Operation alarm method and system of electric power system

Abstract

The invention discloses an operation alarm method and system for a power system, and belongs to the technical field of power system monitoring, and the method comprises the steps: deploying a distributed intelligent sensing network to collect data in a layered manner, and extracting basic physical characteristics; semantic annotation and hierarchical credibility evaluation are carried out on the basic physical features to generate a credible physical feature set; constructing a semantic causal graph and optimizing an edge weight through disturbance simulation to extract a dynamic risk factor; constructing a layered state vector and a dynamic health threshold vector, and comparing the layered state vector and the dynamic health threshold vector to generate an operation alarm state; and executing attribution analysis according to the operation alarm state to generate a threshold calibration signal for feedback. According to the method, the technical scheme of combining multi-source data fusion, hierarchical credibility evaluation, causal risk modeling and dynamic threshold adaptive calibration is adopted, so that accurate identification and prospective early warning of potential risks of the power system can be realized, and the accuracy and adaptive capability of alarm decision making are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and in particular to an operation alarm method and system for a power system. Background Art

[0002] The power system is the lifeblood of national energy security and the critical infrastructure upon which modern society depends. Real-time monitoring and alerting of its operational status are crucial for ensuring the safe, stable, and economical operation of the power grid. The operational alert system aims to detect potential faults, anomalies, or instabilities by continuously measuring and analyzing the operating parameters of various devices in the power grid. This provides decision-making support to dispatchers and operators, preventing the occurrence and escalation of faults.

[0003] Existing power system operation alarm methods typically rely on monitoring systems deployed at substations or control centers to periodically collect key electrical measurement data, such as voltage, current, frequency, and power. This collected data is transmitted to the master station backend, where the real-time measurements are compared with pre-defined fixed thresholds or logic rules. Alarm signals are triggered when the measured values ​​exceed the specified range.

[0004] However, the above-mentioned existing technologies have obvious deficiencies in practical applications. First, their data sources are relatively single, and there is a lack of effective evaluation mechanisms for the quality and reliability of the data itself. They are easily interfered with by sensor errors or communication problems, resulting in a high false alarm rate of alarms. Secondly, alarm judgments are mainly based on static, pre-set thresholds, which cannot dynamically adapt to changes in the operation mode of the power grid or the evolution of the health status of the equipment. They are slow to respond to slow-developing potential failures or complex chain risks. In addition, traditional methods often process data from each measuring point in isolation, and lack the ability to model and analyze the mutual influence and risk transmission relationship between equipment. Summary of the Invention

[0005] To solve the above problems, the present invention provides an operation alarm method and system for an electric power system. It adopts a technical solution that combines multi-source data fusion, hierarchical credibility assessment, causal risk modeling and dynamic threshold adaptive calibration, which can realize accurate identification and forward-looking warning of potential risks in the electric power system, and improve the accuracy and adaptability of alarm decision-making.

[0006] The above objectives can be achieved through the following solutions: A method for operating alarm of a power system, comprising: deploying a distributed intelligent sensing network on physical entities of the power system, the distributed intelligent sensing network switching between different trigger modes, collecting electrical measurement data, environmental state data and topological relationship data in layers, pre-processing and physical feature extraction are performed on each layer of data to obtain basic physical features; performing feature semantic annotation on the basic physical features, associating the physical entities with the topological relationship data, and performing hierarchical credibility evaluation on the basic physical features to quantify data source reliability and generate a set of credible physical features; based on the set of credible physical features and the topological relationship data, constructing and initializing a semantic causal graph, applying multi-source disturbance simulation device anomalies and environmental mutations on the semantic causal graph, using adaptive edge weight optimization to iteratively update the causal edge weight according to the node influence degree, and extracting dynamic risk factors; constructing the basic physical features, the set of credible physical features and the dynamic risk factors into a hierarchical state vector, constructing a dynamic health threshold vector according to the topological relationship data, comparing the hierarchical state vector with the dynamic health threshold vector, and generating an operating alarm state; according to the operating alarm state, performing attribution analysis and parameter tracing to generate a threshold calibration signal for adjusting the dynamic health threshold vector.

[0007] Optionally, the obtaining of the basic physical features comprises: collecting low-frequency aggregated data in a baseline monitoring mode by the distributed intelligent sensing network, and collecting high-frequency waveform data in a high-density event mode by the distributed intelligent sensing network; performing time-domain statistical analysis on the low-frequency aggregated data to extract first physical features, and performing time-frequency domain analysis on the high-frequency waveform data to extract second physical features; and combining the first physical features and the second physical features together as the basic physical features.

[0008] Optionally, the distributed intelligent sensing network comprises: a multi-modal sensor array, an edge computing core and a self-organizing network communication interface, wherein: the multi-modal sensor array is configured to perform hierarchical collection to obtain the electrical measurement data and the environmental state data; the edge computing core is configured to perform switching logic between different trigger modes and perform data preprocessing and physical feature extraction; and the self-organizing network communication interface is configured to transmit data and control instructions between nodes of the distributed intelligent sensing network.

[0009] Optionally, the generating of the set of credible physical features comprises: associating the basic physical features with the topological relationship data to generate semantic physical features; performing three-level credibility evaluation on the semantic physical features, including source layer, data layer and network layer, to calculate a multi-dimensional credibility vector; and binding the semantic physical features with the multi-dimensional credibility vector to generate the set of credible physical features.

