Power system comprehensive monitoring and diagnosis method based on electrical signal processing algorithm

Abstract

The application relates to the technical field of power systems, and discloses a power system comprehensive monitoring and diagnosis method based on an electrical signal processing algorithm, which comprises the following steps: abstracting a physical connection of a power system into an electrical water surface topology structure, establishing a three-dimensional electrical coordinate system to generate an electrical breathing track and an electrical breathing envelope surface of equipment, generating a power layered health ring based on the envelope surface; detecting node abnormities according to the health ring, triggering an electrical ripple detection mechanism to construct a ripple time difference matrix, and determining a fault source position; constructing a fault symptom jigsaw puzzle template library and symptom block association rules, matching a current symptom sequence, reasoning about missing symptoms, resolving diagnosis conflicts, and outputting a diagnosis result; and the application can realize the transformation of a power system from traditional post-maintenance to predictive maintenance.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and more specifically, to a method for comprehensive monitoring and diagnosis of power systems based on electrical signal processing algorithms. Background Technology

[0002] As modern power systems develop towards high voltage, large capacity, and multi-energy grid integration, the operating status of core equipment such as transformers, circuit breakers, busbars, and cables directly determines the safety and stability of the power grid. Real-time monitoring and fault diagnosis have become key links in ensuring reliable system operation and reducing operation and maintenance costs. However, the operating conditions of power systems are becoming increasingly complex. Factors such as load mutations (e.g., large motor starting, electric arc furnace switching), grid disturbances (e.g., lightning strikes, short-circuit faults), and equipment aging are intertwined, causing "transient-harmonic-load" triple nonlinear coupling distortion in multidimensional electrical signals such as voltage, current, and power factor. Traditional monitoring and diagnostic methods based on steady-state analysis or single frequency domain characteristics are no longer suitable for this complex scenario.

[0003] In the prior art, Chinese patent application CN117333028A discloses an evaluation system and method based on the deployment characteristics of a power system. This system collects various data from the power system, processes them, and then constructs a three-dimensional visualized power grid topology model based on deep learning to visualize the characteristics of deployed nodes. Simultaneously, it calculates node parameter evaluation coefficients to generate a attention table, compares the evaluation coefficients with risk thresholds, and issues an anomaly warning if the threshold is exceeded. During the warning, the corresponding node flashes to assist management personnel in locating and repairing the fault. Chinese patent application CN119598247A discloses an intelligent fault diagnosis method and system for power systems. This system collects waveform data from continuous sampling points under different operating conditions according to a preset fault start time and collection time window, and determines a fault matrix. It adds fault type label vectors to the fault matrix to construct a dataset, divides it into training and testing sets, embeds a CNN model within the training rules, and deploys the trained model to the power system. It collects fault data and diagnoses fault types during current surges.

[0004] However, the aforementioned existing technologies still have insurmountable technical limitations: CN117333028A relies on steady-state node parameters to calculate evaluation coefficients, failing to consider the dynamic impact of "transient-harmonic-load" coupling distortion on the parameters. When sudden load changes occur, such as the starting of a large motor, transient components can cause node parameters to deviate from the normal range in the short term, leading to false triggering of warnings by the evaluation coefficients, i.e., "false fault activation." Furthermore, the subtle parameter changes caused by early faults such as equipment insulation degradation can be masked by harmonics and load fluctuations, making it impossible to identify the evaluation coefficients, i.e., "fault feature submersion." Moreover, it does not establish a correlation between parameters and the location of the fault source, requiring blind investigation even after a warning; CN119598247A adopts... Acquiring waveform data using a fixed sampling rate cannot dynamically decouple transient processes from steady-state characteristics. For nanosecond-level voltage spike precursor information in millisecond-level transients such as circuit breaker contact pre-breakdown, fixed sampling is difficult to capture. Furthermore, its fault diagnosis does not combine the spatial location of the fault source to screen symptom sequences. When load fluctuations such as distributed photovoltaic power oscillations cause multi-node transients, normal load transients are easily misjudged as equipment faults. At the same time, it is impossible to exclude physically impossible fault types through location information, further exacerbating diagnostic biases. The above problems together lead to low fault diagnosis accuracy and early fault detection rate under complex operating conditions, becoming a core technical obstacle restricting the transformation of power systems from "reactive maintenance" to "predictive maintenance". Summary of the Invention

[0005] To overcome the aforementioned deficiencies in existing technologies, this invention provides a comprehensive monitoring and diagnosis method for power systems based on electrical signal processing algorithms. By constructing an electrical surface topology and a power hierarchical health loop, it achieves dynamic decoupling of "transient-harmonic-load" coupled signals. Combined with an electrical ripple detection mechanism and symptom mosaic matching logic, it accurately captures fault precursor information and associates it with the location of the fault source. This effectively avoids the phenomena of "fault feature flooding" and "false fault activation," significantly improving the accuracy of fault diagnosis and the early fault detection rate under complex operating conditions. It provides key technical support for the transformation of power systems from "reactive maintenance" to "predictive maintenance."

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A comprehensive monitoring and diagnosis method for power systems based on electrical signal processing algorithms includes:

[0008] The physical connections of the power system are abstracted into an electrical surface topology, a three-dimensional electrical coordinate system is established, and electrical breathing trajectories of the equipment are generated in the three-dimensional electrical coordinate system. An electrical breathing envelope is generated based on the electrical breathing trajectories, and a power-layered health ring is generated based on the electrical breathing envelope.

[0009] Anomaly detection is performed on nodes in the electrical surface topology based on the power-layered health ring. The anomaly detection result triggers the electrical ripple detection mechanism to construct the ripple time difference matrix. The location of the fault source is determined based on the ripple time difference matrix.

[0010] Construct a fault symptom puzzle template library and symptom block association rules. Obtain the current symptom sequence based on the fault source with a determined location. Match the current symptom sequence with the fault symptom puzzle template library to calculate the matching degree. Based on the matching degree and symptom block association rules, perform missing symptom block reasoning and diagnostic conflict resolution, and output the fault diagnosis result.

[0011] Furthermore, in the described electrical surface topology, devices are the nodes, and the electrical connections between devices are the propagation paths.

[0012] Furthermore, the method for generating the electrical breathing trajectory of the device in a three-dimensional electrical coordinate system includes:

[0013] Monitoring points are set at each node of the electrical water surface topology. Based on the set monitoring points, the periodic electrical parameters of the equipment are collected at preset time intervals during the complete operation cycle of the equipment. The periodic electrical parameters are converted into coordinate points in a three-dimensional electrical coordinate system. The coordinate points of adjacent time moments are connected to generate the electrical breathing trajectory of the equipment.

[0014] Furthermore, the three-dimensional electrical coordinate system takes the rated power point of the equipment as the origin O, the X-axis is defined as the active power deviation, the Y-axis is defined as the reactive power deviation, and the Z-axis is defined as the harmonic distortion rate.

[0015] Furthermore, the method for generating the electrical breathing envelope based on the electrical breathing trajectory includes:

[0016] Normal operation trajectory points are extracted from the electrical breathing trajectory to form a normal point set. The three-dimensional coordinates of the normal point set are input into a three-dimensional convex hull algorithm. The three-dimensional convex hull algorithm calculates the convex hull of the normal point set and outputs the vertex coordinates that constitute the convex hull. Then, a closed convex polyhedron surface is constructed based on the vertex coordinates. The convex polyhedron surface is the electrical breathing envelope surface.

[0017] Furthermore, the method for generating a power-layered health ring based on the electrical breathing envelope includes: slicing the electrical breathing envelope according to a percentage of the device's rated power to form multiple power slices, and generating a power-layered health ring on each power slice.

[0018] Furthermore, the method for generating the power slice includes: determining the active power value corresponding to the slice based on the rated power percentage, converting the active power value corresponding to the slice into a fixed X-axis coordinate in a three-dimensional electrical coordinate system, and constructing a plane perpendicular to the X-axis of the three-dimensional electrical coordinate system using the fixed X-axis coordinate. The plane is the power slice corresponding to the rated power percentage.

[0019] Furthermore, the method for anomaly detection of nodes in the electrical water surface topology includes:

[0020] Acquire instantaneous electrical parameters collected at monitoring points. Based on the instantaneous electrical parameters and the health loop of power stratification, determine whether there are any abnormalities in the nodes of the electrical water surface topology and define the nodes with abnormalities as abnormal nodes.

