Electrical equipment automatic fault analysis method and system based on big data

Abstract

The invention discloses an electrical equipment automation fault analysis method and system based on big data, and relates to the technical field of fault analysis, and the method comprises the following steps: constructing an inverter knowledge spectrogram, screening to obtain first data, setting a fault type, simulating to obtain a corresponding output waveform, classifying related characteristics, and constructing a first mapping relation. According to the method, historical data are filtered in advance through the knowledge graph, low-information redundant samples are automatically removed, the subsequent calculation amount is remarkably reduced, the big data processing efficiency is improved, one-key fault injection is carried out on an HIL platform in a fault type and driving voltage table look-up mode, zero-loss and high-throughput fault simulation is achieved, and the fault simulation efficiency is improved. The output waveform is used as an anchor point to complete data automatic classification and feature extraction, an interpretable feature and fault mapping relation is established, on-site rapid positioning is facilitated, the mapping relation is embedded into a knowledge graph in a triple form, online incremental updating is supported, only local data needs to be supplemented when an equipment model or topology is changed, and the maintenance cost is low.

Description

Technical Field

[0001] This invention relates to the technical field of fault analysis, and in particular to a method and system for automated fault analysis of electrical equipment based on big data. Background Technology

[0002] In recent years, electrical equipment automation fault analysis technology has developed rapidly, especially with the widespread adoption of models such as Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), and Autoencoders (AE). These models automatically extract nonlinear fault features from complex time-series data and high-dimensional features, significantly improving the accuracy of fault identification and prediction. Combining CNN with LSTM, or introducing transfer learning techniques, allows models trained on one device to be transferred to other similar devices, significantly improving the efficiency of fault diagnosis across regions and devices. Through data mining algorithms such as cluster analysis and association rule mining, a fault feature library is constructed to achieve early identification and warning of potential faults.

[0003] Currently, Chinese invention patent CN112782512A discloses a method and device for judging the state and diagnosing faults of electrical equipment. This method determines a matching standard parameter curve based on the target operating state of the target electrical equipment within a target time period. The standard parameter curve is generated by the electrical quantities of the target electrical equipment during normal operation in the target operating state. The target parameter curve is compared with the standard parameter curve to generate a parameter comparison result. The fault type of the target electrical equipment within the target time period is determined based on the parameter comparison result. However, the related technology does not screen historical data according to screening criteria, which is not conducive to the effectiveness of the data, the rapid response of the analysis, and the establishment of a rapid mapping relationship based on the analysis results, which is not conducive to the timeliness of obtaining the analysis results. Summary of the Invention

[0004] The technical problem solved by this invention is that related technologies do not screen historical data according to screening criteria, which is not conducive to the effectiveness of the data, the rapid responsiveness of the analysis, and the lack of a rapid mapping relationship based on the analysis results, which is not conducive to the timeliness of obtaining the analysis results.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Firstly, a method for automated fault analysis of electrical equipment based on big data, comprising the following steps: Step S100: Construct an inverter knowledge graph based on historical operating data, and filter the historical operating data based on the inverter knowledge graph to obtain the first data; Step S200: Set the fault type, match the corresponding drive voltage according to the fault type, simulate the fault according to the drive voltage, and obtain the corresponding output waveform; Step S300: Classify the first data according to the output waveform, extract the relevant features of the classified first data, and construct the first mapping relationship between the relevant features and the corresponding fault types.

[0006] As a preferred embodiment of the big data-based electrical equipment automation fault analysis method of the present invention, the historical operating data includes historical input three-phase voltage, historical neutral point voltage, historical interval duration and historical inverter temperature. The historical interval duration is the interval duration for inverter state switching, which is calculated to include a first duration, a second duration and a third duration. The historical inverter temperature is the historical temperature after the historical interval duration. The three-phase voltages are denoted as the first-phase voltage, the second-phase voltage, and the third-phase voltage, and the three phases are labeled as phase u, phase v, and phase w, respectively. The neutral point is the reference point of the voltage in the main circuit of the three-phase inverter, denoted as N. The topology of the main circuit of the three-phase inverter is a fully controlled bridge topology. The bridge arms corresponding to phases u, v and w are denoted as bridge arm a, bridge arm b and bridge arm c. The neutral point is represented as a three-phase three-wire neutral point. According to the voltage law of the fully controlled bridge topology, the voltage change law and flow direction law of the bridge arm corresponding to the first phase voltage, the second phase voltage and the third phase voltage are the same. The first set of data is obtained by filtering historical operating data based on historical neutral point voltage.

