Intelligent identification method for hidden danger of secondary equipment based on wide-area wave recording file

Abstract

A secondary equipment hidden danger intelligent identification method based on a wide area wave recording file comprises the following steps: in response to a power grid primary equipment fault event, carrying out multi-stage plant station topology analysis based on a power grid topology model, and determining peripheral plant stations associated with the fault event; automatically calling wave recording files of a fault plant station and peripheral plant stations at the fault moment; analyzing the recording file and extracting channel information and waveform image data for hidden danger identification; and based on the extracted channel information and waveform image data, carrying out parallel multi-dimensional hidden danger intelligent identification on the secondary equipment of the plant station through a rule engine and a deep learning model. According to the method, single-point post analysis is improved into wide-area linkage and multi-dimensional parallel intelligent screening, and particularly, AI identification based on waveform images is introduced, so that complex hidden dangers which are difficult to capture by a traditional method can be found, the transformation from post analysis to pre-warning is realized, and the efficiency and depth of secondary equipment hidden danger identification are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of power system relay protection and automation technology, and specifically relates to a method for intelligent identification of potential hazards in secondary equipment by using fault event triggering to link multiple substations for waveform data analysis and combining waveform image artificial intelligence recognition technology. Background Technology

[0002] In power systems, secondary equipment such as relay protection and measurement and control systems are crucial for ensuring the safe and stable operation of the power grid. When primary equipment fails, the relevant secondary equipment should operate correctly to isolate the fault. Fault recorders record valuable data such as electrical quantities and switching quantities of the power grid before and after the fault, serving as an important basis for accident analysis and equipment condition assessment.

[0003] Currently, the main problems in investigating potential hazards in secondary equipment are as follows: 1. Post-fault and isolated nature: The analysis is usually carried out after the fault occurs, and is mostly limited to the waveform data of the fault station, lacking synchronous analysis of the status of secondary equipment in related surrounding plants and stations.

[0004] 2. Reliance on human experience: The analysis work heavily depends on the experience of professionals, resulting in low efficiency and a high risk of overlooking deeper, interconnected potential problems due to fatigue or negligence. In particular, the analysis of complex waveform anomalies requires experts to spend a significant amount of time on manual interpretation.

[0005] 3. Limited identification dimensions: Traditional analysis often focuses on the direct cause of the fault, lacking systematic and multi-dimensional (such as sampling circuits, action logic, setpoint coordination, input circuits, etc.) automated hazard screening methods.

[0006] 4. Inability to identify complex waveform patterns: Existing automated methods are mostly based on rules and threshold judgments, and lack effective automatic identification means for complex waveform distortions and transient process anomalies caused by CT (Current Transformer) saturation, ferroresonance, changes in the mechanical characteristics of circuit breakers, etc.

[0007] 5. Inability to form preventive early warning: Due to the lack of systematic automated analysis tools, it is difficult to detect equipment defects that have not yet caused accidents (such as slight abnormalities in sampling circuits, slight deviations in protection action time, early mechanical fault characteristics, etc.), and it is impossible to realize the transformation from "post-event analysis" to "pre-event early warning".

[0008] Therefore, there is an urgent need for a technical solution that can automatically, quickly, multi-dimensionally, and across plants and stations systematically identify potential hazards in secondary equipment. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides an intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files, thereby solving the problems of low analysis efficiency, single dimension, lack of linkage and prevention in existing technologies.

[0010] The present invention adopts the following technical solution. The present invention provides a method for identifying potential hazards in secondary equipment based on waveform recording files, comprising the following steps: Fault event triggering and topology expansion include: responding to a fault event of primary equipment in the power grid, performing multi-level substation topology analysis based on the power grid topology model, identifying surrounding substations associated with the fault event, and forming a list of associated substation groups centered on the fault point; Multi-station waveform data retrieval includes: automatically retrieving waveform files from the faulty plant and surrounding plants at the time of the fault based on the list of associated plant groups; Waveform data analysis and feature extraction, including: analyzing waveform files and extracting channel information and waveform image data for hazard identification; Multi-dimensional hazard intelligent identification includes: based on the extracted channel information and waveform image data, multi-dimensional hazard intelligent identification is carried out in parallel on the secondary equipment of the plant through a rule engine and a deep learning model, respectively.