[0010] Optionally, the calculating the multi-dimensional credibility vector comprises: calculating a source credibility component and a data quality credibility component according to a data source attribute and a data quality index of the semantic physical feature; generating a network consistency credibility component based on the semantic physical feature and the topological relationship data, and combining the source credibility component, the data quality credibility component, and the network consistency credibility component into the multi-dimensional credibility vector.

[0011] Optionally, the extracting the dynamic risk factor comprises: initializing a semantic causal graph based on the set of trusted physical features and the topological relationship data, taking the physical entity as a node and the topological relationship data as an edge; generating an influence degree matrix by applying multi-source disturbance simulation on the semantic causal graph and iteratively updating the semantic causal graph according to the influence degree matrix; and calculating and extracting the dynamic risk factor based on the updated semantic causal graph.

[0012] Optionally, the method further comprises: identifying a key risk contribution node according to a contribution degree order based on a trigger mode switching frequency of the distributed intelligent sensing network and the influence degree matrix.

[0013] Optionally, the generating the running alarm state comprises: combining the basic physical feature, the set of trusted physical features, and the dynamic risk factor to construct a hierarchical state vector; generating a reference health threshold vector according to the topological relationship data, and dynamically tightening a dimension corresponding to the key risk contribution node in the reference health threshold vector to generate a dynamic health threshold vector; and performing a dimension-by-dimension comparison between the hierarchical state vector and the dynamic health threshold vector to generate the running alarm state.

[0014] Optionally, the generating the threshold calibration signal for adjusting the dynamic health threshold vector comprises: receiving a treatment feedback result of the running alarm state, determining the running alarm state according to the treatment feedback result to obtain an alarm determination result; and calculating and generating a threshold calibration signal according to the determination result to adjust the dynamic health threshold vector.

[0015] Based on the same inventive concept, the application also provides an operation warning system of a power system, comprising: a physical state sensing module, configured to deploy a distributed intelligent sensing network on a physical entity of the power system, the distributed intelligent sensing network switches between different trigger modes, and collects electrical measurement data, environmental state data and topological relationship data in layers, pre-processes and extracts physical features from the data in each layer to obtain basic physical features; a feature calibration module, configured to perform feature semantic labeling on the basic physical features, associate the physical entity with the topological relationship data, and perform hierarchical credibility evaluation on the basic physical features to quantify data source reliability and generate a set of credible physical features; a causal risk modeling module, configured to construct and initialize a semantic causal graph based on the set of credible physical features and the topological relationship data, simulate device abnormalities and environmental mutations by applying a multi-source disturbance to the semantic causal graph, iteratively update causal edge weights according to node influence degree by using adaptive edge weight optimization, and extract dynamic risk factors; a state evaluation and warning decision module, configured to construct a hierarchical state vector from the basic physical features, the set of credible physical features and the dynamic risk factors, construct a dynamic health threshold vector according to the topological relationship data, compare the hierarchical state vector with the dynamic health threshold vector dimension by dimension, and generate an operation warning state; and an adaptive optimization module, configured to perform attribution analysis and parameter tracing according to the operation warning state, and generate a threshold calibration signal for adjusting the dynamic health threshold vector.

[0016] Compared with the prior art, the application has the following advantages: 1. The application realizes the intelligentization and precision of the warning decision by constructing a complete technical chain from data sensing, credible evaluation to risk modeling, and introducing a feedback calibration mechanism. Through the multi-mode switching and hierarchical credibility evaluation mechanism of the distributed intelligent sensing network, the comprehensiveness and reliability of the input data are ensured, the quality of the input data is improved from the source, false positives and false negatives caused by factors such as sensor failure, communication delay or data damage are effectively suppressed, and the accuracy and reliability of the warning decision are significantly enhanced; 2. The application converts the static power grid topology into a dynamic risk transmission model by constructing a semantic causal graph and extracting dynamic risk factors, which can quantitatively evaluate the mutual influence between different devices and the vulnerability under specific working conditions. This realizes the deep insight and forward-looking prediction of potential fault chains, breaks through the limitation of traditional warning methods which can only make judgments based on the current state, and changes the warning system from passive state monitoring to active risk warning, providing decision support for preventive maintenance; 3、The dynamic health threshold vector and threshold calibration signal mechanism make the sensitivity of the alarm be able to be accurately matched with the real-time risk level.

[0017] Other features and advantages of the present application will be set forth in the following description, and in part will be apparent from the description, or can be learned by practice of the application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0019] Figure 1 is a flow diagram of a power system operation alarm method according to an embodiment of the present application.

[0020] Figure 2 is a flow diagram of a trusted physical feature set generation according to an embodiment of the present application.

[0021] Figure 3 is a physical entity clustering analysis diagram based on an influence degree matrix according to an embodiment of the present application.

[0022] Figure 4 is a key risk contribution node identification and sorting diagram according to an embodiment of the present application.

[0023] Figure 5 is a structural diagram of a power system operation alarm system according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0025] Referring toFigure 1 One embodiment of the present application proposes a power system operation warning method, which adopts a technical scheme combining multi-source data fusion, hierarchical credibility evaluation, causal risk modeling and dynamic threshold adaptive calibration, and can realize accurate identification and forward-looking early warning of potential risks of the power system, and improve the accuracy and adaptive ability of warning decision.