[0021] The method for determining whether there are anomalies in the nodes of the electro-hydraulic surface topology is as follows: determine the power level corresponding to the instantaneous electrical parameter point of the node to be determined, find the health loop parameter corresponding to the power level, calculate the distance d from the instantaneous electrical parameter point to the center of the corresponding power slice surface, and if r in ≤d≤r out If d <r in or d>r out If the condition is met, the node is determined to be an abnormal node; where the instantaneous electrical parameter point is the spatial coordinate point corresponding to the instantaneous electrical parameter in the three-dimensional electrical coordinate system, and the health ring parameter includes the inner diameter r of the health ring. in With outer diameter r out .

[0022] Furthermore, the triggering condition for the electrical ripple detection mechanism is the detection of an anomaly at any node in the electrical water surface topology.

[0023] Furthermore, the method for determining the location of the fault source includes:

[0024] Velocity calibration is performed on each propagation path in the electrical water surface topology to construct a ripple propagation velocity model;

[0025] The location of the fault source is determined based on the ripple time difference matrix and the ripple propagation velocity model.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] This invention provides a systematic solution to the diagnostic problems caused by multidimensional electrical signal coupling distortion in power systems under complex operating conditions by constructing a complete technical system of "health benchmark construction - fault source location - fault diagnosis". It abstracts the physical connections of the power system into an electrical surface topology, and combines the electrical breathing trajectory and electrical breathing envelope generated by the three-dimensional electrical coordinate system to integrate multidimensional electrical signals. Through a power-layered health loop, it dynamically adapts to different load fluctuation scenarios, avoiding the "pseudo-fault activation" problem that easily occurs in traditional fixed threshold or single steady-state analysis methods. Based on the node anomaly detection of the power-layered health loop, it links the electrical ripple detection mechanism with the ripple time difference. The matrix can accurately capture the fault precursor information contained in the transient process, overcoming the limitations of traditional methods that are difficult to decouple transient and steady-state characteristics and cannot effectively identify weak fault signals. It reduces the phenomenon of "fault feature overload" and achieves accurate location of fault sources. Through the fault symptom mosaic template library and symptom block association rules, it performs matching calculations, missing symptom inferences and diagnostic conflict resolution on the current symptom sequence obtained based on the fault source location. This can improve the fault diagnosis accuracy and early fault detection rate in multi-dimensional signal coupling scenarios, provide reliable protection for the stable operation of power grid equipment, and ultimately promote the transformation of power system operation and maintenance mode from "post-event maintenance" to "predictive maintenance". Attached Figure Description

[0028] To more clearly illustrate the technical solutions 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.

[0029] Figure 1 This is a flowchart illustrating the principle of the integrated monitoring and diagnosis method for power systems based on electrical signal processing algorithms in this invention.

[0030] Figure 2 A flowchart illustrating the principle of determining whether there are abnormalities in nodes in an electrical water surface topology, provided in an embodiment of the present invention;

[0031] Figure 3 A flowchart illustrating the principle of determining the location of a fault source, provided in an embodiment of the present invention;

[0032] Figure 4 This is a functional block diagram of the power system integrated monitoring and diagnosis system based on electrical signal processing algorithms in this invention. Detailed Implementation

[0033] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Example 1:

[0035] Please see Figure 1 As shown, this embodiment provides a comprehensive monitoring and diagnosis method for power systems based on electrical signal processing algorithms, including:

[0036] Step S10: Abstract the physical connection relationship of the power system into an electrical surface topology, establish a three-dimensional electrical coordinate system, generate the electrical breathing trajectory of the equipment in the three-dimensional electrical coordinate system, generate the electrical breathing envelope based on the electrical breathing trajectory, and generate a power-layered health ring based on the electrical breathing envelope.

[0037] Further, step S10 includes:

[0038] Step S11: Based on the actual wiring diagram of the power system, the physical connection relationship of the power system is abstracted into an electrical surface topology. In the electrical surface topology, the equipment is a node and the electrical connection between the equipment is a propagation path. A monitoring point is set on each node of the electrical surface topology.

[0039] Electrical surface topology refers to a graph structure model in a power system where various electrical devices are defined as nodes, and the electrical connections between devices are defined as propagation paths. Electrical devices include transformers, circuit breakers, and busbars, while electrical connections include cables and conductors. The abstraction process of electrical surface topology requires first reviewing power system design drawings, equipment ledgers, and conducting on-site surveys to clarify the physical location, model specifications, and connection relationships of each device. Then, graph theory modeling methods are used to map each device as a vertex (node) in the graph, and the connections between devices as edges (propagation paths). Monitoring points are then set on each electrical device corresponding to each node. The selection of monitoring point locations must be based on the key parameter acquisition requirements of the equipment. For example, monitoring points for transformers are set at the high-voltage and low-voltage side outgoing terminals to ensure that electrical parameters reflecting the core operating status of the equipment can be collected. Step S11 aims to address the lack of spatial correlation between devices and the inability to trace fault propagation paths in traditional monitoring methods. By simplifying the complex power network into a computable graph structure, it not only provides a mathematical basis and spatial reference for the electrical ripple propagation analysis in the subsequent step S20, but also achieves precise binding between device monitoring points and physical locations. This ensures that all subsequently collected electrical parameters can be mapped to specific devices and connection paths, avoiding the chaotic fault tracing caused by the disconnect between parameters and device locations in traditional monitoring. Without this step, the subsequent step S20, which constructs the ripple time difference matrix, will lack spatial references for the relative positions of nodes, making it impossible to determine the propagation path length and connection relationships between nodes. This would prevent the hyperbolic localization algorithm from drawing hyperbolas and calculating intersections, thus creating a barrier between the technical features of fault source localization. This barrier stems from the lack of a unified topological spatial framework, making it impossible to correlate the ripple propagation velocity model with the time difference matrix, ultimately preventing the fault source localization function from being realized.

[0040] Step S12: Establish a three-dimensional electrical coordinate system. Based on the set monitoring points, collect the periodic electrical parameters of the equipment at preset time intervals during the complete operation cycle of the equipment. Convert the periodic electrical parameters into coordinate points in the three-dimensional electrical coordinate system and connect the coordinate points at adjacent time points to generate the electrical breathing trajectory of the equipment.

[0041] Step S12: Based on the electrical surface topology constructed in Step S11, a three-dimensional electrical coordinate system is further established and an electrical breathing trajectory is generated. The establishment of the three-dimensional electrical coordinate system requires the rated power point of the equipment as the origin O. The X-axis is defined as the active power deviation, reflecting the difference between the actual output active power and the rated active power of the equipment. The Y-axis is defined as the reactive power deviation, reflecting the difference between the actual output reactive power and the rated reactive power of the equipment. The Z-axis is defined as the harmonic distortion rate, reflecting the degree of power quality distortion, where the harmonic distortion rate is the voltage harmonic distortion rate. The determination of each axis parameter requires first obtaining the rated active power P of the equipment. rated With rated reactive power Qrated P rated and Q rated Harmonic distortion rate (THD) is obtained from the equipment's nameplate, technical specifications, or power system equipment database. It is calculated by collecting voltage or current harmonic components during equipment operation using a harmonic analyzer. During the complete operating cycle from no-load to full-load, periodic electrical parameters are collected at preset time intervals T. These parameters include active power P, reactive power Q, and THD. The preset time interval T is dynamically adjusted based on the equipment type, specifically the frequency of load changes and parameter fluctuation characteristics. For example, for transformers with slower load changes, a 5-minute interval captures periodic parameter changes while avoiding data redundancy; for circuit breakers with frequent operations and significant parameter fluctuations during operation, a 1-minute interval captures sudden parameter changes during operation; and for busbars with relatively stable loads, a 10-minute interval balances monitoring efficiency and data storage costs. The collected periodic electrical parameters need to be converted to coordinates in a three-dimensional electrical coordinate system. The conversion formula is: X-coordinate equals measured active power minus rated active power P. rated The Y-coordinate equals the measured reactive power minus the rated reactive power Q. rated The Z-coordinate equals the measured harmonic distortion rate. The physical meaning of the above conversion formula is as follows: when the X-coordinate is positive, it indicates that the actual active power of the equipment is higher than the rated value, and there may be an overload risk; when the X-coordinate is negative, it indicates that the equipment is in a light-load state; the sign of the Y-coordinate reflects whether the reactive power supply and demand of the equipment are balanced; the magnitude of the Z-coordinate directly reflects the quality of power, and the larger the value, the more serious the harmonic pollution. After the coordinate points are generated, the coordinate points at adjacent times need to be connected with straight lines to form an electrical breathing trajectory. Step 12 aims to address the problem that traditional monitoring methods often use abstract electrical parameters that fail to intuitively reflect the equipment's operating status. By concretizing these abstract parameters into spatial coordinate points and continuous trajectories, the visual tracking of the equipment's operating status is achieved. The monitoring points set in step S11 provide clear physical locations for parameter acquisition in this step, ensuring that the acquired active power, reactive power, and harmonic distortion rate can accurately correspond to specific nodes in the electrical surface topology, avoiding mismatch between parameters and equipment. Without this step, subsequent step S13 would lack the trajectory point data required to generate the electrical breathing envelope, making it impossible to establish the initial baseline for the health status. Simultaneously, the power-layered health loop in step S14 would also fail to generate due to the lack of coordinate point support, thus causing the entire monitoring and diagnostic system to lose its core data foundation.