[0011] Preferably, the rule engine-based hazard identification includes: sampling circuit anomaly identification, action behavior anomaly identification, protection setting anomaly identification, and input / output anomaly identification.

[0012] Preferably, the sampling circuit anomaly identification includes at least one of the following: dual-sampling inconsistency identification, CT saturation identification, and harmonic anomaly identification.

[0013] Preferably, the abnormal action behavior identification includes at least one of the following: abnormal tripping time identification, and inconsistent identification of two sets of actions.

[0014] Preferably, the protection setting anomaly identification includes at least one of the following: protection setting sensitivity coefficient anomaly identification, minimum operating current verification anomaly identification.

[0015] Preferably, the protection setting sensitivity analysis includes the following steps: The fault time is calculated by measuring voltage and current abrupt changes in the recorded waveform data, and the electrical characteristic values ​​at the fault time are also calculated. Extract the setpoints from the waveform HDR file, or retrieve the setpoints of the relay protection device to obtain the impedance setpoints; Calculate the protection sensitivity coefficient: Based on the fault type and protection principle, calculate the ratio of the actual fault quantity to the protection setting value to obtain the sensitivity coefficient K_sen; The calculated sensitivity coefficient K_sen is compared with the minimum sensitivity coefficient K_min required by the power system regulations to verify compliance with the regulations. Analyze whether the timing and sensitivity coordination between the main protection and backup protection are reasonable under the same fault. Output the identification conclusion.

[0016] Preferably, the input / output anomaly identification includes at least one of the following: switch quantity jitter identification and malfunction circuit anomaly identification.

[0017] Preferably, the waveform image intelligent identification based on the deep learning model includes: taking current and voltage waveforms as input, and using a convolutional neural network model to automatically learn and identify potential hazards, including: CT deep saturation characteristic waveform, PT ferroresonant waveform, and current waveform distortion caused by circuit breaker operating mechanism failure.

[0018] Preferably, the waveform image intelligent recognition based on the deep learning model includes the following steps: The waveform image is input into a pre-trained deep learning model to perform feature extraction and pattern recognition. Output waveform anomaly type and confidence level; The identification results are integrated with the hazard identification results based on the rule engine to generate a comprehensive identification report.

[0019] Compared with the prior art, the beneficial effects of the present invention include at least the following: 1. Linked analysis, global perspective: Through fault triggering and topology expansion, it automatically summons waveform data from multiple stations, breaking the limitations of single-station analysis and enabling the discovery of related hidden dangers at the system level.

[0020] 2. Multi-dimensional parallel and in-depth screening: It integrates identification modules of multiple dimensions such as sampling, action, set value, input and output, and performs an in-depth inspection of the secondary system, covering most common types of potential problems in secondary equipment.

[0021] 3. Waveform AI Recognition, Breaking Limits: The introduction of deep learning recognition technology based on waveform images enables the system to automatically learn and discover new patterns from complex, high-dimensional waveform data. It can identify complex hidden dangers that are difficult to describe by traditional methods, such as specific types of CT saturation, ferromagnetic resonance, and early mechanical fault characteristics.

[0022] 4. Automation and Intelligence: The entire process from data retrieval to hazard identification has been automated, which greatly improves the efficiency and accuracy of analysis and reduces the reliance on human experience.