[0026] The method of the embodiment specifically includes: A distributed intelligent sensing network is deployed on physical entities of a power system, the distributed intelligent sensing network switches between different trigger modes, and collects electrical measurement data, environmental state data and topological relationship data in layers, pre-processes and extracts physical features from each layer of data to obtain basic physical features; The basic physical features are subjected to feature semantic labeling, the physical entities are associated with the topological relationship data, and hierarchical credibility evaluation is performed on the basic physical features to quantify data source reliability, and a set of credible physical features is generated; Based on the set of credible physical features and the topological relationship data, a semantic causal graph is constructed and initialized, multi-source disturbance simulation equipment anomalies and environmental mutations are applied to the semantic causal graph, adaptive edge weight optimization is used to iteratively update causal edge weights according to node influence degrees, and dynamic risk factors are extracted; The basic physical features, the set of credible physical features and the dynamic risk factors are constructed into a hierarchical state vector, a dynamic health threshold vector is constructed according to the topological relationship data, the hierarchical state vector and the dynamic health threshold vector are compared dimension by dimension, and an operation warning state is generated; According to the operation warning state, attribution analysis and parameter tracing are performed to generate a threshold calibration signal for adjusting the dynamic health threshold vector.

[0027] The technical scheme combining multi-source data fusion, hierarchical credibility evaluation, causal risk modeling and dynamic threshold adaptive calibration can realize accurate identification and forward-looking early warning of potential risks of the power system, and improve the accuracy and adaptive ability of warning decision.

[0028] Optionally, the basic physical features include: The distributed intelligent sensing network collects low-frequency aggregated data in a baseline monitoring mode, and collects high-frequency waveform data in a high-density event mode; Specifically, in this step, the distributed intelligent sensing network adaptively switches between two working modes. In normal operation, the network works in a baseline monitoring mode, in which data is collected and processed at a lower frequency to form low-frequency aggregated data, such as minute-level averages of line voltage effective value and current effective value. When the low-frequency aggregated data fluctuates beyond the preset stable range, the distributed intelligent sensing network switches to a high-density event mode, in which instantaneous voltage and current waveforms are captured at a high sampling frequency to obtain high-frequency waveform data.

[0029] The low-frequency aggregated data is subjected to time-domain statistical analysis to extract first physical features, and the high-frequency waveform data is subjected to time-frequency domain analysis to extract second physical features; Specifically, for low-frequency aggregated data, the processing process performs time-domain statistical analysis to calculate statistical quantities such as mean and variance within a time window, which constitute the first physical features representing steady-state characteristics. At the same time, for high-frequency waveform data, the processing process performs time-frequency domain analysis, such as Fourier transform to decompose wave components and calculate total harmonic distortion (THD) to quantify waveform distortion degree, whose calculation formula can be expressed as: , wherein, represents the effective value of the fundamental component; represents the effective value of each harmonic component; is the highest order of harmonic analysis, which is an integer preset according to monitoring accuracy requirements. The harmonic content and transient overvoltage amplitude extracted by such analysis constitute the second physical features representing dynamic disturbance characteristics.

[0030] The first physical features and the second physical features are combined together as basic physical features.

[0031] Specifically, this step combines the first physical features representing steady-state operating characteristics and the second physical features representing dynamic disturbance characteristics extracted in the previous steps to form a multi-dimensional basic physical feature set, which provides comprehensive and reliable basis for subsequent alarm decision-making.

[0032] Optionally, the distributed intelligent sensing network comprises a multi-modal sensor array, an edge computing core, and a self-organizing network communication interface, wherein: The multi-modal sensor array is used to perform hierarchical collection to obtain the electrical measurement data and the environmental state data; In particular, the distributed intelligent sensing network in the embodiments of the present application is an intelligent hardware node deployed on a physical entity, which integrates sensing, computing and communication capabilities. A multi-modal sensor array is the sensing front-end of the node, which physically integrates multiple types of sensors. For example, it includes high-frequency current transformers and voltage divider voltage sensors for acquiring electrical measurement data, as well as temperature sensors, humidity sensors and Micro-Electro-Mechanical System (MEMS) vibration sensors for acquiring environmental state data. Each sensor in the array is configured to work at different sampling rates and accuracies according to instructions from the edge computing core to perform hierarchical collection.

[0033] The edge computing core is used to perform switching logic between different trigger modes, and to pre-process and extract physical features from data; In particular, the edge computing core is the local processing unit of the node, which is usually implemented by a low-power Microcontroller Unit (MCU) or System on Chip (SoC). The core performs switching logic between different trigger modes, for example, it analyzes the data collected in the baseline monitoring mode in real time, and once it finds that the feature fluctuation exceeds the internally set dynamic threshold, it immediately triggers switching to the high-density event mode. At the same time, the core is also responsible for performing pre-processing, such as digital filtering and normalization of the collected raw data, and performing physical feature extraction, such as locally calculating voltage effective value, power factor and energy features of transient waveforms, thereby reducing the amount of data that needs to be transmitted through the network.

[0034] The self-organizing network communication interface is used to transmit data and control instructions between the nodes of the distributed intelligent sensing network.

[0035] In particular, the self-organizing network communication interface is the communication unit of the node, which uses a wireless communication protocol that supports mesh network topology. The interface enables each node of the distributed intelligent sensing network not only to communicate with the central aggregation node, but also to relay data between each other. This self-organizing feature ensures that even if some communication links are blocked or fail, data and control instructions can still be transmitted by other nodes, greatly enhancing the communication robustness and reliability of the entire sensing network.

