Overload analysis method based on feeder N-1 risk

Abstract

The invention belongs to the technical field of safety and stability analysis of a power system, and particularly relates to an overload analysis method based on feeder N-1 risk, which comprises the following steps: after simulating an emergency event of an initial component fault, not only determining a main risk index directly related to the event, but also determining an overload risk index based on a system state change caused by the event; a secondary vulnerability index for quantifying the sensitivity of other non-fault components in the system to subsequent disturbance is calculated for the other non-fault components in the system, a single component fault event is converted from a static end point to a dynamic detection process of a power grid vulnerability chain, and a hidden cascading fault path is actively identified and quantified by using the single component fault as a driving force. According to the overload analysis method based on the feeder N-1 risk, the contradiction between safety and complexity, prediction and interference and local and global optimization in the prior art is fundamentally solved, and advanced early warning of cascading failure risks and improvement of the overall risk insight ability of the system are achieved.

Description

Technical Field

[0001] This invention relates to the field of power system safety and stability analysis technology, specifically to an overload analysis method based on feeder N-1 risk. Background Technology

[0002] The safe and stable operation of the power system plays a decisive role in ensuring the normal operation of society. To ensure that the power grid can remain stable even when a single component such as a transmission line or transformer fails, the N-1 safety criterion has been established as the basic principle for power grid planning and operation. This criterion requires that when any single component of the power grid fails and goes out of service, the rest of the system can still operate normally without chain reactions that lead to stability disruption or large-scale power outages. Accurate and efficient N-1 safety analysis of the power grid is a core technical link in ensuring the reliability of the power grid.

[0003] Traditional N-1 security analysis mainly relies on offline power flow calculations and transient stability simulations. Technicians pre-define a series of possible fault scenarios and use large-scale computing software to perform detailed numerical simulations of each scenario. The calculation results of this method have high reliability. However, with the large-scale grid connection of new energy sources, the widespread application of power electronic equipment, and the increasing complexity of load characteristics, the operating state of modern power grids exhibits high dynamism and uncertainty. Traditional offline, full-section scanning analysis methods involve huge computational loads, with a single full-network scan taking several hours, which cannot meet the needs of real-time dispatch decision-making. In addition, its analysis is static and isolated, and can only give a binary conclusion of whether the limit is exceeded, failing to reveal the more destructive chain of cascading fault risks hidden behind a single fault.

[0004] To overcome the real-time bottleneck of traditional methods, in recent years, the industry has begun to explore the application of data-driven technology to power grid security analysis. The basic idea is to use historical operating data and simulation data to train machine learning models to quickly predict or evaluate the system state after a specific N-1 event, thereby replacing time-consuming power flow calculations. However, cascading failures in power systems are highly nonlinear physical processes dominated by electromagnetic and electromechanical transient laws on a millisecond timescale. An initial disturbance can instantly change the topology and state of the entire network, causing models trained based on the state before the disturbance to fail instantly. Existing technologies attempt to replace insights into instantaneous causal physical processes with a correlation model based on historical statistics, which essentially ignores the real-time rigid constraints of the power grid as a physical system.

[0005] Therefore, existing technologies, whether purely physical simulation methods or simply data-driven methods, are limited to assessing the direct consequences of a single N-1 event and fail to effectively address the problem of early warning of cascading failure risks. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides an overload analysis method based on feeder N-1 risk, which solves the problems of slow calculation, isolated risk assessment, and failure to effectively reveal cascading failure paths in existing N-1 risk analysis methods.

[0007] The present invention adopts the following technical solution: An overload analysis method based on feeder N-1 risk is proposed. This method is implemented through a physical-topology dual-core driven adaptive risk detection framework. After simulating an initial N-1 emergency event, the method not only calculates traditional, directly related major risk indicators such as overload, but also innovatively calculates a secondary vulnerability indicator for other components in the power grid that have not directly experienced failure. This indicator does not assess whether the current state exceeds the limit, but rather proactively quantifies the stress critical state that these components enter due to the system state change caused by the initial disturbance, as well as their sensitivity or attractiveness to the next subsequent disturbance. In this way, the present invention shifts the focus of analysis from whether the system is currently safe to where it is most likely to become unsafe in the future, thereby revealing the highest probability cascading failure propagation path.