[0042] Step S13: Generate the electrical breathing envelope surface using a three-dimensional convex hull algorithm based on the electrical breathing trajectory;

[0043] The 3D convex hull algorithm refers to an algorithm that constructs a minimal convex polyhedron surface that completely contains all normal operating trajectory points by calculating the convex hull vertices of the normal operating trajectory points in the electrical breathing trajectory. The specific implementation process is as follows: Normal operating trajectory points are extracted from the electrical breathing trajectory to form a normal point set. The 3D coordinates of the normal point set are input into a 3D convex hull algorithm (such as the Qhull algorithm). The 3D convex hull algorithm calculates the convex hull of the normal point set and outputs the vertex coordinates constituting the convex hull. Then, a closed convex polyhedron surface is constructed based on the vertex coordinates. This convex polyhedron surface is the electrical breathing envelope surface. The selection criteria for normal operating trajectory points are: based on periodic electrical parameters, points exceeding the factory allowable deviation range are removed, such as active power deviation > ±10% Prated, reactive power deviation > ±15% Qrated, and harmonic distortion rate > 5%.

[0044] In traditional monitoring, it is difficult to accurately define the boundaries of electrical parameters for normal equipment operation. On the one hand, equipment operating parameters change dynamically with operating conditions, and fixed thresholds cannot adapt to their individual fluctuation ranges. On the other hand, anomalies caused by instantaneous disturbances can easily mix into normal data, blurring the distinction between normal and abnormal states and failing to provide a reliable benchmark for health status assessment. The 3D convex hull algorithm constructs a minimum convex polyhedron surface, i.e., the electrical breathing envelope, by calculating the convex hull vertices of all normal operating trajectory points in the electrical breathing trajectory. This envelope surface can dynamically adapt to the parameter fluctuation characteristics of the equipment under different operating conditions, replacing traditional fixed thresholds and accurately defining the 3D parameter range for normal equipment operation. This solves the problem that fixed thresholds cannot adapt to the "individualized" electrical characteristics of the equipment, providing a dynamic and accurate benchmark boundary for subsequent health status assessment. The 3D convex hull algorithm generates the envelope surface only based on normal operating trajectory points, naturally excluding anomalies such as jump points caused by instantaneous disturbances. For example, anomalies that deviate from the normal trajectory will not be included in the convex hull vertex calculation. This ensures that the envelope surface accurately reflects the parameter distribution characteristics during stable equipment operation, avoiding the problem of distortion of the normal range benchmark caused by the intrusion of abnormal data, and improving the reliability of the health status benchmark. The generated closed electrical breathing envelope surface is a continuous and complete spatial structure, providing a unified spatial carrier for slicing according to the rated power percentage in step S14. Only based on this complete envelope surface can accurate slicing at different power levels be achieved, thereby generating health rings for each power layer, solving the problem of inconsistent layer benchmarks caused by the lack of a unified normal range boundary, and ensuring the accuracy of subsequent anomaly detection. Step S13 generates the electrical breathing envelope surface through a three-dimensional convex hull algorithm, fundamentally solving the problems of blurred normal parameter boundaries and susceptibility to interference from abnormal points in traditional monitoring, providing an accurate and reliable initial benchmark for the health status of the entire monitoring and diagnostic system.

[0045] Step S14: The electrical breathing envelope is sliced ​​according to the percentage of the device's rated power to form multiple power slices, and a power-layered health ring is generated on each power slice.

[0046] The electrical breathing envelope is sliced ​​according to a percentage of the equipment's rated power. The percentage of rated power set during the slicing process needs to be sufficient to fully cover all load ranges of the equipment, ensuring a corresponding health baseline for each load stage, while also guaranteeing a sufficient number of historical trajectory points within each slice. This avoids overly dense slicing leading to insufficient data in some slices, or overly sparse slicing resulting in incomplete load range coverage. For example, setting the percentage of rated power to 10%, and slicing at 10% intervals of rated power, from 0% rated power (no load) to 100% rated power (full load), results in 11 power slices. The generation process for each power slice is as follows: determine the active power value corresponding to the slice based on the rated power percentage; convert the active power value corresponding to the slice to a fixed X-axis coordinate in a three-dimensional electrical coordinate system; and construct a plane perpendicular to the X-axis of the three-dimensional electrical coordinate system using the fixed X-axis coordinate. This plane is the power slice corresponding to the rated power percentage. For example, the active power value corresponding to the 10% rated power slice is 0.1 × P. rated In a three-dimensional electrical coordinate system, this power slice is perpendicular to the X-axis and passes through X = 0.1 × P. rated -P rated = -0.9×P rated The plane, and the closed curve formed by the intersection of this plane and the electrical breathing envelope plane, is the envelope line under this power.

[0047] On each power slice, the distance distribution from all historical trajectory points to the slice center at the corresponding power level needs to be calculated. Taking slicing at intervals of 10% of the rated power as an example, the coordinates of the slice center are determined as follows: the deviation of the rated active power corresponding to the power is X. center = (k × 10% × P) rated )-P rated Where k is an integer from 0 to 10, representing different power slices, and the rated reactive power deviation is Y. center = (k × 10% × Q) rated )-Q rated The Z-coordinate of the slice center is Z center Z center The average voltage harmonic distortion rate under this power slice during normal operation is calculated from the average harmonic distortion rate of the corresponding power level in historical normal operation data. Therefore, the center coordinate of the slice is (X). center ,Y center r,Z centerThe physical meaning of this is the ideal operating point of the equipment at that power. The distance from the historical trajectory point to the slice center is calculated as a two-dimensional distance. Since the slice surface is perpendicular to the X-axis and the X-coordinate is fixed, only the distance in the Y-axis and Z-axis directions needs to be calculated. The formula is: Where Y i' Z i' Let d be the Y and Z coordinates of the projected coordinates of the historical trajectory point on the power slice plane. i' Let be the distance from the i'th historical trajectory point to the slice center, where i' is the index variable of the historical trajectory point. After the distance distribution is calculated, take the 5th percentile of the distance distribution as the inner diameter r. in The 95th percentile is used as the outer diameter r out The percentile calculation process is as follows: Distance d between all historical trajectory points i' Sort the data in ascending order. Let the total number of points be N. The position corresponding to the 5th percentile is 0.05 × N. Rounding up, the distance to this position is r. in The position corresponding to the 95th percentile is 0.95 × N, which is obtained by rounding up. The distance to this position is r. out The 5% and 95% percentiles were chosen to exclude the influence of extreme outliers on the health loop boundary, ensuring that the health loop covers 90% of the equipment's normal operating conditions and improving the robustness of the benchmark. These extreme outliers include, for example, accidental harmonic exceedances and transient reactive power surges. in and r out The resulting annular region is the health ring at the corresponding power level. The inside of the health ring is the normal operating area, and the outside is the abnormal operating area. Step S14 aims to solve the problem that traditional fixed thresholds cannot adapt to the differences in characteristics of equipment under different loads. By using a power-layered health ring, an anomaly judgment standard that is load-adaptive is achieved. The electrical breathing envelope generated in step S13 provides the basis for slicing in step S14. The integrity of the envelope ensures that each power slice can accurately reflect the normal parameter boundaries under the corresponding load. Without step S14, the anomaly detection in the subsequent step S20 would not be able to distinguish the normal parameter ranges under different power levels. For example, when the equipment is under light load, the normal fluctuation range of the Y-axis is small. If the full-load threshold is used for judgment, normal light-load parameters may be misjudged as abnormal, or under full load, the threshold may be too strict, leading to missed detection and compromising the accuracy of anomaly detection. This step S14, through a "divide and conquer" load stratification strategy and a "tailor-made" health loop benchmark design, completely breaks through the technical limitations of the traditional fixed threshold "one-size-fits-all" approach. It achieves precise adaptation between the anomaly judgment benchmark and the differences in electrical characteristics of the equipment under different loads, effectively improving the accuracy and robustness of the judgment of abnormal equipment operating status, and providing reliable load adaptive benchmark support for the anomaly detection in the subsequent step S20.