[0036] Optionally, the generating a set of trusted physical features comprises: associating the basic physical features with the topological relationship data to generate semantic physical features; Specifically, this step aims to further process the basic physical features obtained in the previous step to endow them with physical semantics. The processing process associates each data point from the basic physical features with the power system topology relationship data. The topology relationship data describes in detail the unique identification of each physical entity in the power grid and their electrical connection relationship with each other. Through this association operation, an abstract measurement value, such as a temperature data, is accurately labeled as its source, such as "the temperature of the A-phase winding of the No. 1 main transformer in a certain substation", thereby generating a semantic physical feature with clear physical meaning.

[0037] A multi-dimensional credibility vector is calculated by performing a three-level credibility evaluation on the semantic physical features, including a source layer, a data layer, and a network layer; Specifically, for the semantic physical features that have completed semantic labeling, the processing process performs a three-level credibility evaluation to calculate a multi-dimensional credibility vector. The first layer is the source layer evaluation, which mainly assesses the reliability of the data acquisition device itself, such as the factory calibration accuracy of the reading sensor, the last calibration date, and other attributes, and quantifies it as a source credibility component. The second layer is the data layer evaluation, which focuses on evaluating the quality of the data itself, such as the real-time and completeness of the data, and calculates the data quality credibility component. The third layer is the network layer evaluation, which is a cross-validation process based on physical laws. Using topology relationship data, multiple semantic physical features at the same electrical node or associated path are compared for consistency. For example, according to Kirchhoff's current law, the sum of the currents flowing into a node should be zero. The network consistency credibility component is generated by calculating the deviation of the actual measurement values.

[0038] The semantic physical features are bound to the multi-dimensional credibility vector to generate a set of credible physical features.

[0039] Specifically, this step is to bind each semantic physical feature to its corresponding multi-dimensional credibility vector. This binding operation encapsulates the value, semantics, and credibility of the feature into a unified data structure, forming the final set of credible physical features and providing a solid data foundation for subsequent analysis and decision-making, such as Figure 2 As shown, the nodes in the outer circle of the figure represent key data entities and processing modules, and the flow between the nodes clearly reveals the complete path of data from input, through layer-by-layer processing and evaluation, to final fusion output.

[0040] Optionally, the calculation of the multi-dimensional credibility vector includes: According to the data source attributes and data quality indicators of the semantic physical features, the source credibility component and the data quality credibility component are calculated; In particular, this step aims to deeply quantify the reliability of the semantic physical features. The processing process calculates the source credibility component and the data quality credibility component according to the data source attribute and the data quality index embedded in each semantic physical feature. The data source attribute mainly refers to the static parameters of the sensor collecting the feature itself, such as its factory accuracy level, calibration period, and service life, etc. These attributes are quantified and then comprehensively calculated to obtain the source credibility component. The data quality index refers to the dynamic attributes generated during data transmission and recording, such as data transmission delay and data packet integrity rate. Through the evaluation of these indexes, the data quality credibility component is calculated.

[0041] Based on the semantic physical features, the consistency of the topological relationship data is compared, the network consistency credibility component is generated, and the source credibility component, the data quality credibility component, and the network consistency credibility component are combined into a multi-dimensional credibility vector.

[0042] In particular, the core of generating the network consistency credibility component is to cross-verify the semantic physical features by using the inherent physical laws of the power system. Based on the topological relationship data, a group of physically related semantic physical features is identified. For example, for an electrical node in the power grid, according to Kirchhoff's current law, at any time, the phasor sum of all line currents flowing into the node should theoretically be zero. The residual error of the actual measured current is calculated by the following formula to evaluate the consistency of the data: , wherein, is the current phasor of the first connecting line from the semantic physical feature; is the total number of lines connected to the node, which is an integer obtained by parsing the topological relationship data. The smaller the calculated residual error , the higher the consistency of the measurement data of each sensor. This residual value is then mapped to a network consistency credibility component between 0 and 1. Finally, the source credibility component, the data quality credibility component, and the network consistency credibility component calculated by the above steps are combined into a structured multi-dimensional credibility vector.

[0043] Optionally, the extraction of dynamic risk factors includes: Based on the set of credible physical features and the topological relationship data, the physical entity is taken as a node, and the topological relationship data is taken as an edge to initialize and construct a semantic causal graph. Specifically, this step aims to convert the static topology of the power system into a dynamic causal relationship model. The processing process uses the set of trusted physical features obtained in the previous step and the topology relationship data of the power system to take the physical entities in the power grid, such as generators, transformers, and lines, as the nodes of the graph, and the electrical connection relationship between these entities, i.e., the topology relationship data, as the initial edges of the graph, thereby constructing a basic semantic causal graph. At this time, the structure of the graph only reflects the physical connection, and the edges have not been assigned a quantitative causal strength.