[0008] Preferably, the method employs a dual-core computing strategy to balance real-time performance and accuracy. The first core is a rapid global risk scan, which does not perform time-consuming deterministic power flow calculations. Instead, it abstracts the power grid model into a physical empowerment graph and uses efficient graph theory algorithms or pre-trained graph neural network models to quickly identify several suspected areas with the most drastic changes in topological importance or stress after the redistribution of power flow paths across the entire network due to an initial fault within sub-second time. The second core is a focused detailed investigation of high-risk areas, which only initiates computationally expensive and precise physical model analysis for a few key components selected by the first core to accurately verify and quantify the vulnerability of these components.

[0009] Preferably, the quantitative calculation of the secondary vulnerability index comprehensively considers three major factors: first, the change in topological importance, which quantifies the structural importance of the component to global power flow by calculating the dynamic power flow weighted betweenness centrality of the component in the new topology after the fault; second, the dramatic change in physical stress, which captures the physical impact directly caused by the disturbance by calculating the absolute changes in key electrical state variables such as power flow, voltage, and power angle of the component before and after the fault; and third, the existing safety margin, which achieves a nonlinear weighting by dividing the change in physical stress by the safety margin of the component before the fault, so that when the same disturbance is applied to a component with a smaller margin, it will produce a beacon with exponentially increased strength.

[0010] Preferably, to further enhance the depth of early warning, the detailed investigation phase of high-risk areas employs quasi-dynamic physical stress gradient analysis. This analysis goes beyond simply calculating a final static stable state; instead, it simulates the system's dynamic response process within hundreds of milliseconds after a fault by solving a simplified set of differential-algebraic equations. Thus, it not only obtains the final stress change value but also calculates the rate or gradient of stress change. By introducing this stress gradient as an enhancing factor into the calculation of secondary vulnerability indicators, this invention can identify dynamic instability risks, including power angle instability and voltage collapse, earlier and more accurately.

[0011] Preferably, the method encapsulates the analysis results in a structured luminescent risk package data structure. This data package not only contains traditional overload data, but more importantly, it contains the most critical cascading failure paths composed of all high-intensity secondary vulnerability indicators. In addition, the data package also contains a comprehensive luminescent risk package index, which not only considers the beacon strength of the weakest link in the risk chain, but also introduces the destructive force reflecting the length of the cascading path and the race relationship between risk propagation time and scheduling intervention time, thereby condensing the peak, breadth, and time dimensions of risk into a single quantifiable and comparable benchmark.

[0012] The present invention has at least the following beneficial effects: This invention proactively detects and reveals the next component most likely to fail by introducing the concept of secondary vulnerability indicators. This transforms analysis from static verification to dynamic prediction of system vulnerability chains, significantly improving the early warning capability for cascading failures compared to traditional methods. By employing a physical-topology dual-core driven architecture, it avoids indiscriminate and time-consuming detailed calculations on all network components. Instead, it precisely allocates computing resources to a few high-risk targets identified by topology analysis, achieving an order-of-magnitude improvement in computational efficiency without sacrificing physical accuracy, thus meeting the needs of real-time scheduling decisions. By outputting structured, luminescent risk packages containing risk paths, risk levels, and risk narratives, it provides a unified measure of risk severity, indicating the focus of intervention for dispatchers and lowering the cognitive threshold for decision-making, greatly improving human-computer interaction efficiency and decision accuracy. Furthermore, by introducing quasi-dynamic analysis and stress gradient concepts, it can capture and assess rapid dynamic processes brought about by new energy sources and power electronic equipment, effectively warning of risk types that traditional steady-state analysis methods cannot cover, such as power angle instability, significantly improving the reliability of analysis in the context of new power systems. Attached Figure Description

