies with i and d. If d is a fixed parameter (such as dimension), then 2i / d determines the scaling exponent. This form is common in positional encoding (such as Positional Encoding in Transformer) to generate sine waves of different frequencies.

[0078] 4) Voltage-related factors ;in, This represents the absolute value of the voltage change, indicating the amplitude of the voltage fluctuation. This is the rated voltage (nominal voltage), used to normalize voltage variation. This value reflects the proportion of voltage fluctuation relative to the rated voltage and may be used to modulate the frequency or amplitude of a sine wave.

[0079] (2) Noise suppression. In order to improve the model's ability to distinguish real faults, this invention also constructs a noise suppression mechanism. For noise data such as small voltage fluctuations, since their value for fault discrimination is low, reducing their weight effectively weakens the possible accidental correlation between these noise data and communication indicators.

[0080] The second step is the Graph Neural Network (GNN) spatial modeling module.

[0081] Traditional power grid fault detection typically relies on threshold rules for individual device states (e.g., "alarm when transformer load > 90%)", which can only identify anomalies in single devices and cannot analyze how faults propagate throughout the power grid. For example, even if transformer A is detected to be overloaded, it cannot predict whether this will lead to overload of downstream line B or power outage for user group C. Furthermore, traditional methods struggle to dynamically adjust device risk weights; for instance, older transformers may be more prone to failure due to long-term operation, but static rules cannot provide specific consideration for them.

[0082] Graph Neural Networks (GNNs) model the physical connections of the power grid (such as the power supply path of "substation → transformer → user group") to simulate the cascading propagation process of faults. For example, if transformer A fails, GNN can infer that its power supply line B may be overloaded, further affecting user group C. Simultaneously, GNNs can automatically learn equipment risk weights; for instance, old transformers that have been in operation for more than 15 years receive 50% more attention in the calculation, making the model more focused on highly vulnerable nodes, thereby accurately locating the source of the fault and predicting the propagation path.

[0083] Graph Neural Networks (GNNs) model fault propagation by understanding the physical connections of the power grid, overcoming the limitations of traditional single-device detection. Their core principle lies in abstracting the power grid as a graph structure composed of nodes (equipment / user groups) and edges (power supply relationships): nodes contain key equipment attributes (such as substation voltage levels, transformer lifespan, and user group offline rates), while edges describe the power supply relationships between equipment (such as load ratios or Boolean power supply identifiers). This structured representation can explicitly capture the cascading effects of faults in the power grid topology; for example, a transformer fault can trigger downstream line overloads and user power outages through power supply edges.

[0084] GNN constructs a power grid topology model using two data structures: nodes and edges. At the node level, the system abstracts power grid entities such as substations, distribution transformers, and user groups as graph nodes. Each node stores its key operating parameters; for example, substations record voltage levels and load rates, distribution transformers include their service life and three-phase balance, and user groups are represented by statistics such as optical modem offline rates and electricity consumption patterns. Power grid nodes (power grid equipment) are shown in Table 1.

[0085] Table 1 Power Grid Nodes In terms of edge relationship modeling, the system accurately represents the power supply connection relationships between devices through directed edges. Two main edge types are defined: the power transmission edge from substation to transformer records the load carrying ratio, and the power supply affiliation edge from transformer to user group uses Boolean values ​​to mark direct power supply relationships. This structured edge connection method fully presents the transmission path of electrical energy from the transmission network to the distribution network and then to the end user. The power supply relationships between grid devices are shown in Table 2.

[0086] Table 2 Power supply relationships between power grid equipment Spatial modeling of GNNs is achieved through a dynamic message passing mechanism, which includes three key steps.

[0087] (i) Neighbor Status Aggregation: The device monitors the load, latency, and other statuses of neighboring nodes in real time. If a neighbor overload is detected, the system calculates a risk score based on the number of neighbors, topology weights, and historical data. If the score exceeds a threshold, it is marked as high risk.

[0088] (ii) Attention Weight Calculation: The system uses an attention mechanism to dynamically assess the importance of neighboring nodes and automatically identify critical devices. Overloaded devices have significantly increased weights, while normally operating devices have lower weights. This mechanism comprehensively considers real-time load, network topology location, and historical operating status, prioritizing the amplification of the impact of high-risk nodes to ensure more accurate risk assessment, thereby optimizing early warning strategies and improving the overall reliability of the system.