[0044] A multi-source disturbance simulation is applied to the semantic causal graph to generate an influence degree matrix, and the semantic causal graph is iteratively updated according to the influence degree matrix. Specifically, this step aims to dynamically optimize the semantic causal graph. The processing process applies multi-source disturbances to the constructed semantic causal graph in a computational simulation manner. Disturbance simulation can be based on mature power grid simulation analysis software, such as PSASP, BPA, etc. The type and parameters of the disturbance can be set to typical severe faults in the power grid plan, such as setting a 100ms three-phase metallic short-circuit fault on a critical transmission line, simulating an N-1 trip event of an important generator set, or increasing a large industrial load by 50% instantaneously. For each simulated disturbance, the processing process calculates the propagation effect of the disturbance in the entire network based on the physical model of the power grid operation, i.e., the influence degree of the state change of a node on the state of all other nodes. By applying multiple types of disturbances to different nodes and analyzing their global influence, an influence degree matrix can be generated, each element of which quantifies the mutual influence strength between nodes. Subsequently, according to the influence degree matrix, the processing process iteratively updates the causal edge weight of the semantic causal graph, as shown in Figure 3 An example of an influence degree matrix is shown, with rows and columns being physical entities in the system, and the size and color of the bubbles representing the influence strength between entities. The cluster tree diagram on the graph automatically groups highly correlated entities according to the influence pattern, thereby identifying potential risk coupling groups.

[0045] Based on the updated semantic causal graph, dynamic risk factors are calculated and extracted.

[0046] Specifically, this step aims to calculate and extract dynamic risk factors based on the iteratively updated semantic causal graph. These risk factors are key indicators for quantifying the risk state of the power grid, and are usually calculated through graph theory algorithms. For example, by calculating the weighted out-degree of each node in the graph, the ability of the node as a risk source to spread influence outward can be obtained, which is a dynamic risk factor. By calculating the weighted in-degree of the node, the vulnerability of the node to the influence of other nodes can be obtained, which is another dynamic risk factor.

[0047] Optionally, the method further comprises: According to the contribution degree, a key risk contribution node is identified by sorting the trigger mode switching frequency of the distributed intelligent sensing network and the influence degree matrix.

[0048] Specifically, this step aims to deeply sort the risk priority of all physical entity nodes in the power grid, which does not rely on a single indicator, but integrates two kinds of information with different properties: one is the trigger mode switching frequency of the distributed intelligent sensing network representing the current running activity and stability of the node; the other is the influence degree matrix representing the potential global influence of the node. The processing process first counts the number of times that the distributed intelligent sensing network associated with each physical entity switches from the baseline monitoring mode to the high-density event mode within a preset time window, obtaining the trigger mode switching frequency of the node as its "dynamic activity" indicator. At the same time, the processing process calls the influence degree matrix generated in the previous step to extract the comprehensive influence degree of each node on all other nodes in the network as its "static influence" indicator. The core step is to integrate the normalized "dynamic activity" indicator and "static influence" indicator through a weighted fusion model to obtain a unified contribution degree score. Finally, the processing process ranks all physical entities in descending order according to the contribution degree, and the top-ranked nodes are identified as key risk contribution nodes, as shown in Figure 4 The result of contribution degree sorting of all physical entities is shown, each entity corresponds to a "lollipop", and the length represents the final comprehensive contribution degree score of the entity. Through the sorting from high to low, the key nodes with the largest contribution to the overall system risk can be identified.

[0049] Optionally, the generating the operation alarm state comprises: Combining the basic physical features, the set of trusted physical features, and the dynamic risk factor to construct a hierarchical state vector; Specifically, this step aims to integrate three types of key information to construct a comprehensive hierarchical state vector. The three types of information are: the basic physical features that directly reflect the physical quantity measurement values; the set of trusted physical features that have been processed by semanticization and trust quantification; and the dynamic risk factor that prospectively reveals vulnerability and risk transmission path. By combining these three according to the dimension of physical entities, a hierarchical state vector is formed that contains not only the current real state, but also the data trustworthiness and future risk trend. Each dimension of the vector accurately corresponds to a specific parameter of a specific device in the power grid.

[0050] Generating a baseline health threshold vector according to the topological relationship data, and dynamically tightening the dimensions corresponding to the key risk contribution nodes in the baseline health threshold vector to generate a dynamic health threshold vector; Specifically, this step aims to construct a benchmark for comparison, namely the dynamic health threshold vector. The process begins by generating an initial health threshold for each dimension of the hierarchical state vector based on the topological relationship data of the power system and industry operating procedures. These thresholds collectively form the benchmark health threshold vector. Subsequently, the process locates the dimensions in the benchmark health threshold vector associated with the key risk-contributing nodes identified in the previous step. For these specific dimensions, the process performs a dynamic tightening operation, which involves a moderate reduction in the allowable range of their health thresholds. In this way, the static benchmark health threshold vector evolves into a dynamic health threshold vector that can reflect the risk distribution in real time.

[0051] Comparing the hierarchical state vector with the dynamic health threshold vector generates an operational alarm state.

[0052] Specifically, this step aims to perform a dimension-by-dimension comparison operation. The process compares each element value in the hierarchical state vector with the dynamically tightened threshold value at its corresponding position in the dynamic health threshold vector. Once any element of the state vector is found to exceed its corresponding dynamic health threshold, the process determines that there is an anomaly. Based on the number of elements exceeding the threshold, the extent of the exceedance, and the dynamic risk factors associated with these elements, a clear operational alarm state is ultimately generated, such as normal, pre-warning, or severe warning.