[0013] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0014] Figure 1 A structural block diagram of a system architecture provided in one embodiment of the present invention; Figure 2 A flowchart illustrating a core processing method provided in one embodiment of the present invention; Figure 3 This is a block diagram of the internal structure of a BRP dual-core risk calculation engine provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of the data structure of a light emission risk packet provided in one embodiment of the present invention; Figure 5 This is a schematic diagram of the user interface of a risk visualization and decision support module provided in one embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] Please see Figure 1 As shown, an embodiment of the present invention provides an overall architecture of a system 100. The system is deployed in a centralized server cluster of a power grid dispatch center. It includes a data acquisition and integration module 101, a real-time state estimation and model calibration module 102, a system orchestration and task management module 103, a BRP dual-core risk calculation engine 104, a risk visualization and decision support module 105, and an intervention measure recommendation and simulation module 106. Its core objective is to achieve second-level early warning of potential cascading failure risks after an N-1 fault.

[0017] The data acquisition and integration module 101 collects operational data from the SCADA, WAMS, and EMS systems of all substations in the network at a preset time period using a standard communication protocol.

[0018] The protocol parsing submodule 101a inside module 101 is used to decode heterogeneous data streams from different sources into a unified internal standard data frame structure. After receiving the data packet, the data alignment and buffering submodule 101b immediately timestamps it with the synchronous phasor measurement unit data from other key measurement points in the network. The alignment accuracy meets the preset requirements to ensure the cross-sectional consistency of subsequent state estimation.

[0019] The pre-processed data is sent to the real-time state estimation and model calibration module 102. The core of the module 102 is the state estimation solver submodule 102b based on the weighted least squares method. After receiving the data and its synchronous dataset, the data cleaning and verification submodule 102a is started first. It is used to preprocess the input data through the Kalman filter algorithm to filter out random noise and to remove bad data using preset criteria.

[0020] The solver submodule 102b loads a pre-stored power grid topology model based on a data snapshot of the instant before the fault occurs, and performs state estimation calculations. The convergence criterion for the calculation is set as the maximum power imbalance of all nodes in the network being less than a preset convergence threshold. After several iterations of calculation, the module 102 successfully generates a high-fidelity power grid model snapshot object, which accurately reproduces the power grid state before the fault occurs. The object is stored in an in-memory database, and its handle is pushed to the system orchestration and task management module 103.

[0021] The system orchestration and task management module 103 serves as the central processing unit of the system. Its internal event-driven engine submodule 103a continuously monitors state updates from module 102. When it detects an update to the power grid model snapshot and parses out a state change event of a key component, it immediately classifies it as a critical N-1 event. The task generation and distribution submodule 103b is then activated, creating a computational task object containing the associated power grid snapshot identifier and a list of single faults, setting the analysis depth to full analysis. After serialization, the task object is sent to the BRP dual-core risk calculation engine 104 via an application programming interface.

[0022] Please see Figure 3As shown, the internal structure of the engine 104 is illustrated. After receiving a task, the task is sent to the physical-topology graph constructor submodule 301. The submodule reads the specified power grid snapshot and constructs a post-fault power grid topology graph in memory based on the fault information. The weight of each edge in the graph is dynamically coupled with its real-time physical margin. Subsequently, the first core processing stage, namely the global risk fast scan submodule 302, is entered. The role of this submodule is to quickly and globally identify the region where the risk is most likely to converge. It does not perform time-consuming power flow calculations, but instead uses graph theory algorithms, specifically parallel optimized algorithms, to calculate the dynamic power flow weighted betweenness centrality of all elements in the post-fault topology.

[0023] Furthermore, the engine 104 enters the second core processing stage, namely the high-risk area focused detailed investigation submodule 303. This submodule receives the global scan results from the submodule 302 and applies a preset beacon strength threshold for filtering. It only initiates precise physical stress analysis for those components with drastic changes in betweenness centrality and high initial load. For each locked component, the submodule 303 performs a fast decoupling power flow method to calculate the precise state after the fault, thereby obtaining the absolute changes in various physical stresses on the component.