[0007] As a preferred embodiment of the big data-based electrical equipment automation fault analysis method of the present invention, the method for constructing the inverter knowledge graph includes: Set the switch transistor number. After the switch transistor number is set, set the switch state according to the voltage difference between any bridge arm and the neutral point voltage. Mark the circuit operation state according to the switch state. Set the inverter knowledge spectrum according to the marked circuit operation state. The switch states include a first state, a second state, and a third state.

[0008] As a preferred embodiment of the big data-based automated fault analysis method for electrical equipment described in this invention, the method for numbering the switching transistors includes: The switching transistor connecting the U-phase input AC voltage and the a-arm bridge is denoted as S in the direction of the positive input current. u1 S u2 S u3 and S u4 The switching transistor connecting the V-phase input AC voltage and the B-arm bridge is denoted as S in the direction of the positive input current. v1 S v2 S v3and S v4 The switching transistor connecting the W-phase input AC voltage and the C-bridge arm is denoted as S in the direction of the positive input current. w1 S w2 S w3 and S w4 ; The methods for setting the switch status include: Select the circuit corresponding to bridge arm a, and input a positive voltage from the voltage input port. This positive voltage is half of the total voltage and is denoted as the first voltage, S. u1 and S u2 The call was connected, S u3 and S u4 When disconnected, the neutral point voltage is detected as 0, the equipotential of bridge arm a is the first voltage, the first difference between the first voltage and 0 is calculated, and the switch state corresponding to the first difference is set as the first state. A negative voltage is input from the voltage input port. The absolute value of this negative voltage is half of the total voltage, denoted as the second voltage, S. u1 and S u2 Disconnected, S u3 and S u4 When the circuit is switched on, the neutral point voltage is detected to be 0, the equipotential of the a-arm is the second voltage, the first difference between the second voltage and 0 is calculated, and the switch state corresponding to the second difference is set to the second state. A positive voltage and a negative voltage are input from the voltage input port, respectively. The positive voltage is half of the total voltage, and the absolute value of the negative voltage is half of the total voltage. These are denoted as the first voltage and the second voltage, respectively. u2 and S u3 The call was connected, S u1 and S u4 When the circuit is switched on, the neutral point voltage is detected as the sum of the first voltage and the second voltage. The equipotential of the a-arm is the sum of the first voltage and the second voltage. The sum of the first voltage and the second voltage is recorded as the first sum. The third difference between the first sum and the first sum is calculated. The switch state corresponding to the third difference is set as the third state. The switch state is marked as the circuit operating state, which includes the first state, the second state, and the third state.

[0009] As a preferred embodiment of the big data-based electrical equipment automation fault analysis method of the present invention, the historical operating data is filtered according to the marking logic of the circuit operating state. The filtering method for the historical operating data includes: calculating the corresponding first difference, second difference, and third difference; calculating the difference between the first difference and the first voltage; calculating the difference between the second difference and the second voltage; calculating the difference between the third difference and the third voltage, and recording them as the first error, second error, and third error, respectively; setting 0.5 as the error threshold; comparing the first error, second error, and third error with 0.5 respectively; when the first error, second error, or third error is greater than 0.5, the corresponding historical operating data is deleted; when all the first error, second error, and third error are greater than 0.5, the corresponding historical operating data is retained. Select the retained historical operation data, and in chronological order, select the adjacent historical operation data corresponding to the transition from the first state to the second state. Obtain the historical interval duration of the adjacent historical operation data. Iterate through each historical interval duration and calculate the first average of the historical interval duration. Record the first average as the first duration. Select the adjacent historical operation data corresponding to the transition from the second state to the third state, and obtain the historical interval duration of the adjacent historical operation data. Iterate through each historical interval duration and calculate the second average of the historical interval duration. Record the second average as the second duration. Take the adjacent historical operation data corresponding to the transition from the first state to the third state, obtain the historical interval duration of the adjacent historical operation data, iterate through each historical interval duration, calculate the third average of the historical interval duration, and record the third average as the third duration. The first duration is set as the interval between switching from the first state to the second state, the second duration is set as the interval between switching from the second state to the third state, and the third duration is set as the interval between switching from the first state to the third state. Here, switching means that the on switch corresponding to the previous state turns off at the same time, and after the corresponding interval, the off switch corresponding to the next state turns on at the same time.