[0053] Optionally, the generating of the threshold calibration signal for adjusting the dynamic health threshold vector comprises: Receiving the handling feedback result of the operational alarm state and determining the operational alarm state based on the handling feedback result to obtain an alarm determination result; Specifically, this step aims to build a feedback-driven self-optimizing closed loop to ensure that the alarm system can continuously learn and improve from historical experience. The process begins by receiving and analyzing the handling feedback result for the previous operational alarm state. This handling feedback result can come from human operation logs, which detail the conclusions of on-site checks by operation personnel, or from action records of power system automation devices. The process determines the previous generated operational alarm state based on this handling feedback result, thereby obtaining a clear alarm determination result, which is usually classified as valid alarm, false alarm, or missed alarm.

[0054] According to the determination result, a threshold calibration signal is calculated and generated to adjust the dynamic health threshold vector.

[0055] Specifically, after obtaining the alarm decision result, the processing process enters the core calculation stage to generate a threshold calibration signal with a clear adjustment direction and amplitude. The calculation process follows a set of adjustment logic: if the alarm decision result is a false alarm, it indicates that the current dynamic health threshold vector is too strict, and the processing process will calculate an adjustment amount for relaxing the threshold; on the contrary, if the decision result is a missed alarm, it indicates that the threshold is too loose, and the processing process will calculate an adjustment amount for tightening the threshold; if the decision is a valid alarm, the adjustment amount can be set to zero. The size of the adjustment amount is also controlled by a learning rate to avoid excessive oscillation due to a single event. Finally, the calculation result containing the specific adjustment direction and value, i.e. the threshold calibration signal, is used to update the dynamic health threshold vector.

[0056] To verify the feasibility of the application in implementation, the application is applied to the operation monitoring and alarm of a certain municipal regional power grid. In order to ensure the power supply reliability of the core area of the city, the key physical entities such as the substation and the transmission line under jurisdiction need to be monitored in real time to realize early warning and accurate positioning of potential faults. The traditional alarm system has problems such as fixed threshold, one-sided alarm information and difficulty in distinguishing fault severity, which often leads to missed alarms of early problems and false alarms of non-emergency disturbances. The municipal power grid dispatching center hopes to use the method of the application to build a closed-loop alarm system from intelligent sensing, data purification, risk deduction to dynamic decision-making.

[0057] In this embodiment, the power grid dispatching center deploys a distributed intelligent sensing network of the application on multiple key substations and important transmission lines. The network collects voltage, current, device temperature, environmental humidity and other electrical quantity measurement data and environmental state data through a multi-modal sensor array, and performs localized feature extraction and mode switching by the edge computing core. The entire alarm system has been running for 6 months, during which various operating conditions of the power grid are recorded, and the effect is compared with that of the traditional fixed threshold alarm system running at the same time.

[0058] In the embodiment, the application successfully captures and analyzes an early latent fault of a transformer. The distributed intelligent sensing network deployed on the No. 2 main transformer of the southern substation runs in the baseline monitoring mode to collect low-frequency aggregated data such as voltage effective value and active power at a frequency of seconds. At about 10:20, the edge computing core monitors the voltage effective value of the low-voltage side of the transformer to appear a small but continuous fluctuation outside the preset stable range, immediately executes the switching logic between different trigger modes, and switches the multi-modal sensor array to the high-density event mode to capture the instantaneous voltage and current high-frequency waveform data at a high sampling rate of kilohertz.

[0059] The method then performs parallel processing on the two types of data. For the low-frequency aggregated data before mode switching, time-domain statistical analysis is performed to calculate a 15% increase in voltage standard deviation; for the high-frequency waveform data collected after mode switching, time-frequency domain analysis is performed, and the total harmonic distortion rate is calculated to be 4.2% through Fourier transform, which is significantly higher than the normal value of 2.5%. These two features are combined as the basic physical features reflecting the abnormal state of the transformer.

[0060] Next, the method performs feature calibration on the basic physical features. First, these features are associated with the topological relationship data and are semantically labeled as "voltage standard deviation of low-voltage side of No. 2 main transformer of Chengnan substation" and "THD of low-voltage side of No. 2 main transformer of Chengnan substation", generating semantic physical features. Subsequently, hierarchical reliability assessment is performed on the features: source layer assessment confirms that the sensor calibration record is good; data layer assessment confirms that the data transmission delay is low and there is no packet loss; network layer assessment uses topological relationship for consistency comparison, cross- verifies the current of each line flowing into the bus of the substation, and the residual error is minimal, excluding the possibility of other line sensor failure. These three components are combined into a multi-dimensional reliability vector and bound with the semantic physical features, confirming the reliability of the abnormal data and forming a set of reliable physical features.

[0061] Based on this set of reliable physical features, the method updates the pre-constructed regional power grid semantic causal graph. Through historical multi-source disturbance simulation, an influence matrix has been generated, which shows that the No. 2 main transformer of Chengnan substation has a high influence degree on the downstream multiple important loads. At the same time, it is statistically found that the trigger mode switching frequency of the distributed intelligent sensing network of this node has occurred 3 times in the past week, which is much higher than that of other nodes. By integrating these two indicators and sorting them according to the contribution degree, the No. 2 main transformer of Chengnan substation is identified as a key risk contribution node. Based on the updated semantic causal graph, the dynamic risk factor representing the risk transmission capacity of the node is calculated and extracted through the graph aggregation function.