[0024] At this point, the calculation of the secondary vulnerability index, namely the vulnerability attraction beacon strength, is invoked. Its calculation comprehensively considers the dynamic power flow weighted betweenness centrality, the absolute changes of various physical stresses, and the existing safety margin of the components, and is coupled through a nonlinear activation function. Finally, the BRP synthesizer submodule 304 is invoked, which is used to trace from high to low beacon strength to construct the most critical cascading failure path. Then, it uses a comprehensive risk index calculation formula to calculate the comprehensive risk index. The input of the formula includes the maximum beacon strength, the cascading failure path length, the risk aversion coefficient, the estimated risk propagation time, and the scheduling intervention response time constant. A natural language generator based on a predefined template creates a risk narrative to describe the nature and transmission logic of the risk in human-readable language.

[0025] Please see Figure 4 As shown, it illustrates the data structure of a complete BRP object. After the object is constructed, it is pushed to the risk visualization and decision support module 105 in real time through an interface. In the module 105, a rendering engine receives the BRP object.

[0026] Please see Figure 5As shown, a user interface diagram is displayed. On the large screen 500 in the dispatch center, the icon 501 representing the faulty component is displayed as disconnected, while the cascading failure path composed of high-risk components 502 and 503 is visually alerted at a specific frequency, the alert frequency being proportional to the comprehensive risk index. The risk narrative and comprehensive risk index are displayed in the information panel 504.

[0027] The BRP object is also synchronously sent to the intervention recommendation and simulation module 106. Based on the risk issues revealed in the most critical cascading failure path, the module 106 queries its internal control-response sensitivity matrix to find the control resources that have the most significant impact on the power flow of the risky line. The module 106 then generates a specific intervention suggestion and gives the expected control effect. The dispatcher can adopt the suggestion and issue instructions through the energy management system, thereby forming a closed-loop control from risk identification to effective intervention.

[0028] In a preferred embodiment, the global risk fast scanning submodule 302 can employ a pre-trained graph convolutional neural network model, namely a GCN topology sensitivity predictor. The GCN model contains several graph attention convolutional layers and uses a preset activation function. The model is trained on a dataset containing a large number of offline simulated fault scenarios. Its training objective is to take the pre-fault topology, fault information, and pre-fault state as input and directly output a vector. Each element of the vector corresponds to a power grid element, and its value is the predicted value of the element's dynamic power flow weighted betweenness centrality after the fault. When the post-fault topology graph is fed into the GCN model, it performs a forward propagation on the graphics processor, which outputs the predicted centrality values ​​of all elements in the entire network.

[0029] As a preferred implementation, the high-risk area focused detailed investigation submodule 303 may employ a quasi-dynamic physical stress gradient analyzer. This analyzer is used to capture the dynamic process within a short period after a fault. Upon receiving a list of selected high-risk components, for each component, it solves a simplified set of differential-algebraic equations to simulate the system's dynamic response within a specific time window after the fault. By performing numerical integration with a preset step size, the analyzer can not only obtain the final state but also calculate the rate of change of each physical stress, i.e., the stress gradient. This gradient information is used to correct the calculation of the vulnerability attraction beacon strength. Specifically, the original beacon strength is multiplied by an enhancement term related to the normalized stress gradient norm and the gradient enhancement coefficient.

[0030] In a preferred implementation, the system can be deployed in a distributed system based on edge computing. In this architecture, the system consists of a central cloud coordinator and several edge computing nodes deployed in key substations, forming a hierarchical federated system. Each edge computing node is a simplified risk detection unit, while the central cloud coordinator is responsible for the integration and in-depth analysis of the global situation. When a local failure occurs, the edge computing node does not need to wait for central instructions and immediately triggers a local BRP calculation task on its own node to perform rapid analysis on its neighborhood power grid model and generate a preliminary risk packet for reporting. The central cloud coordinator is responsible for aggregating local intelligence from all edge computing nodes in the entire network, combining it with the global power grid model for in-depth analysis, and piecing together a complete panoramic view of cascading failures.

[0031] To verify the performance and robustness of the method described in this invention under extreme conditions, this embodiment further analyzes two typical extreme scenarios.