[0089] (III) Comprehensive Risk Assessment: The system calculates the spatial correlation strength index by weighted aggregation of signals such as load and offline rate of neighboring nodes. When critical equipment simultaneously experiences abnormalities such as high load and user offline, a high correlation risk value will be output. If this value exceeds the set threshold, the fault source will be automatically located.

[0090] This method particularly emphasizes the dynamic perception of high-risk equipment. For example, old transformers that have been in operation for over 15 years will have their weight increased by 50% in the model, enabling the GNN to proactively identify vulnerable points. Through this end-to-end graph structure learning, the propagation path and impact range of power grid faults can be quantitatively analyzed, providing a basis for proactive defense decisions.

[0091] The third step is the dynamic decision-making module.

[0092] Dynamic decision-making achieves highly reliable power outage event determination by integrating temporal, spatial, and environmental characteristics and combining them with adaptive threshold adjustments. Its core is similar to an "intelligent scoring system," employing a weighted fusion and dynamic threshold strategy to ensure accurate decision-making in different scenarios.

[0093] (I) Multi-dimensional Feature Input: The dynamic decision-making system achieves intelligent judgment of power outage events by integrating spatiotemporal features. Its core inputs include two key dimensions: time-aligned confidence, which is used to accurately assess the temporal correlation between power system anomalies and communication network interruptions, and establishes direct evidence of causal relationships through cross-modal analysis; and spatial correlation strength, which quantifies fault propagation paths based on the power grid topology and identifies potential impact ranges and key nodes through the connection relationships between devices. Among them, time-aligned confidence focuses on verifying the temporal logic of the event occurrence to ensure the timeliness and accuracy of fault judgment; and spatial correlation strength reveals the diffusion law of faults in the power grid, providing structured analysis for locating fault sources and predicting the scope of impact. These two dimensions of features mutually verify and complement each other, together constituting the spatiotemporal analysis framework for power outage event judgment.

[0094] (II) Weighted Fusion and Dynamic Judgment: This system adopts a dynamic weighted fusion mechanism, emphasizing the decision-making value of different feature dimensions: time alignment features are assigned the highest weight of 60%, as they directly reflect key temporal evidence of power anomalies and communication interruptions, serving as the core basis for determining power outages; spatial correlation features are allocated a weight of 40%, verifying fault propagation paths through power grid topology analysis and providing auxiliary support for locating the fault source. The weighted fusion results are transformed into power outage probabilities in the 0-1 range using the Sigmoid function, ensuring both the probabilistic expression of the decision results and maintaining the interpretability of the model output. This weight allocation scheme fully considers the information value of different feature dimensions, with the time dimension serving as the primary judgment basis and the spatial dimension as a supplementary verification, forming a hierarchical decision-making logic.

[0095] (III) Scenario Adaptive Optimization: This system employs an intelligent dynamic threshold adjustment mechanism, automatically optimizing judgment criteria through real-time operational status assessment: a higher judgment threshold is used under standard operating conditions to ensure accuracy during routine operation; when abnormal operating condition signals are detected, the judgment threshold is automatically lowered to improve system response sensitivity. This adaptive adjustment strategy ensures judgment accuracy under basic scenarios while promptly capturing potential risks under special operating conditions. Through the organic combination of operational status perception and decision-making logic, optimal decision-making balance is achieved under different scenarios. This mechanism specifically considers the dynamic changes in equipment operating status, enabling the system to possess progressive early warning capabilities and providing intelligent, tiered protection for the safe operation of the power grid.

[0096] The fourth step is system implementation and verification.

[0097] This system employs a modular architecture to achieve efficient fusion and collaborative analysis of multi-source data. It utilizes a distributed computing framework to process massive amounts of power SCADA data and communication optical modem data, ensuring real-time performance and reliability. It deeply integrates Transformer temporal modeling capabilities with GNN spatial modeling capabilities; the time alignment module achieves millisecond-level event matching, while the spatial modeling module supports real-time calculation of power grid topology with tens of millions of nodes.

[0098] In this embodiment, a cross-modal time alignment method based on self-attention mechanism, its dynamic weight adjustment strategy and noise suppression mechanism; topology sensing algorithm in power grid spatial modeling, especially equipment risk dynamic assessment system and fault propagation path prediction method; these innovations together constitute a full-chain technical solution covering time series analysis, spatial modeling and intelligent decision-making, which has significant non-obviousness and technical effect improvement compared with the existing technology.