[0048] Step S10 constructs a personalized health benchmark system for equipment through a four-order logic of "spatial binding - parameter visualization - dynamic boundary - load adaptation." This not only overcomes the diagnostic dilemma of the traditional fixed threshold "one-size-fits-all" approach but also achieves accurate capture of the "personalized" electrical characteristics of different equipment under different operating conditions. Furthermore, each sub-step is interconnected to form a complete logical chain: First, S11, based on the power system wiring, constructs an electrical surface topology through graph theory modeling (setting different equipment as independent nodes, connecting components as propagation paths, and setting monitoring points at key equipment locations). This not only achieves precise spatial binding between the benchmark and the equipment to distinguish different devices but also provides a spatial framework for S12-S14, ensuring that all parameter acquisition and health benchmarks are associated with specific equipment nodes. Second, S12 constructs a system with the equipment's rated power point as the origin. A three-dimensional electrical coordinate system is established, and the acquisition interval is dynamically adjusted according to the equipment type. Periodic electrical parameters are acquired and converted into coordinate points to generate an electrical breathing trajectory. This not only concretizes the abstract parameters but also provides a data foundation for the generation of the S13 envelope. Next, S13 uses a three-dimensional convex hull algorithm to generate an electrical breathing envelope that can dynamically adapt to the operating conditions based on the normal trajectory points of the equipment. This forms a dynamic reference boundary and quantifies the health status through the volume of the envelope, providing a boundary basis for the S14 power-layered health ring. Finally, S14 slices the envelope according to the percentage of rated power and customizes a health ring that matches the load on each slice. This not only achieves reference adaptation under load subdivision but also provides a layered judgment standard for subsequent S20 anomaly detection. The whole constitutes a closed-loop logic of "spatial correlation - parameter concretization - health quantification - layered reference". Without step S10, the entire power system integrated monitoring and diagnosis method will lose its core health benchmark support, and subsequent anomaly detection and fault location in S20 and fault diagnosis in S30 will be impossible, causing the entire technical solution to fail. Step S10 provides accurate, dynamic, and hierarchical health judgment basis for subsequent links, realizing a breakthrough from traditional "post-event maintenance" to "predictive maintenance".

[0049] Step S20: Perform anomaly detection on nodes in the electrical surface topology based on the power-layered health loop; trigger the electrical ripple detection mechanism based on the anomaly detection results; construct the ripple time difference matrix; and determine the location of the fault source based on the ripple time difference matrix.

[0050] Further, step S20 includes:

[0051] Step S21: Obtain the instantaneous electrical parameters collected at the monitoring point. Based on the instantaneous electrical parameters and the health loop of power layering, determine whether there are any abnormalities in the nodes of the electrical water surface topology and define the nodes with abnormalities as abnormal nodes.

[0052] The core of step S21 is to determine the anomalies of nodes in the electrical surface topology based on the instantaneous electrical parameters collected from the monitoring points and the power hierarchical health loop generated in step S14. There is a fundamental difference between instantaneous electrical parameters and periodic electrical parameters: periodic electrical parameters are periodic routine monitoring data, such as data collected every 5 minutes for transformers, used to generate the electrical breathing trajectory of equipment in long-term operation, with a low collection frequency and focusing on the steady-state characteristics of the equipment; instantaneous electrical parameters are high-frequency instantaneous monitoring data, with a collection frequency usually in the range of microseconds to milliseconds, for example, set to 1 millisecond / time, focusing on capturing the instantaneous changes in electrical parameters when a fault occurs, such as the sudden increase in current and the sudden increase in harmonic distortion rate caused by a short circuit fault, with a high collection frequency and focusing on the transient abnormal characteristics of the equipment. The difference between the two in terms of collection frequency and monitoring purpose ensures comprehensive coverage of the steady-state operation characteristics and transient fault characteristics of the equipment.

[0053] Please see Figure 2 As shown, the method for determining whether there are anomalies in nodes of the electrical surface topology is as follows: First, determine the power level corresponding to the instantaneous electrical parameter point of the node to be judged, that is, calculate the rated power percentage range based on the instantaneously collected active power value. The instantaneous electrical parameter point is the spatial coordinate point corresponding to the instantaneous electrical parameter in the three-dimensional electrical coordinate system, and the node to be judged is the node in the electrical surface topology to be judged for anomalies. For example, following the example in step S14, slice the power at 10% intervals of rated power. When the instantaneous active power is 0.35×Prated, it corresponds to the 30%-40% rated power range. The division of the power range must be consistent with the percentage interval of the power slice in step S14 to ensure accurate matching between the power level and the health loop parameter; secondly, find the health loop parameter corresponding to the power level, that is, the inner diameter r. in With outer diameter r out Then, calculate the distance d from the instantaneous electrical parameter point to the center of the corresponding power slice surface. The method for determining the coordinates of the power slice surface center is consistent with step S14. If r in ≤d≤r out If d <r in or d>r out If so, the node to be judged is determined to be an abnormal node.

[0054] Step S21 achieves node anomaly determination by collecting high-frequency instantaneous electrical parameters and combining them with a power-layered health loop, solving the core problems of "difficulty in capturing transient faults" and "poor load adaptability" in traditional anomaly detection. First, instantaneous electrical parameters and periodic electrical parameters complement each other in terms of acquisition frequency and monitoring purpose: high-frequency acquisition characteristics can accurately capture millisecond-level parameter mutations during fault occurrence, such as the sudden current surge caused by short-circuit faults or the sudden increase in harmonic distortion rate caused by circuit breaker contact erosion. This avoids the shortcomings of traditional low-frequency acquisition, which misses transient anomaly characteristics due to excessively long time intervals, reducing the fault anomaly perception delay to the millisecond level and gaining critical response time for subsequent fault location. Second, the rated power percentage range of the parameter point is calculated using instantaneous active power, and the corresponding health loop parameters are retrieved. Then, the distance d from the instantaneous parameter point to the slice center is calculated using a two-dimensional distance formula. Finally, based on d and r... in r out The relationship between the node status and the fault location is determined. This "load-adaptive benchmark" design completely solves the drawbacks of the traditional fixed threshold "one-size-fits-all" approach, avoiding the problems of misjudging normal parameters due to excessively wide thresholds under light loads and missing abnormal parameters due to excessively strict thresholds under full loads. This improves the accuracy of fault detection to a level that adapts to different load characteristics. At the same time, this step relies on the electrical surface topology structure constructed in step S11 to ensure that the fault detection result can be accurately associated with specific device nodes, providing clear trigger objects and initial node information for the electrical ripple detection mechanism in step S22. If this step is missing, step S22 will lose the trigger signal of the abnormal node and will not be able to start the construction of the ripple time difference matrix; step S24, fault source location, will also lose the basis for starting the entire fault location process due to the lack of initial abnormal nodes and time sequence starting points, thus breaking the technical chain of "fault identification-fault location".

[0055] Step S22: When an anomaly is detected in any node of the electrical water surface topology, the electrical ripple detection mechanism is triggered to construct the ripple time difference matrix.

[0056] Step S22, based on the abnormal node identified in step S21, triggers the electrical ripple detection mechanism and constructs a ripple time difference matrix. The electrical ripple detection mechanism refers to simulating the propagation characteristics of fault signals in the power system. When a fault occurs, the abnormal electrical signal will spread to surrounding nodes along the electrical connection path like ripples, centered on the fault source. The abnormal electrical signal is such as harmonics and transient currents. By monitoring the arrival time difference of abnormal signals at different nodes, the technical mechanism of tracing the location of the fault source in reverse is achieved.