[0032] The first scenario involves system disturbances under high-proportion renewable energy penetration. In this scenario, a large photovoltaic power station experiences a significant and rapid drop in output power due to cloud cover. While the drop itself does not constitute an N-1 fault, it injects a substantial power disturbance into the system. The quasi-dynamic physical stress gradient analyzer in the method described in this invention can effectively capture this rapid change process. Specifically, it calculates the power flow rate of critical branches and the power angle rate of generator units by solving a simplified system of differential-algebraic equations.

[0033] The results show that even if no branch experiences immediate steady-state overload after the disturbance, the stress gradients of several series compensation devices and long-distance transmission lines located in power flow channels significantly exceed the safety threshold. The method described above, by introducing this stress gradient as an enhancement factor into the calculation of the vulnerability attraction beacon, can identify these dynamically stressed components in advance and mark them as high-risk units. This demonstrates that the method described in this invention can effectively provide early warning of transient stability risks caused by the volatility of new energy sources, which are difficult to detect using traditional steady-state analysis methods. This provides dispatchers with valuable pre-control time to take measures such as starting standby units or adjusting energy storage output to maintain system stability.

[0034] To further refine this process, the quasi-dynamic physical stress gradient analyzer, when handling this scenario, incorporates a simplified model of fast power control for the photovoltaic inverter and a second-order classical model of the synchronous generator set into its internal dynamic model. When a photovoltaic power drop event is detected, the analyzer initializes the system's differential-algebraic equations, where the injected power of the photovoltaic power station is modeled as a ramp function, the slope of which corresponds to the gradient of the power drop. The system is then numerically integrated using methods such as the Euler method or the fourth-order Runge-Kutta method, with a time step set to, for example, 10 milliseconds, and the simulation is sustained for 500 milliseconds. At each integration step, the system records the power flow values ​​of key transmission sections and the power angle of the synchronous generator set.

[0035] After the simulation, the rate of change of each quantity, i.e., the stress gradient, is obtained by differentiating these time-series data. For example, if the power flow of a certain line increases linearly from 500 MW to 550 MW within 200 milliseconds, its stress gradient is quantified as 250 MW / s. This gradient value is then normalized and multiplied by a preset gradient enhancement coefficient, such as 1.5, and finally used as an enhancement factor in calculating the secondary vulnerability index of the line. This significantly improves its risk level based on the static analysis, thereby ensuring the effective capture of such rapid dynamic processes.

[0036] The second scenario involves network attacks targeting the measurement system. In this scenario, attackers simulate injecting forged measurement data into the state estimation system through network penetration techniques. This is primarily to mask the actual overload situation of a heavily loaded line or to fabricate a non-existent line tripping event. The method described in this invention demonstrates inherent robustness against attacks that forge line tripping logs. When the system receives forged line tripping information and initiates BRP analysis, its physical-topology dual-core driven architecture provides cross-validation.

[0037] Specifically, the global risk rapid scanning submodule, whether based on graph theory algorithms or a pre-trained GCN model, predicts a theoretical power flow transfer distribution and risk convergence area based on the fabricated topology changes. However, when the high-risk area focused detailed investigation submodule performs precise physical stress analysis using tamper-proof real-time synchronized phasor data from the entire network's PMUs, it discovers a significant discrepancy between the actual system state and the theoretical prediction. For example, parallel lines that were predicted to handle a large amount of transferred power flow show no significant change in their measured power flow values. This mismatch between prediction and measurement will be identified by the system as a high-confidence anomaly.

[0038] The BRP synthesizer submodule quantifies this deviation into a physical consistency verification index. When the index is lower than a preset threshold, it not only suppresses conventional cascading fault alarms but also generates a higher-level measurement system reliability alarm, prompting the dispatcher to pay attention to the authenticity of the data source. This proves that the method described in this invention can not only analyze physical faults but also utilize its inherent physical-topology coupling model to have a certain perception and defense capability against attacks at the information layer, thereby improving the depth of information-physical fusion security protection of the power system.