[0023] 5. Preventive maintenance: It can promptly detect potential defects that are "operating with problems" (such as minor deviations in action time, abnormal early waveform characteristics, etc.), providing a basis for decision-making for predictive maintenance of equipment and preventing problems before they occur. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for intelligent identification of potential hazards in secondary equipment based on wide-area waveform recording files, provided in accordance with an embodiment of the present invention. Figure 2 This is a flowchart of the protection setting sensitivity analysis submodule provided according to an embodiment of the present invention; Figure 3 This is a flowchart of an AI waveform recognition unit provided according to an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0026] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying potential hazards in secondary equipment based on wide-area waveform recording files, including the following steps: Step 1, Fault Event Triggering and Topology Expansion, includes: In response to a primary equipment fault event in the power grid, performing multi-level substation topology analysis based on the power grid topology model to identify surrounding substations associated with the fault event, forming a "group of associated substations" centered on the fault point. Preferably, but not restrictively, Step 1 specifically includes: Step 1.1: When the dispatch center or substation monitoring system reports a device fault, such as, but not limited to, a line short circuit or bus fault, the system automatically captures the fault event.

[0027] Step 1.2, forming an “associated plant and substation group”, includes: when a fault occurs in a primary device in the power grid and is recorded, the plant or substation where the fault occurred is taken as the root node, a pre-set power grid topology model service is called to perform topology analysis and form an “associated plant and substation group” centered on the fault point.

[0028] Further preferred, but not limited, the topology analysis is not limited to directly electrically connected opposite substations, but also includes adjacent substations that may be affected by power flow transfer and protection coordination, preferably but not limited to the other substation on the double circuit where the faulty line is located, and the superior or same-level substation that provides backup protection for it.

[0029] In other words, the "associated power plant group" is a set of surrounding power plants associated with the current fault, determined through multi-level expansion based on the power grid topology model with the faulted power plant as the root node. This multi-level expansion includes both directly electrically connected power plants, such as the peer power plant, and adjacent power plants that may be affected by the fault or require protection coordination.

[0030] For example, but not in a limiting sense, the topology analysis automatically generates a three-level list of associated power plants by parsing the CIME model of the power grid and performing a breadth-first search based on graph theory algorithms. The "associated power plant group" contains a total of 23 power plants, providing targets for subsequent waveform data retrieval.

[0031] Step 2, Multi-station waveform data retrieval, including: based on the "Associated Plant Group" list, automatically retrieve waveform files of the faulty plant and surrounding plants at the time of the fault.

[0032] Preferably, but not limitingly, the system automatically and in parallel sends recording file recall commands to the protection devices or fault recorders of all plants in the "Associated Plant Group" list, based on the associated plant list generated in step 1, through standard communication protocols, such as, but not limited to, IEC 61850, 103 protocols.

[0033] Further, but not restrictively, the instruction precisely specifies the fault timestamp and sets a reasonable time window to obtain data for the entire process before, during, and after the fault, including protection startup waveform recordings and fault waveform recordings. Specifically, the instruction automatically retrieves all startup waveform recordings and fault waveform recordings within a certain period before and after the fault time from the faulted plant and all surrounding plants in the "Associated Plant Group" list determined in step 1.

[0034] For example, but not limited to, the data acquisition module uses multi-threading technology to simultaneously send data request requests to hundreds of protection devices and fault recorders in 23 plants and stations, and performs integrity verification upon receiving the files to ensure the comprehensiveness and reliability of the data.

[0035] Step 3, waveform data analysis and feature extraction, includes: analyzing the waveform file and extracting channel information and waveform image data for hazard identification.

[0036] Preferably, but not limitingly, after receiving the waveform recording files from each plant, the system invokes a unified analysis engine to decode all the retrieved waveform recording files. After analysis, key channel information is extracted as input features for subsequent intelligent identification. More preferably, but not limitingly, these features include: Analog signal characteristics: instantaneous waveforms, power frequency RMS values, phase angles, and harmonic content of each analog signal channel, i.e., each phase current and voltage channel.

[0037] Switching characteristics: The precise action sequence, change time and duration of each switching channel, namely various protection trip outputs, circuit breaker positions, channel alarms and other switching quantities.

[0038] Meanwhile, to support waveform recognition based on artificial intelligence, as one of the outstanding substantive features of this invention, in step 3, the specified key analog quantity channels, such as but not limited to, the waveform data of line current and bus voltage, will be rendered into standardized waveform images, such as but not limited to, PNG format, according to a unified time scale and amplitude scale, as input data for the deep learning model.