[0057] T window The value needs to be determined based on the scale of the power system and the signal propagation speed. For example, in a medium- and low-voltage distribution network with a coverage area of ​​less than 10 kilometers, the propagation speed of a fault signal is approximately 3 × 10⁻⁶. 8Meters per second, a 100-millisecond time window can cover a propagation distance of 30 kilometers, which is sufficient to capture the arrival time of abnormal signals from all nodes. If T window If the time interval is too short, such as 10 milliseconds, abnormal signals from distant nodes may be missed. If the time interval is too long, such as 1 second, other irrelevant interference signals may be mixed in, such as transient signals generated by load switching.

[0058] The specific implementation process of step S22 includes: marking ripple candidate sources, that is, recording the first node that detects an anomaly and its anomaly start time t0, defining the first abnormal node as a ripple candidate source. The reason for selecting the first abnormal node as a ripple candidate source is that it has the highest probability of being closest to the fault source, which can narrow the scope of subsequent propagation time monitoring. The ripple candidate source provides a clear time starting point for subsequent reverse tracing of the fault propagation direction, avoiding the confusion of fault source tracing caused by multiple nodes alarming at the same time; setting a preset time window T. window In T window Inside, all nodes in the electrical surface topology are continuously monitored. When node j detects an anomaly, the start time t of the anomaly at node j is recorded. j If in T window If no anomaly is detected, the time of the anomaly at node j is marked as "not triggered," where j is the index variable of the node in the electric water surface topology. Subsequently, a ripple time difference matrix D is constructed. D is an N×N square matrix, where N is the total number of nodes in the electric water surface topology, and the matrix elements D... ij Represents the abnormal time t of node i i The abnormal time t of node j j The absolute value of the difference, if both node i and node j trigger an anomaly, i.e., there exists a valid t. i t j Then D ij =|t i -t j |; If the abnormal moment of either node i or node j is "not triggered", then D ij Assigned the value T window +1, assigned based on the maximum effective time difference being T. window T window +1 is significantly greater than all valid differences, enabling subsequent positioning algorithms to quickly identify invalid node pairs and avoid invalid data interfering with distance difference calculations. Here, i is the index variable of a node in the electrical water surface topology, 1≤i≤N. When i=j, it means that the ripple has propagated from its own position to its own position, at which point the propagation distance is 0, and the corresponding time difference is 0, i.e., D. ij=0. The ripple time difference matrix D accurately quantifies the temporal correlation of the anomalous signal spreading from the ripple candidate source to surrounding nodes, providing core time-series data for step S24 to infer the fault source location through distance difference. This solves the problem that traditional methods, relying solely on single-node anomalous information, cannot distinguish between the fault source and affected nodes.

[0059] Simultaneously record the anomaly intensity value I of each node. i I i Let I be the anomaly intensity value of node i. i Defined as the degree to which the instantaneous electrical parameter point of node i deviates from the healthy loop, the specific calculation formula is as follows: Where Δd is the distance from the instantaneous electrical parameter point of node i to the boundary of the health loop, and R is the radius of the health loop. Δd i The calculation formula is: Where, d i Let I be the distance from the instantaneous electrical parameter point of node i to the center of the corresponding power slice. i The larger the value, the more serious the deviation of the instantaneous electrical parameter point, and the higher the fault level.

[0060] The purpose of step S22 is to obtain the propagation time difference of the fault signal at different nodes, providing time-dimensional data support for subsequent fault source localization. The abnormal nodes identified in step S21 provide the triggering conditions and initial candidate sources for step S22. Without the abnormality determination in step S21, step S22 cannot determine the start time. The electrical water surface topology clarifies the total number of nodes and the connection relationship between nodes, ensuring that the ripple time difference matrix D can completely cover all nodes, avoiding incomplete localization information caused by missing matrix dimensions. If step S22 is missing, the triangulation algorithm in the subsequent step S24 will lose the crucial time difference data, making it impossible to calculate the distance difference from the fault source to each reference point, resulting in the inability to achieve the localization function; in addition, the abnormal intensity value I i The records not only provide a basis for subsequent fault level assessment, but also help verify the direction of the fault source. When the fault signal propagates outward from the source point, the signal weakens due to line loss, and the abnormal intensity will decrease as the propagation distance increases. By comparing the abnormal intensity values ​​of different nodes, the approximate direction of the fault source can be preliminarily determined, providing auxiliary constraints for subsequent positioning algorithms and improving positioning efficiency.

[0061] Step S23: Calibrate the velocity of each propagation path in the electrical water surface topology and construct a ripple propagation velocity model;

[0062] Step S23 calibrates the velocity of each propagation path in the electrical water surface topology, constructing a ripple propagation velocity model to address the positioning error caused by the assumption of a uniform fault signal propagation velocity in traditional positioning methods. In actual power systems, different propagation paths have different impedances, materials, and cross-sectional areas, resulting in significant differences in the propagation velocity of fault signals. A uniform velocity assumption leads to substantial positioning errors. The specific implementation of step S23 includes: first, selecting a light-load period during normal system operation for calibration. Light-load periods have low load current, stable line impedance, and minimal grid interference, ensuring accurate calibration results. During the light-load period, a standard test pulse signal is injected into a designated node in the electrical water surface topology. The pulse signal parameters must meet the principle of "not affecting normal grid operation and being accurately captured." For example, a pulse width of 1 millisecond avoids overlap with normal transient grid signals while ensuring sufficient time for the monitoring point to capture the signal. The pulse amplitude is set to 1% of the rated voltage to avoid impacting grid equipment while ensuring accurate acquisition by the sensors at the monitoring point. Subsequently, the time delay Δt from the injection node to each adjacent node is recorded. test The time delay is measured using a high-precision timestamp module.

[0063] The physical length l of the direct connection path between nodes i and j in the electro-hydraulic surface topology was measured using a field laser rangefinder. ij According to l ij and Δt test Calculate the ripple propagation speed v for each directly connected path. ij =l ij / Δt test v ij The ripple propagation speed is represented by the direct connection path between nodes i and j. Then, a model is established to show the relationship between the ripple propagation speed and the path parameters. The ripple propagation speed v... ij With line impedance value Z ij They are inversely proportional, meaning the ripple propagation speed model is v ij =k' / Z ij Among them, Z ij Let k' represent the line impedance value of the direct connection path between nodes i and j, and k' be the propagation constant. The determination of k' requires least-squares fitting: collect at least 10 sets of line impedance values ​​and corresponding ripple propagation velocities for different propagation paths, where the line impedance values ​​are obtained through impedance measurement. Based on the collected 10 sets of data, each set including the measured line impedance value and measured ripple propagation velocity of the path, construct an error function. Among them, v i' Z represents the measured ripple propagation speed in the i'th data set. i'Let represent the measured line impedance value in the i'th data set, where 1 ≤ i' ≤ 10. The error function E is used to quantify the deviation between the measured ripple propagation speed and the theoretical speed calculated based on the "propagation speed-impedance inverse relationship". The propagation constant k' is obtained by taking the partial derivative of the error function E with respect to the propagation constant k' and setting it to zero.

[0064] The purpose of step S23 is to construct a ripple propagation velocity model that matches the actual path characteristics, providing accurate velocity parameters for subsequent positioning algorithms. The ripple propagation velocity model constructed in step S23 provides the basis for calculating the path velocity corresponding to the ripple time difference matrix D in step S22, ensuring that D... ij With v ij Matching accuracy is crucial. Without this step, the triangulation algorithm in step S24 will only be able to use a uniform propagation speed, leading to errors in the calculation of distance differences between different paths. Furthermore, the impedance-based ripple propagation speed model allows for dynamic adjustment: when the line impedance changes due to temperature and humidity variations, such as increased cable temperature in summer leading to increased impedance, there is no need to recalibrate the pulse. Simply measuring the real-time line impedance value allows the calculation of the real-time propagation speed using the ripple propagation speed model, reducing maintenance workload.

[0065] Step S24: Determine the location of the fault source based on the ripple time difference matrix and the ripple propagation velocity model.