[0062] In the alarm decision stage, the method combines the basic physical features, the set of trusted physical features, and the dynamic risk factor to construct a hierarchical state vector. The benchmark health threshold vector is generated based on the topological relationship data. Since the No. 2 main transformer of Chengnan substation is identified as a key risk contribution node, the method dynamically tightens the dimension corresponding to this node in the benchmark health threshold vector to generate a dynamic health threshold vector. For example, the benchmark health threshold of the total harmonic distortion rate is 5.0%, and after dynamic tightening, it becomes 3.5%. At this time, the measured THD value 4.2% in the hierarchical state vector is compared with the dynamic health threshold 3.5% dimension by dimension, and it is found that it has exceeded the limit, and an operation alarm state of “warning” level is generated at 10:23, which clearly points out that the transformer exists waveform distortion risk. The traditional alarm system at the same period is not triggered due to its fixed 5.0% threshold, and cannot find this abnormality.

[0063] The dispatch center receives the alarm and sends operation and maintenance personnel to check. The disposal feedback result shows that there is a slight insulation aging sign in the internal winding of the transformer. The disposal feedback result is input, and it is determined as an “effective alarm”, and the alarm determination result is obtained. Due to the accuracy of the alarm, according to the determination result, the threshold calibration signal is calculated and generated, and the adjustment logic of the current dynamic health threshold vector is positively confirmed, so that more accurate judgment can be made in similar future events.

[0064] Through 6 months of comparative operation, the application shows significant advantages in the accuracy, forward-looking and adaptive ability of the alarm. See Tables 1 to 3 for specific data.

[0065] Table 1 Key risk contribution node identification data table Table 2 Dynamic health threshold adjustment and alarm triggering comparison table Table 3 Operation effect comparison data table The above Tables 1 to 3 record the actual application data of the application in the regional power grid, and detailedly show the performance of the method in identifying key risks, dynamic decision and overall performance.

[0066] As can be seen from Table 1 data, the application can effectively integrate the real-time “dynamic activity” and potential “static influence” of the node, accurately identify the No. 2 main transformer of Chengnan substation as the most worthy of attention risk source, and provide a basis for subsequent differentiated monitoring.

[0067] Table 2 clearly shows the core advantage of the present application. For the traditional alarm system, 4.2% THD is considered normal at a fixed threshold of 5.0%. However, due to the identification of the transformer as a key risk contribution node, the present application dynamically tightens the threshold to 3.5%, thereby successfully identifying and issuing an early warning at the early stage of failure. This proves the decisive role of the dynamic health threshold mechanism in improving alarm sensitivity.

[0068] The overall performance data in Table 3 macroscopically confirms the technical effect of the present application. Through a closed-loop intelligent processing flow, not only can problems be detected earlier, but the alarm quality is also significantly improved, with accuracy improved while false positives and false negatives are significantly reduced. This fully proves that the present application can provide more reliable, intelligent and forward-looking technical support for the safe and stable operation of the power system.

[0069] Based on the same inventive concept, the present application also provides an operation alarm system for a power system, as shown in Figure 5 The system comprises: a physical state perception module for deploying a distributed intelligent perception network on physical entities of the power system, the distributed intelligent perception network switching between different trigger modes and hierarchically collecting electrical measurement data, environmental state data and topological relationship data, pre-processing and physical feature extraction of each layer of data to obtain basic physical features; a feature calibration module for performing feature semantic annotation on the basic physical features, associating the physical entities with the topological relationship data, and performing hierarchical credibility evaluation on the basic physical features to quantify data source reliability and generate a set of trusted physical features; a causal risk modeling module for constructing and initializing a semantic causal graph based on the set of trusted physical features and the topological relationship data, imposing multi-source disturbance simulation equipment anomalies and environmental mutations on the semantic causal graph, iteratively updating causal edge weights according to node influence degree using adaptive edge weight optimization, and extracting dynamic risk factors; a state evaluation and alarm decision module for constructing a hierarchical state vector from the basic physical features, the set of trusted physical features and the dynamic risk factors, constructing a dynamic health threshold vector based on the topological relationship data, comparing the hierarchical state vector with the dynamic health threshold vector dimension by dimension, and generating an operation alarm state; an adaptive optimization module for performing attribution analysis and parameter tracing according to the operation alarm state, and generating a threshold calibration signal for adjusting the dynamic health threshold vector.

[0070] It should be noted that the function division and information interaction among the above various modules are logical, and can be integrated in the same software platform or distributed in physical implementation. The connection between them represents the data flow and control flow, and aims to cooperatively realize the object of the present application. The above description is only exemplary embodiments of the present application, and cannot limit the protection scope of the present application.

Claims

1. An operation alarm method for a power system, characterized in that: The method comprises: Deploy a distributed intelligent sensing network on the physical entities of the power system. The distributed intelligent sensing network switches between different trigger modes and collects electrical measurement data, environmental status data, and topological relationship data in layers. It preprocesses and extracts physical features from the data at each layer to obtain basic physical features. Performing feature semantic annotation on the basic physical features, associating the physical entities with the topological relationship data, and performing hierarchical credibility assessment on the basic physical features to quantify the reliability of the data source and generate a credible physical feature set; Based on the trusted physical feature set and the topological relationship data, a semantic causal graph is constructed and initialized. Multi-source disturbances are applied to the semantic causal graph to simulate equipment anomalies and environmental mutations. The causal edge weights are iteratively updated according to the node influence using adaptive edge weight optimization to extract dynamic risk factors. Constructing a hierarchical state vector from the basic physical features, the trusted physical feature set, and the dynamic risk factor, constructing a dynamic health threshold vector based on the topological relationship data, and performing a dimension-by-dimension comparison between the hierarchical state vector and the dynamic health threshold vector to generate an operation alarm state; According to the operation alarm state, attribution analysis and parameter tracing are performed to generate a threshold calibration signal for adjusting the dynamic health threshold vector.