[0039] Step 4, multi-dimensional intelligent identification of potential hazards, includes: based on the extracted channel information and waveform image data, multi-dimensional intelligent identification of potential hazards is carried out in parallel on the secondary equipment of the plant through a rule engine and a deep learning model, respectively.

[0040] As one of the prominent substantive features of this invention, the multi-dimensional hazard identification carried out in parallel includes at least: anomaly identification based on a rule engine and intelligent waveform image identification based on a deep learning model.

[0041] It is worth noting that rule-based anomaly identification uses pre-defined logical rules and thresholds for rapid judgment, efficiently handling relatively logically clear potential problems such as inconsistent sampling between two sets of circuits or abnormal tripping times. Preferably, but not restrictively, rule-based anomaly identification includes: sampling circuit anomaly identification, action behavior anomaly identification, protection setting anomaly identification, and input / output anomaly identification.

[0042] Further preferred, but not limiting, the sampling loop anomaly identification includes at least one of the following: dual-sampling inconsistency identification, CT saturation identification, and harmonic anomaly identification.

[0043] Furthermore, the dual-sampling inconsistency identification includes: firstly, aligning the waveforms of the dual protection recordings at the same interval according to voltage changes or start-up positions; then, interpolating the second recordings one by one according to the channel; comparing the effective values ​​of the aligned and interpolated waveform data with the first recording data; if the difference in effective values ​​reaches the abnormal threshold, it is determined to be an "inconsistent dual-sampling" abnormality.

[0044] For example, but not limited to, the dual-sampling inconsistency identification method aligns and interpolates the two waveforms of the dual protection systems on the same line, calculates the effective value of current or voltage at the corresponding time for each data point, and if the difference in effective value is greater than 5% and the absolute error is greater than 0.05A or 2V, the channel is judged as having a "dual-sampling inconsistency" anomaly.

[0045] Furthermore, the CT saturation identification includes: when the ratio of the second harmonic amplitude to the fundamental amplitude exceeds a set threshold value, it is determined to be an "CT saturation" anomaly.

[0046] Furthermore, the harmonic anomaly includes: when the amplitude of multiple harmonics and their ratio to the fundamental amplitude exceed a set threshold value, it is determined to be a "harmonic anomaly".

[0047] Further preferred, but not limiting, the abnormal action behavior identification includes at least one of the following: abnormal tripping time identification, and inconsistent identification of two sets of actions.

[0048] Furthermore, the trip time anomaly identification first calculates the first trip command time and the first trip position change time through switch quantity change calculation, and calculates the difference between the two. If the difference is greater than the set threshold value, it is determined to be an "abnormal trip time".

[0049] For example, but not limited to, the trip time anomaly identification calculates the difference between the time of the first trip position change and the time of the first trip command. If the difference is greater than the set threshold value of 80ms, it is determined to be an "abnormal trip time".

[0050] Furthermore, the identification of inconsistent actions between the two sets of protection includes: when one set of protection operates while the other does not, it is determined to be an "inconsistent action between the two sets of protection" anomaly.

[0051] Further preferred, but not limiting, the protection setting anomaly identification includes at least one of the following: protection setting sensitivity coefficient anomaly identification, minimum operating current verification anomaly identification.

[0052] Furthermore, the abnormal identification of the protection setting sensitivity coefficient includes: firstly, calculating the electrical quantity characteristic value at the time of the fault by measuring the voltage and current sudden changes in the recorded waveform data, thereby calculating the protection sensitivity coefficient, and comparing it with the minimum sensitivity coefficient required by the power system regulations, and analyzing whether the action sequence and sensitivity coordination of the main protection and backup protection are reasonable under the same fault.

[0053] Preferred, but not limiting, such as Figure 2 As shown, the sensitivity analysis of the protection setting includes the following steps: Step A1: Calculate the fault time by measuring voltage and current abrupt changes in the recorded waveform data, and calculate the electrical characteristic values ​​at the fault time, such as, but not limited to, the fault current amplitude I_fault and the fault voltage U_fault.