[0066] Step S24, based on the ripple time difference matrix D and the ripple propagation velocity model, determines the location of the fault source using an improved triangulation algorithm. This algorithm adds an error correction mechanism to the traditional triangulation method, improving the robustness of the location. (See also...) Figure 3 As shown, the specific implementation process needs to be carried out in stages: First, select the positioning reference points. From the ripple time difference matrix D, select the three nodes with the smallest time difference from the ripple candidate source as reference points. The smallest time difference indicates that the propagation path between these nodes and the fault source is the shortest, and they are most likely to be close to the real fault source, which can reduce the impact of propagation time measurement error on the positioning results. Second, calculate the distance difference from the fault source to each reference point. For each pair of reference points i* and j*, based on the abnormal time t of reference point i*... i* Abnormal time t relative to the reference point j* j* The absolute value of the difference D i*j* The ripple propagation speed v of the path between reference points i* and j* i*j* Calculate the distance difference Δd i*j* =D i*j* ×v i*j* , where D i*j* Extract from the ripple time difference matrix D, v i*j* This is obtained through the ripple propagation velocity model. The physical meaning of the distance difference is the distance d from the fault source to the reference point i*. i* Distance d from the reference point j*j* The absolute value of the difference.

[0067] In a two-dimensional plane, if the coordinates of two fixed points are known, and it is determined that the difference in distance from a certain point to these two fixed points is always equal to Δd. i*j* Then, connecting all possible positions of that point will form a curve, which is a hyperbola. Based on the principle of drawing a hyperbola, a hyperbola is drawn on the electrical surface topology. The distance difference from the two reference points i* and j* within the electrical surface topology is Δd. i*j* The trajectory of the points is a hyperbola, therefore the fault source must be located on this hyperbola. Three pairs of reference points (i*,j*), (i*,k*), and (j*,k*) generate three hyperbolas. Theoretically, the intersection of these three hyperbolas is the location of the fault source, where i*, j*, and k* are index variables of the reference points, and i*≠j*≠k*. However, considering the existence of time synchronization errors and speed calibration errors in actual measurements, the three hyperbolas may not intersect precisely at a single point. In this case, it is necessary to calculate the intersection of the three hyperbolas. The three intersection points of the curves are used for each pair of intersections. For example, hyperbola 1 is drawn based on reference points (i*, j*), hyperbola 2 is drawn based on reference points (i*, k*), and hyperbola 3 is drawn based on reference points (j*, k*). Hyperbola 1 and hyperbola 2 intersect at point A, hyperbola 1 and hyperbola 3 intersect at point B, and hyperbola 2 and hyperbola 3 intersect at point C. The centroid of the triangle formed by these three intersection points is used as the location of the fault source. The centroid method can effectively smooth the error of a single intersection point and improve the stability of the positioning results. The purpose of step S24 is to accurately determine the physical location of the fault source and solve the problem that traditional methods cannot trace the source of the fault. The ripple time difference matrix in step S22 provides time difference data, and the ripple propagation velocity model in step S23 provides velocity data. The distance difference can only be calculated by combining the two. Without either data, the hyperbola cannot be generated. The electrical water surface topology provides a spatial reference for the drawing of the hyperbola and the calculation of the intersection points. If step S24 is missing, the fault source location target of step S20 cannot be achieved. The abnormal node information, time difference data, and velocity model collected in the preceding steps S21-S23 cannot be converted into specific fault source locations. As a result, the subsequent step S30 cannot eliminate physically impossible fault types based on the fault source location. For example, if the fault source is located near the transformer, but is diagnosed as a circuit breaker fault, it will lead to diagnostic conflicts.

[0068] Step S20 constructs a complete fault source localization system through a four-order logic of "abnormal node identification - ripple time difference acquisition - propagation speed calibration - triangulation localization," solving the problem of "hidden source and scattered manifestations" of power system faults in traditional monitoring and achieving accurate fault source localization. From an overall collaborative perspective, the power hierarchical health loop of step S21 and step S14 is combined to ensure the load adaptability of anomaly judgment; step S22 uses the abnormal node of step S21 as a trigger and combines it with the electrical surface topology of step S11 to construct a ripple time difference matrix and obtain propagation time dimension data; step S23 performs speed calibration based on the propagation path of step S11 to construct a speed model that conforms to reality; step S24 integrates the time difference of step S22 and the speed of step S23, and combines it with the topological space benchmark of step S11 to achieve localization. Each step is interconnected, forming a closed-loop logic of "abnormal triggering - data acquisition - model construction - localization calculation." This invention achieves full-condition adaptability for fault source location. Regardless of light load, full load, or different equipment types, it maintains high location accuracy through dynamic health loops and differentiated speed calibration, while traditional methods are only effective under specific conditions. Without step S20, the entire power system integrated monitoring and diagnosis method will lose its core "location" link, causing a disconnect between health monitoring in step S10 and fault diagnosis in step S30, making it impossible to determine the fault location and ultimately failing to achieve the transformation from "reactive maintenance" to "predictive maintenance."

[0069] Step S30: Construct a fault symptom puzzle template library and symptom block association rules; obtain the current symptom sequence based on the fault source whose location has been determined; match the current symptom sequence with the fault symptom puzzle template library to calculate the matching degree; perform missing symptom block reasoning and diagnostic conflict resolution based on the matching degree and symptom block association rules; and output the fault diagnosis result.

[0070] The fault symptom template library refers to a structured database that integrates historical fault case data of various equipment in the power system under different operating conditions. It contains a mapping relationship between fault types and corresponding multi-dimensional symptom blocks. The various equipment types cover node devices in the electrical surface topology defined in step S11, such as transformers, circuit breakers, and busbars. Different operating conditions include light load, full load, harmonic interference, and line aging. A symptom block is defined as the minimum set of electrical parameters that can characterize fault features, specifically including the instantaneous electrical parameter anomaly type, anomaly duration τ, the trend of anomaly intensity value, the time difference of the fault signal on the propagation path, and the health loop parameters of the corresponding power slice. Instantaneous electrical parameter anomaly types include active power deviation X exceeding the standard, sudden change in reactive power deviation Y, and sudden increase in harmonic distortion rate Z. The trend of anomaly intensity value changes includes continuous increase, pulse fluctuation, and stable maintenance. The time difference of the fault signal on the propagation path is extracted from the constructed ripple time difference matrix D. The health loop parameters include the inner diameter r. inWith outer diameter r out .

[0071] The construction of the fault symptom puzzle template library requires the collection of valid fault cases covering common operating years of the power system. Case selection must meet the requirements of covering different equipment types, different fault causes, and different environmental conditions to ensure the diversity and representativeness of the cases. Different fault causes include overload, insulation damage, and operational errors; different environmental conditions include normal temperature and humidity ranges. Subsequently, each fault case is decomposed into symptom blocks. The decomposition process requires combining the electrical characteristics of the corresponding equipment and the health loop parameters of the power layer to determine the parameter range of each symptom block. For example, in the case of a transformer inter-turn short circuit fault, the parameter range of "harmonic distortion rate Z exceeding the standard" in the decomposed symptom block needs to be determined by statistical analysis of the harmonic distortion rate Z value of the transformer model during normal operation. Specifically, the preset percentile of the preset number of normal trajectory points of the equipment in a recent period is taken. The selection of this percentile must ensure that it can cover more than 90% of the normal operating conditions of the equipment to avoid misjudging normal fluctuations as abnormalities. After the symptom block is decomposed, a set of symptom blocks is established for each type of fault. The mapping relationship is clearly defined in a table. Only symptom blocks that appear in all cases of the same type of fault above a preset proportion are retained as core symptom blocks, and those with a probability of occurrence below another preset proportion are retained as secondary symptom blocks, so as to avoid random symptoms from interfering with the diagnostic results.

[0072] Symptom block association rules refer to the statistical correlation rules between symptom blocks constructed using the Apriori association rule mining algorithm based on symptom block combination data from historical failure cases. These rules describe the logical relationship that "if symptom block A and symptom block B exist, then symptom block C is likely to exist." The construction of this rule requires first encoding the symptom blocks of all failure cases in the template library, converting each symptom block into a binary variable, with 1 indicating its existence and 0 indicating its absence. Then, support and confidence thresholds for the association rule are set. Support is defined as the proportion of cases containing both the antecedent (e.g., symptom blocks A and B) and the consequent (e.g., symptom block C) to the total number of cases. Confidence is defined as the proportion of cases containing both the antecedent and consequent to the proportion of cases containing only the antecedent. The thresholds are determined through statistical analysis of 1000 valid failure cases: the support and confidence distributions for different symptom block combinations are calculated. When the support is ≥5%, it indicates that the symptom block combination occurs frequently enough and has statistical significance, avoiding rule failure caused by rare combinations. When the confidence is ≥85%, it indicates that the antecedent has a strong predictive ability for the consequent and can be used as a valid association basis. Rules below this threshold are removed due to low predictive reliability.