[0099] Example 4 Figure 5 This is a schematic diagram of the structure of a power system outage detection device according to Embodiment 4 of the present invention. Figure 5As shown, the device includes: a confidence level determination module 51, a correlation strength determination module 52, and a detection result determination module 53.

[0100] The confidence determination module 51 is used to acquire power time-series data and communication time-series data, and to determine the time alignment confidence between the power time-series data and the communication time-series data through a self-attention mechanism.

[0101] The correlation strength determination module 52 is used to determine the power grid equipment and the power supply correlation between the power grid equipment. It constructs a power grid topology model according to the power grid equipment and the power supply correlation through a graph neural network, and determines the spatial correlation strength based on the power grid topology model.

[0102] The detection result determination module 53 is used to determine the power outage probability of the power system based on the time alignment confidence and spatial correlation strength, and generate the power outage detection result according to the power outage probability.

[0103] The technical solution of this invention acquires power time-series data and communication time-series data through a confidence determination module, determines the time alignment confidence of the power time-series data and communication time-series data through a self-attention mechanism, determines the power grid equipment and the power supply correlation between the power grid equipment through a correlation strength determination module, constructs a power grid topology model according to the power grid equipment and the power supply correlation through a graph neural network, determines the spatial correlation strength according to the power grid topology model, determines the power outage probability of the power system according to the time alignment confidence and spatial correlation strength, and generates power outage detection results according to the power outage probability, thereby achieving accurate detection and early warning of power system outages. At the same time, by capturing the time correlation of power time-series data and communication time-series data, false judgments are reduced and the user experience is improved.

[0104] In one embodiment, the confidence determination module 51 includes: The data acquisition unit is used to collect power time series data from the monitoring and data acquisition system as power time series data, and to collect offline rate series data as communication time series data. The data mapping unit is used to map power time-series data into query vectors and communication time-series data into key vectors and value vectors. The confidence determination unit is used to determine the time alignment confidence between power time series data and communication time series data based on a self-attention mechanism using query vectors, key vectors, and value vectors.

[0105] In one embodiment, the confidence determination unit is specifically used for: The dot product of the query vector and the key vector is determined as the initial similarity score. The initial similarity score is then scaled according to a preset scaling factor to obtain the target similarity score. The target similarity scores are normalized to obtain the attention weights. The value vectors are then weighted and summed according to the attention weights to obtain the time alignment confidence between the power time series data and the communication time series data.

[0106] In one embodiment, the power system outage detection device further includes: The voltage information acquisition module is used to determine the absolute value of voltage change according to power time series data, and to determine the ratio of the absolute value of voltage change to the rated voltage as a voltage correlation factor. The weight determination module is used to determine the product of the voltage-related factor and the preset time position encoding scaling factor as the target value, and to determine the sine value of the target value as the power event intensity weight. The weight update module is used to update the attention weight according to the power event intensity weight when the ratio of the absolute value of the voltage change to the rated voltage is greater than a preset threshold.

[0107] In one embodiment, the association strength determination module 52 includes: The topology graph construction unit is used to extract power grid equipment and the power supply relationships between power grid equipment. It constructs a power grid topology graph structure with each power grid equipment as a node and the power supply relationships as directed edges. Each node stores the corresponding operating status information, and each directed edge stores the power supply topology relationship. The association strength determination unit is used to spatially model the power grid topology using a graph neural network, aggregate the operating status information of each node's neighboring nodes along the power supply association relationship, and obtain the spatial association strength of each node.

[0108] In one embodiment, the detection result determination module 53 includes: The probability determination unit is used to obtain the power outage probability of the power system by weighted fusion of time alignment confidence and spatial correlation strength; The threshold determination unit is used to acquire the environmental meteorological risk index and determine the power outage judgment threshold based on the environmental meteorological risk index. The result determination unit is used to determine the comparison result between the power outage probability and the power outage judgment threshold, and to determine the power outage detection result according to the comparison result.

[0109] The power system outage detection device provided in the embodiments of the present invention can execute the power system outage detection method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0110] Example 5 Figure 6This is a schematic diagram of the structure of an electronic device implementing the power system outage detection method of this invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0111] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0112] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0113] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as power system outage detection methods.

[0114] In some embodiments, the power system outage detection method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the power system outage detection method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the power system outage detection method by any other suitable means (e.g., by means of firmware).