[0039] To make this mechanism more concrete, the physical consistency verification index can be calculated based on the weighted sum of squared residuals between the theoretically calculated state and the actual measured state. For each key component that is focused on for detailed investigation, the system calculates the difference between the theoretical predicted values ​​of its power flow, voltage, and other state quantities and the actual values ​​obtained through high-precision measurement equipment such as PMUs. The difference is then squared and normalized by dividing by the variance of the measured value. After weighted summation of the normalized residuals of all key components, a comprehensive deviation metric is obtained.

[0040] The physical consistency verification index is defined as the reciprocal of this deviation metric or a function negatively correlated with it. When the system receives forged data, the large-scale mismatch between theoretical prediction and actual measurement will cause the deviation metric to increase sharply, thereby causing the physical consistency verification index to drop to an extremely low level, such as below the preset threshold of 0.5, thus triggering a measurement system credibility alarm. This mechanism can not only identify simple tripping forgeries, but also detect abnormal trends in the physical consistency verification index that deviate from the normal fluctuation range by performing time series analysis on more complex, covert attacks that are mainly aimed at slowly manipulating the system state estimation results, thereby realizing the perception of deep network attacks.

[0041] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An overload analysis method based on feeder N-1 risk, characterized in that, The method includes the following steps: Step S1: In the computing device, simulate an initial emergency event in which a preset initial component in the power system fails; Step S2: Based on the simulation of the initial emergency event, determine at least one major risk indicator directly related to the initial emergency event; Step S3: Based on the change in the state of the power system caused by the initial emergency event, calculate a secondary vulnerability index for at least one other component in the power system that has not experienced a fault. The secondary vulnerability index is used to quantify the sensitivity of the other component to a subsequent disturbance. Step S4: Encapsulate the primary risk indicators and the secondary vulnerability indicators in an emergency response data packet structure to identify cascading failure paths.

2. The overload analysis method based on feeder N-1 risk according to claim 1, characterized in that, The secondary vulnerability index includes an attraction index, which is calculated based on the magnitude of changes in one or more electrical state variables of the other components before and after the initial emergency event, and weighted according to the proximity of the one or more electrical state variables to the preset operating limits of the other components.

3. The overload analysis method based on feeder N-1 risk according to claim 1, characterized in that, The method is executed through a dual-core computing strategy, which includes: a global risk rapid scanning core for quickly identifying suspected areas without performing precise power flow calculations, and a high-risk area focused detailed investigation core that initiates precise physical model analysis only for components within the suspected areas to quantify the secondary vulnerability indicators.

4. The overload analysis method based on feeder N-1 risk according to claim 2, characterized in that, The calculation of the attraction index further includes: calculating the dynamic power flow weighted betweenness centrality change of the other components in the post-failure topology, and using the change as a weighting factor to correct the attraction index.

5. The overload analysis method based on feeder N-1 risk according to claim 2, characterized in that, The method further includes a quasi-dynamic physical stress gradient analysis step, which calculates the rate of stress change by solving a simplified set of differential algebraic equations and uses the rate as an enhancement factor to correct the attraction index in order to identify the risk of dynamic instability.

6. The overload analysis method based on feeder N-1 risk according to claim 1, characterized in that, The emergency response data packet structure is a structured data object that includes at least: a list of the main risk indicators, one or more sequentially arranged cascading failure paths composed of higher-level vulnerability indicator components, and a comprehensive risk index.

7. The overload analysis method based on feeder N-1 risk according to claim 6, characterized in that, The method further includes at least one of the following steps: generating a risk heatmap on the topology of the power system based on the secondary vulnerability index for visualization; or simulating a corrective measure for the power system and recalculating the secondary vulnerability index to evaluate the effectiveness of the corrective measure.

8. A system for performing an overload analysis method based on feeder N-1 risk according to any one of claims 1-7, characterized in that, The system includes: one or more processors; and a memory storing instructions that, when executed by the one or more processors, enable the system to perform an overload analysis method based on feeder N-1 risk according to any one of claims 1-7.

9. The system of the overload analysis method based on feeder N-1 risk according to claim 8, characterized in that, The instructions further enable the system to: generate a risk heatmap of the power system and visualize the risk heatmap on a user interface, wherein the secondary vulnerability indicators are overlaid on the topology of the power system.

Images