[0073] The acquisition of the current symptom sequence requires using the fault source location determined in step S24 as the spatial reference. Only the electrical parameters and abnormal information of the nodes corresponding to the fault source and the nodes on the propagation path directly connected to it are collected to avoid interference from irrelevant node data in the diagnostic process. The specific process is as follows: First, the instantaneous electrical parameters of the fault source node and its adjacent nodes are extracted from the instantaneous electrical parameters collected from the monitoring points, and then combined with the corresponding power slice health loop parameter r. in and r out First, determine whether each parameter exceeds the health loop range, and record the anomaly start time and anomaly duration τ, where τ is the time difference from the anomaly start time to the current data acquisition time. Second, retrieve the anomaly intensity value and its change curve of the above nodes, calculate the rate of change of the anomaly intensity value, and determine the trend of the anomaly intensity value based on the rate of change of the anomaly intensity value. If the rate of change is stable and not zero, it is determined to be a linear increase or decrease; if the rate of change fluctuates periodically, it is determined to be a pulse fluctuation. Then, extract the anomaly triggering time difference between the fault source node and adjacent nodes from the ripple time difference matrix D, and clarify the specific value of the time difference. Finally, arrange the above-determined symptom blocks in the order of the anomaly occurrence time to form the current symptom sequence. Each symptom block in the current symptom sequence is closely related to the spatial location, propagation characteristics, and health benchmark of the fault source, avoiding the problems of lack of spatial specificity in symptom acquisition and the disconnect between parameters and equipment characteristics in the existing technology.

[0074] The multidimensional matching calculation of the overall matching degree needs to be carried out from three dimensions: symptom block existence, parameter value deviation, and temporal consistency. The weight of each dimension is determined by the Analytic Hierarchy Process (AHP)—a predetermined number of experts with rich experience in power system fault diagnosis are invited to compare the importance of each dimension in the diagnostic process pairwise, construct a judgment matrix, and calculate the weight of each dimension after passing the consistency test. Among them, the symptom block existence dimension has the highest weight, because the existence or absence of the core symptom block directly determines the correlation between the fault type and the symptom combination; the parameter value deviation dimension has the second highest weight, which reflects the degree of deviation between the parameter value of the current fault symptom block and the mean parameter value of the corresponding fault type symptom block in the template library. The smaller the deviation, the higher the similarity between the current fault and the template fault; the temporal consistency dimension has the third highest weight. The temporal consistency dimension reflects the degree of sequential matching between the current symptom sequence and the template sequence by calculating the edit distance between the two. The smaller the edit distance, the higher the temporal consistency. The edit distance is the minimum number of insertion, deletion, and replacement operations required to convert the current sequence into the template sequence. For example, the weight for existence matching is 0.4, the weight for numerical deviation matching is 0.3, and the weight for temporal consistency matching is 0.3. The scores for each dimension are calculated as follows: For the symptom block existence dimension, if the current symptom sequence contains all the core symptom blocks of the corresponding fault type, the score is full. If core symptom blocks are missing, points are deducted proportionally based on the number of missing blocks. The presence or absence of secondary symptom blocks does not affect the score for this dimension. For the parameter numerical deviation dimension, the relative deviation between the parameter value of the current symptom block and the mean parameter value of the corresponding symptom block in the template is calculated. The relative deviation is the ratio of the absolute value of the difference between the current parameter value and the mean to the standard deviation of the template parameters. The standard deviation is an indicator of the dispersion of the parameters of the corresponding symptom block in the template. If the relative deviation is less than or equal to 1, the score is negatively correlated with the relative deviation, that is, the smaller the relative deviation, the higher the score. The calculation formula can be 1 - relative deviation. If the relative deviation is greater than 1, the score is zero. For the temporal consistency dimension, the score is negatively correlated with the edit distance, that is, the smaller the edit distance, the higher the score. Specifically, the temporal consistency score is equal to 1 minus the ratio of the edit distance to the total number of symptom blocks in the template. The overall matching degree is the sum of the products of the scores of each dimension and their corresponding weights. This value is used to quantify the degree of matching between the current symptom sequence and each fault type in the template library.

[0075] The missing symptom block inference is based on the association rules of symptom blocks and the characteristics of the equipment type corresponding to the location of the fault source. When a symptom block is missing in the current symptom sequence due to temporary sensor failure or signal interference at the monitoring point, the system first searches the association rule base for rules that contain the existing symptom block as antecedents. The rule with the highest confidence is selected as the initial inference basis to obtain candidate results for the missing symptom block. Subsequently, the system is verified by combining the electrical characteristics of the fault source equipment type and the ripple propagation speed model. Based on the propagation path length and ripple propagation speed between the fault source node and its adjacent nodes, the theoretical timing difference of the fault signal on the propagation path is calculated. If the parameter range of the candidate missing symptom block contains the theoretical timing difference, the validity of the candidate result is confirmed. If not, the association rule with the second highest confidence is re-selected, and the verification process is repeated until a missing symptom block that matches the electrical characteristics of the equipment is obtained, thus avoiding inference errors caused by not distinguishing equipment characteristics in the prior art.

[0076] For example, if the current symptom sequence contains the symptom blocks "harmonic distortion rate Z value > 5%" and "abnormal intensity value increases linearly", but lacks the symptom block "adjacent node timing difference", the rule in the association rule base with the highest confidence is "if there is 'harmonic distortion rate Z value > 5%' and 'abnormal intensity value increases linearly', then there is 'adjacent node timing difference ≤ 15μs'". Therefore, the initial inference is that the missing symptom block is "adjacent node timing difference ≤ 15μs". Subsequently, verification is performed based on the characteristics of the fault source equipment type. If the fault source is a transformer node, its propagation path length with the adjacent bus node is calibrated to 400 meters in step S23, and the ripple propagation speed v = 3 × 10⁻⁶. 8 m / s, then the theoretical timing difference = path length / v = 400 / (3 × 10) 8 The time difference is 1.33 μs, which matches the inference result of "≤15 μs", further confirming the validity of the missing symptom block; if the fault source is a bus node, its path length to the adjacent circuit breaker node is 800 meters, and the theoretical timing difference is 800 / (3×10). 8 =2.67μs, which also matches the inference result, ensuring that the inference is not affected by the equipment type and is verified only based on the actual electrical characteristics, thus solving the problem of inference errors caused by not distinguishing equipment characteristics in the prior art.

[0077] Diagnostic conflict resolution is used to solve problems where multiple fault templates have a high degree of overall matching. Specifically, it is carried out through three levels: the correlation strength of symptom blocks, the physical characteristics of the fault source location, and the fit of healthy loop parameters. First, the confidence levels of the association rules for the core symptom blocks corresponding to each high-matching template are compared. Templates with higher confidence levels have a stronger correlation between their symptom block combinations and fault types, and thus higher diagnostic priority. Second, considering the physical characteristics of the fault source location, the equipment type corresponding to the fault source location must be consistent with the typical equipment of the template fault type. Simultaneously, considering the characteristic that "the abnormal intensity value decreases with distance when the fault signal propagates outward from the fault source," it is verified whether the distribution of the abnormal intensity values ​​of each node is consistent with the distribution pattern of the abnormal intensity values ​​of the template fault type. For example, if the fault source of the template fault type is a transformer, then the abnormal intensity value of the transformer node should be greater than the abnormal intensity value of the adjacent bus node. If they are inconsistent, the priority of the template is reduced. Finally, the fit between each template and the health ring parameters is verified. The similarity between the deviation of the current symptom block parameters from the health ring and the corresponding deviation in the template is calculated. The higher the similarity, the higher the fit and the higher the priority. Through these three layers of screening, a unique matching fault type is finally determined, achieving diagnostic conflict resolution.