[0054] Step A2: Extract the setpoints from the waveform HDR file, or retrieve the setpoints of the relay protection device to obtain the impedance setpoints Z_set, etc.

[0055] Step A3, calculate the protection sensitivity coefficient: Based on the fault type and protection principle, calculate the ratio of the actual fault quantity to the protection setting value to obtain the sensitivity coefficient K_sen.

[0056] Step A4, Compliance Verification: Compare the calculated sensitivity coefficient K_sen with the minimum sensitivity coefficient K_min required by the power system regulations.

[0057] Step A5, Protection Coordination Analysis: Analyze whether the timing and sensitivity coordination between the main protection and backup protection are reasonable under the same fault.

[0058] Step A6: Output the identification conclusion, such as: "The sensitivity coefficient of the main protection is 1.2, which is lower than the requirement of 1.5 in the regulations, and is judged as insufficient sensitivity" or "The backup protection acts before the main protection, and is judged as protection coordination mismatch".

[0059] Furthermore, the minimum operating current verification anomaly identification includes: first extracting the maximum value of the fault current, and then checking whether the protection device has issued a start or operation signal. If it has not started or operated correctly, it is determined to be "minimum operating current verification anomaly".

[0060] Further preferred, but not limiting, the input / output anomaly identification includes at least one of the following: switch quantity jitter identification, and malfunctioning circuit anomaly identification.

[0061] Furthermore, the switch quantity jitter identification includes: detecting switch quantity changes; if a switch quantity changes frequently more than a set number of times within a certain time period, it is considered to be "switch quantity jitter".

[0062] For example, but not in a limiting sense, the switch jitter identification detects switch position changes. If a switch position changes frequently more than 3 times within 20ms, it is considered to be "switch jitter".

[0063] Furthermore, the failure circuit anomaly identification means that if the corresponding bus protection or circuit breaker protection does not receive a failure start signal after the line or main transformer operates, it is determined to be an "abnormal failure circuit".

[0064] It is worth noting that the waveform image intelligent identification based on the deep learning model includes: taking current and voltage waveforms as input, and automatically learning and identifying them using a convolutional neural network model, preferably but not limited to complex hidden dangers such as CT deep saturation feature waveforms, PT ferroresonant waveforms, and current waveform distortion caused by circuit breaker operating mechanism faults.

[0065] As one of the prominent and essential features of this invention, preferably but not limitingly, such as Figure 3 As shown, the intelligent waveform image recognition based on the deep learning model includes the following steps: Step B1: Input the waveform image into the pre-trained deep learning model to perform feature extraction and pattern recognition on the model.

[0066] Step B2: Output the waveform anomaly type and confidence level.

[0067] Step B3: Integrate the recognition results with the results from other modules to generate a comprehensive recognition report.

[0068] Further, but not limited to, the standardized waveform image generated in step 3 is input into a convolutional neural network (CNN) model. This model has been trained on a large dataset of historical waveforms labeled with various abnormal waveforms, preferably but not limited to "normal waveforms", "CT saturation", and "ferromagnetic resonance". It can automatically extract deep features from the waveforms and classify or regress the images, thereby identifying complex hidden danger patterns that are difficult for the human eye or traditional algorithms to detect.

[0069] Traditional methods for addressing CT saturation issues involve calculating harmonic content, but mild, specific types of saturation may not be significant. In contrast, the significant advancements this invention brings to the prior art include at least the ability of a deep learning-based AI model to accurately identify saturation directly from visual features such as the "shoulder" shape and distortion initiation point of the current waveform.

[0070] Besides comparing time differences, the significant advancements this invention brings to the prior art regarding the potential hazards of slow circuit breaker tripping include at least the following: the AI ​​model can analyze the current waveform image of the trip coil and, through the shape changes of its rising and falling edges, can predict in advance whether there are potential faults such as jamming in the mechanism.