2. The operation alarm method of a power system according to claim 1, characterized in that: The basic physical characteristics are obtained as follows: The distributed intelligent perception network collects low-frequency aggregate data in a baseline monitoring mode, and the distributed intelligent perception network collects high-frequency waveform data in a high-density event mode; Performing a time-domain statistical analysis on the low-frequency aggregated data to extract a first physical feature, and performing a time-frequency domain analysis on the high-frequency waveform data to extract a second physical feature; The first physical feature and the second physical feature are combined to form a basic physical feature.

3. The operation alarm method of a power system according to claim 1, characterized in that: The distributed intelligent sensing network includes: a multimodal sensor array, an edge computing core and a self-organizing network communication interface, wherein: The multimodal sensor array is used to perform hierarchical acquisition to obtain the electrical measurement data and the environmental status data; The edge computing core is used to execute the switching logic between different trigger modes and perform data preprocessing and physical feature extraction; The self-organizing network communication interface is used to transmit data and control instructions between nodes of the distributed intelligent perception network.

4. The operation alarm method of a power system according to claim 1, characterized in that: Generating a credible physical feature set includes: Associating the basic physical features with the topological relationship data to generate semantic physical features; For the semantic physical features, a three-level credibility evaluation is performed at the source layer, data layer, and network layer to calculate a multi-dimensional credibility vector; The semantic physical feature is bound to the multi-dimensional credibility vector to generate a credible physical feature set.

5. The operation alarm method of a power system according to claim 4, characterized in that: The calculation to obtain the multidimensional credibility vector includes: Calculating a source credibility component and a data quality credibility component based on the data source attributes and data quality indicators of the semantic-physical features; Based on the semantic physical features, a consistency comparison is performed with the topological relationship data to generate a network consistency credibility component, and the source credibility component, the data quality credibility component, and the network consistency credibility component are combined into a multidimensional credibility vector.

6. The operation alarm method of a power system according to claim 1, characterized in that: The extracting of dynamic risk factors comprises: Based on the trusted physical feature set and the topological relationship data, a semantic causal graph is constructed by initializing the physical entities as nodes and the topological relationship data as edges; Applying multi-source disturbance simulation on the semantic causal graph to generate an influence matrix, and iteratively updating the semantic causal graph according to the influence matrix; Based on the updated semantic causal graph, dynamic risk factors are calculated and extracted.

7. The operation alarm method of a power system according to claim 6, characterized in that: The method further comprises: The trigger mode switching frequency of the distributed intelligent perception network and the influence matrix are combined and sorted according to contribution to identify key risk contribution nodes.

8. The operation alarm method of a power system according to claim 7, characterized in that: The generating operation alarm status includes: Combining the basic physical feature, the trusted physical feature set, and the dynamic risk factor to construct a layered state vector; Generating a baseline health threshold vector based on the topological relationship data, and dynamically tightening the dimension corresponding to the key risk contribution node in the baseline health threshold vector to generate a dynamic health threshold vector; The hierarchical state vector is compared dimension by dimension with the dynamic health threshold vector to generate an operation alarm state.

9. The operation alarm method of a power system according to claim 1, characterized in that: Generating a threshold calibration signal for adjusting the dynamic health threshold vector includes: receiving a handling feedback result of the operation alarm state, and determining the operation alarm state according to the handling feedback result to obtain an alarm determination result; According to the determination result, a threshold calibration signal is calculated and generated to adjust the dynamic health threshold vector.

10. An operation alarm system for a power system, applied to an operation alarm method for a power system according to any one of claims 1 to 9, characterized in that: The system comprises: The physical state perception module is used to deploy a distributed intelligent perception network on the physical entities of the power system. The distributed intelligent perception network switches between different trigger modes and collects electrical measurement data, environmental state data, and topological relationship data in layers. It preprocesses and extracts physical features from each layer of data to obtain basic physical features. a feature calibration module, configured to perform feature semantic annotation on the basic physical features, associate the physical entities with the topological relationship data, perform hierarchical credibility assessment on the basic physical features, quantify the reliability of the data source, and generate a credible physical feature set; A causal risk modeling module is configured to construct and initialize a semantic causal graph based on the trusted physical feature set and the topological relationship data, apply multi-source disturbances to the semantic causal graph to simulate equipment anomalies and environmental mutations, iteratively update causal edge weights based on node influence using adaptive edge weight optimization, and extract dynamic risk factors; a state assessment and alarm decision module, configured to construct a hierarchical state vector from the basic physical characteristics, the trusted physical characteristic set, and the dynamic risk factor, construct a dynamic health threshold vector based on the topological relationship data, and compare the hierarchical state vector with the dynamic health threshold vector dimension by dimension to generate an operation alarm state; The adaptive optimization module is used to perform attribution analysis and parameter tracing according to the operation alarm state, and generate a threshold calibration signal for adjusting the dynamic health threshold vector.

Images