[0078] The output of the fault diagnosis results should include four parts: fault type, fault severity, fault impact range, and recommended handling measures. The fault severity is determined based on the combination of the maximum value of the abnormal intensity value and the abnormal duration τ. Through statistical analysis of historical fault consequences, multiple threshold combinations of the maximum value of the abnormal intensity value and τ are set, with different combinations corresponding to different fault severity levels. The fault impact range is determined based on the distribution of abnormal nodes in the electrical surface topology. Abnormal nodes are defined as nodes where symptom blocks exist in the comprehensive matching degree calculation and the abnormal intensity value reaches the preset level. Recommended handling measures should be formulated in combination with the fault type and severity. For severe faults, it is recommended to immediately shut down for maintenance and identify the core components to be inspected. For general faults, it is recommended to shut down for maintenance within a preset time. For minor faults, it is recommended to increase the monitoring frequency.

[0079] Step S30 enables multi-symptom collaborative and accurate diagnosis to avoid the ambiguity of traditional single-parameter diagnosis. Existing technologies often rely solely on single electrical parameters such as excessive harmonic distortion rate Z to determine faults, ignoring key symptoms such as abnormal intensity values ​​and timing differences, leading to misdiagnosis. This step, however, constructs a template library containing multi-dimensional symptom blocks, ensuring that each fault type corresponds to a unique combination of core symptom blocks. Secondary symptom blocks are used to narrow the diagnostic scope. Furthermore, the similarity between the current fault and the template fault is quantified through three dimensions: symptom block existence, parameter value deviation, and timing consistency. By leveraging the fault "feature profile" constructed from multiple symptom blocks, the fault type is uniquely determined, significantly improving diagnostic accuracy. The parameter range of the symptom blocks in the fault symptom puzzle template library is based on the health loop parameter r generated in step S14. in r outTo ensure accurate diagnosis, the system matches the rated power and load conditions of the equipment, avoiding diagnostic biases from generic templates. The current symptom sequence only collects data from the fault source and adjacent nodes identified in step S24, reducing data redundancy from irrelevant nodes and ensuring a direct correlation between symptoms and the physical location of the fault. Step S10's power-layered health loop provides personalized anomaly judgment benchmarks for different loads, preventing missed diagnoses due to overly broad generic thresholds under light loads and misdiagnoses due to overly strict thresholds under full loads. Step S24's fault source localization focuses symptom collection on the core fault area, reducing data processing volume and improving diagnostic efficiency. Step S30 possesses the ability to supplement missing symptoms and resolve conflicts to enhance diagnostic robustness. To address the issues of diagnostic interruption or uncertain results in existing technologies when symptoms are missing or multiple templates match well, this step infers missing symptoms by using symptom block association rules built from a large number of historical cases that meet confidence thresholds. This is combined with verification based on the electrical characteristics of the equipment to ensure accurate supplementation. Furthermore, a multi-layered logical process, considering symptom block association strength, physical characteristics of the fault source location, and the fit of health loop parameters, is used to filter high-matching templates, ensuring unique diagnostic results and avoiding misjudgments due to similar symptoms. Given the significant differences in diagnostic accuracy under different load conditions in existing technologies, this step, in conjunction with the power-layered health loop in step S14, addresses this by providing different power... The slicing plane sets an appropriate range of symptom block parameters, ensuring that the anomaly judgment benchmark is consistent with the different load characteristics of the equipment, eliminating the influence of operating conditions on diagnostic accuracy. Furthermore, the diagnostic results include the fault impact range based on the electrical surface topology, allowing maintenance personnel to directly locate the nodes requiring repair and the propagation path based on the topology diagram. This avoids the problem of traditional diagnostics that only know the fault type but blindly troubleshoot, significantly shortening maintenance time. Step S30 ensures a closed loop for the entire power system's comprehensive monitoring and diagnostic system, eliminating obstacles to the integration of technical features. Without step S30, the health monitoring in step S10 can only determine node anomalies, and the fault source location in step S20 can only determine the fault. The inability to translate "anomaly" and "location" into specific fault types prevents the technical solution from reaching the "problem detection - location determination" stage and achieve the core goal of "predictive maintenance." Furthermore, the anomaly intensity value, ripple time difference matrix D, and health loop cannot be correlated with the fault type. This step, however, uses a template library and association rules to connect health baseline parameters, anomaly data, fault location, and fault type, forming a complete closed loop of "anomaly detection - fault location - fault diagnosis." This enables a shift from traditional "reactive maintenance" to "predictive maintenance," while eliminating barriers to the integration of various technical features and improving the practicality and operability of the entire solution.

[0080] Example 2:

[0081] This embodiment, based on Embodiment 1, provides a comprehensive power system monitoring and diagnosis system based on electrical signal processing algorithms, such as... Figure 4 As shown, it includes:

[0082] The health baseline construction module is used to abstract the physical connection relationship of the power system into an electrical surface topology, establish a three-dimensional electrical coordinate system, generate the electrical breathing trajectory of the equipment in the three-dimensional electrical coordinate system, generate the electrical breathing envelope based on the electrical breathing trajectory, and generate a power-layered health ring based on the electrical breathing envelope.

[0083] Fault source location module: Used to perform anomaly detection on nodes in the electrical surface topology based on the power-layered health loop, trigger the electrical ripple detection mechanism based on the anomaly detection results, construct the ripple time difference matrix, and determine the location of the fault source based on the ripple time difference matrix;

[0084] Fault diagnosis module: It is used to build a fault symptom puzzle template library and symptom block association rules. It obtains the current symptom sequence based on the fault source with a determined location, matches the current symptom sequence with the fault symptom puzzle template library to calculate the matching degree, and performs missing symptom block reasoning and diagnosis conflict resolution based on the matching degree and symptom block association rules, and outputs the fault diagnosis results.

[0085] Furthermore, in the health benchmark construction module, the method for generating the electrical breathing envelope surface based on the electrical breathing trajectory includes: extracting normal operating trajectory points from the electrical breathing trajectory to form a normal point set; inputting the three-dimensional coordinates of the normal point set into a three-dimensional convex hull algorithm; the three-dimensional convex hull algorithm calculates the convex hull of the normal point set and outputs the vertex coordinates constituting the convex hull; and then constructing a closed convex polyhedron surface based on the vertex coordinates, wherein the convex polyhedron surface is the electrical breathing envelope surface.

[0086] Furthermore, in the fault source location module, the method for anomaly detection of nodes in the electrical surface topology based on the power-layered health loop includes:

[0087] Determine the power level corresponding to the instantaneous electrical parameter point of the node to be judged, find the health loop parameter corresponding to the power level, and calculate the distance d from the instantaneous electrical parameter point to the center of the corresponding power slice. If r in ≤d≤r out If d <r in or d>r out If the node to be judged is an abnormal node, then the node to be judged is determined to be an abnormal node; where, the instantaneous electrical parameter point is the spatial coordinate point corresponding to the instantaneous electrical parameter in the three-dimensional electrical coordinate system, and the health ring parameter includes the inner diameter r of the health ring. in With outer diameter r out .

[0088] The methods and systems of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated.

[0089] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0090] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for comprehensive monitoring and diagnosis of power systems based on electrical signal processing algorithms, characterized in that, The method comprises: The method comprises: According to the power layered health ring, the node in the electrical water surface topology is detected for anomaly, including: acquiring the instantaneous electrical parameter collected at the monitoring point, determining whether the node in the electrical water surface topology exists anomaly according to the instantaneous electrical parameter and the power layered health ring, and defining the node existing anomaly as an abnormal node; the method for determining whether the node in the electrical water surface topology exists anomaly is: determining the power level corresponding to the instantaneous electrical parameter point of the node to be determined, searching for the health ring parameter corresponding to the power level, calculating the distance d from the instantaneous electrical parameter point to the center of the corresponding power slice surface, if r in ≤d≤r out , it is determined that the node to be determined is in the normal operation area, if d in or d>r out , it is determined that the node to be determined is an abnormal node; wherein the instantaneous electrical parameter point is the space coordinate point corresponding to the instantaneous electrical parameter in the three-dimensional electrical coordinate system, the health ring parameter includes the inner diameter r in and the outer diameter r out of the health ring. The method comprises: The method comprises:

2. The method for comprehensive monitoring and diagnosis of power systems based on electrical signal processing algorithms according to claim 1, characterized in that, The method comprises: The method comprises:

3. The method for comprehensive monitoring and diagnosis of power systems based on electrical signal processing algorithms according to claim 2, characterized in that, The method comprises:

4. The power system integrated monitoring and diagnosis method based on electrical signal processing algorithm according to claim 3, characterized in that, The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises: The method comprises:

Images