[0115] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0116] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0117] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0118] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0119] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0120] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0121] In one embodiment, the present invention further includes a computer program product, which includes a computer program that, when executed by a processor, implements the power system outage detection method of any embodiment of the present invention.

[0122] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0123] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0124] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A power system outage detection method, characterized by, include: Acquire power time-series data and communication time-series data, and determine the time alignment confidence between the power time-series data and the communication time-series data through a self-attention mechanism; Determine the power grid equipment and the power supply correlation between the power grid equipment, construct a power grid topology model according to the power grid equipment and the power supply correlation using a graph neural network, and determine the spatial correlation strength based on the power grid topology model; The power outage probability of the power system is determined based on the time alignment confidence and the spatial correlation strength, and the power outage detection result is generated according to the power outage probability.

2. The method of claim 1, wherein, The acquisition of power time-series data and communication time-series data, and the determination of the time alignment confidence between the power time-series data and the communication time-series data through a self-attention mechanism, include: The power time series data collected from the monitoring and data acquisition system is used as power time series data, and the offline rate series data is used as communication time series data. The power time series data is used as a query vector, and the communication time series data is mapped as a key vector and a value vector. Based on the self-attention mechanism, the time alignment confidence between the power time series data and the communication time series data is determined using the query vector, the key vector, and the value vector.

3. The method of claim 2, wherein, The method of determining the time alignment confidence between power time-series data and communication time-series data based on the self-attention mechanism using the query vector, the key vector, and the value vector includes: The dot product of the query vector and the key vector is determined as the initial similarity score, and the initial similarity score is scaled according to a preset scaling factor to obtain the target similarity score. The target similarity score is normalized to obtain the attention weight, and the value vector is weighted and summed according to the attention weight to obtain the time alignment confidence between the power time series data and the communication time series data.

4. The method of claim 3, wherein, After normalizing the target similarity score to obtain the attention weight, the method further includes: The absolute value of voltage change is determined based on power time-series data, and the ratio of the absolute value of voltage change to the rated voltage is determined as a voltage correlation factor. The product of the voltage correlation factor and the preset time position encoding scaling factor is determined as the target value, and the sine value of the target value is determined as the power event intensity weight. When the ratio of the absolute value of the voltage change to the rated voltage is greater than a preset threshold, the attention weight is updated according to the power event intensity weight.

5. The method of claim 1, wherein, The process of determining power grid equipment and the power supply relationships between power grid equipment involves constructing a power grid topology model using a graph neural network based on the power grid equipment and the power supply relationships, and determining the spatial correlation strength based on the power grid topology model, including: Extract the power grid equipment and the power supply relationships between the power grid equipment, and construct a power grid topology graph structure with each power grid equipment as a node and the power supply relationships as directed edges; wherein, each node stores the corresponding operating status information, and each directed edge stores the power supply topology relationship; The power grid topology is spatially modeled using a graph neural network. The operating status information of each node's neighboring nodes is aggregated along the power supply association relationship to obtain the spatial association strength of each node.

6. The method of claim 1, wherein, The step of determining the power system outage probability based on the time alignment confidence and the spatial correlation strength, and generating outage detection results according to the outage probability, includes: The power outage probability of the power system is obtained by weighted fusion of time alignment confidence and spatial correlation strength. Obtain the environmental meteorological risk index, and determine the power outage judgment threshold based on the environmental meteorological risk index; The comparison result between the power outage probability and the power outage determination threshold is determined, and the power outage detection result is determined according to the comparison result.

7. A power system outage detection apparatus characterized by comprising: include: The confidence determination module is used to acquire power time-series data and communication time-series data, and determine the time alignment confidence between the power time-series data and the communication time-series data through a self-attention mechanism. The correlation strength determination module is used to determine the power grid equipment and the power supply correlation between the power grid equipment. It constructs a power grid topology model according to the power grid equipment and the power supply correlation relationship through a graph neural network, and determines the spatial correlation strength based on the power grid topology model. The detection result determination module is used to determine the power outage probability of the power system based on the time alignment confidence and the spatial correlation strength, and generate the power outage detection result according to the power outage probability.

8. An electronic device, comprising: The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the power system outage detection method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the power system outage detection method according to any one of claims 1-6.

10. A computer program product, characterised in that, The computer program product includes a computer program that, when executed by a processor, implements the power system outage detection method according to any one of claims 1-6.

Images