[0071] As one of the most prominent substantive features of this invention, it elevates single-point post-event analysis to wide-area linkage and multi-dimensional parallel intelligent screening. In particular, it introduces AI recognition based on waveform images, which can discover complex hidden dangers that are difficult to capture by traditional methods. This realizes the transformation from "post-event analysis" to "pre-event warning", significantly improving the efficiency and depth of secondary equipment hidden danger identification.

[0072] It is worth noting that in the embodiments of the present invention, "step + number" or "step + letter + number" is only an expression for clearly describing the specific implementation of the secondary equipment hidden danger identification method based on waveform recording files, and is not an absolute restriction on the order of each step. Under the guidance of the core concept of the present invention, changing the order of these steps to obtain the same or similar technical effects all fall within the scope of the present invention.

[0073] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent identification of potential hazards in secondary equipment based on wide-area waveform recording files, characterized in that, Includes the following steps: Fault event triggering and topology expansion include: responding to a fault event of primary equipment in the power grid, performing multi-level substation topology analysis based on the power grid topology model, identifying surrounding substations associated with the fault event, and forming a list of associated substation groups centered on the fault point; Multi-station waveform data retrieval includes: automatically retrieving waveform files from the faulty plant and surrounding plants at the time of the fault based on the list of associated plant groups; Waveform data analysis and feature extraction, including: analyzing waveform files and extracting channel information and waveform image data for hazard identification; Multi-dimensional hazard intelligent identification includes: based on the extracted channel information and waveform image data, multi-dimensional hazard intelligent identification is carried out in parallel on the secondary equipment of the plant through a rule engine and a deep learning model, respectively.

2. The intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files according to claim 1, characterized in that: Hazard identification based on rule engine includes: identification of sampling circuit anomalies, identification of action behavior anomalies, identification of protection setting anomalies, and identification of input and output anomalies.

3. The intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files according to claim 2, characterized in that: The sampling circuit anomaly identification includes at least one of the following: dual-sampling inconsistency identification, CT saturation identification, and harmonic anomaly identification.

4. The intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files according to claim 2 or 3, characterized in that: The abnormal action behavior identification includes at least one of the following: abnormal tripping time identification, and inconsistent identification of two sets of actions.

5. The intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files according to claim 2 or 3, characterized in that: The protection setting anomaly identification includes at least one of the following: protection setting sensitivity coefficient anomaly identification, minimum operating current verification anomaly identification.

6. The intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files according to claim 5, characterized in that: The sensitivity analysis of the protection setpoint includes the following steps: The fault time is calculated by measuring voltage and current abrupt changes in the recorded waveform data, and the electrical characteristic values ​​at the fault time are also calculated. Extract the setpoints from the waveform HDR file, or retrieve the setpoints of the relay protection device to obtain the impedance setpoints; Calculate the protection sensitivity coefficient: Based on the fault type and protection principle, calculate the ratio of the actual fault quantity to the protection setting value to obtain the sensitivity coefficient K_sen; The calculated sensitivity coefficient K_sen is compared with the minimum sensitivity coefficient K_min required by the power system regulations to verify compliance with the regulations. Analyze whether the timing and sensitivity coordination between the main protection and backup protection are reasonable under the same fault. Output the identification conclusion.

7. The intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files according to claim 2 or 3, characterized in that: The input / output anomaly identification includes at least one of the following: switch quantity jitter identification, and malfunctioning circuit anomaly identification.

8. The intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files according to claim 1, characterized in that: Intelligent waveform image recognition based on deep learning models includes: taking current and voltage waveforms as input, using convolutional neural network models to automatically learn and identify potential hazards, including: CT deep saturation characteristic waveforms, PT ferroresonant waveforms, and current waveform distortions caused by circuit breaker operating mechanism faults.

9. The intelligent identification method for potential hazards in secondary equipment based on wide-area waveform recording files according to claim 8, characterized in that: The waveform image intelligent recognition based on the deep learning model includes the following steps: The waveform image is input into a pre-trained deep learning model to perform feature extraction and pattern recognition. Output waveform anomaly type and confidence level; The identification results are integrated with the hazard identification results based on the rule engine to generate a comprehensive identification report.

Images