[0010] As a preferred embodiment of the big data-based electrical equipment automation fault analysis method of the present invention, the method comprises: obtaining the corresponding historical inverter temperature, setting a first value as a temperature threshold, setting a second value as a change amount, wherein the change amount and duration have the same dimension, comparing the historical inverter temperature with the first value, and when the historical inverter temperature is greater than the first value, continuously weighting the corresponding interval duration and the change amount to obtain new interval durations, continuously re-filtering historical operating data under the new interval duration and continuously obtaining new historical inverter temperatures until the new historical inverter temperature is less than or equal to the first value, stopping the continuous weighting calculation of the corresponding interval duration and the change amount, and setting the interval duration at this time as the comprehensive interval duration when the corresponding circuit operating state switches. Establish a correspondence between the switching combinations of circuit operating states and the corresponding interval durations, and set the method of filtering historical operating data and the correspondence as the inverter knowledge spectrum.

[0011] As a preferred embodiment of the big data-based electrical equipment automation fault analysis method of the present invention, the method for obtaining the first data by filtering historical operating data according to the inverter knowledge spectrum includes: Based on the method of filtering historical operating data, the retained historical operating data is obtained; Based on the corresponding relationship, the comprehensive interval duration corresponding to the switching between different circuit operating states is obtained; Set the retained historical operation data and comprehensive interval duration as the first data.

[0012] As a preferred embodiment of the big data-based electrical equipment automation fault analysis method of the present invention, the fault types are set, including: any switch open circuit fault of any bridge arm in the same phase; one switch device in the upper half and one in the lower half of the bridge arm in the same phase are simultaneously open; two switch devices in the upper half of the bridge arm in different phases are simultaneously open; two switch devices in the lower half of the bridge arm in different phases are simultaneously open; two switch devices in positions 1 and 4 of the cross bridge arm in different phases are simultaneously open; two switch devices in positions 2 and 3 of the cross bridge arm in different phases are simultaneously open; two switch devices in positions 1 and 3 of the cross bridge arm in different phases are simultaneously open; and two switch devices in positions 2 and 4 of the cross bridge arm in different phases are simultaneously open. Matching the corresponding drive voltage according to the type of fault, the method for matching the drive voltage includes: For any open-circuit fault of any switch in any bridge arm of the same phase: set the drive voltage signal of the switch to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of one switching device in the upper half and one in the lower half of the same phase bridge arm: set the drive voltage signal of the corresponding switching device in the upper half and the lower half to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices in the upper half of the bridge arm in different phases: set the drive voltage signal of the switching device in the upper half of the bridge in the two different phases to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices in the lower half of the bridge arm in different phases: set the drive voltage signal of the switching device in the lower half of the bridge in the two different phases to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices at positions 1 and 4 on different phase cross-bridge arms: set the drive voltage signal of the switching devices at positions 1 and 4 to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices at positions 2 and 3 on different phase cross-bridge arms: set the drive voltage signal of the switching devices at positions 2 and 3 to a continuous low level or a high impedance state; For a simultaneous open-circuit fault of two switching devices at positions 1 and 3 on different phase cross-bridge arms: set the drive voltage signal of the switching devices at positions 1 and 3 to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices at positions 2 and 4 on different phase cross-bridge arms: set the drive voltage signal of the switching devices at positions 2 and 4 to a continuous low level or a high impedance state. Wherein, the low voltage means that the driver output voltage is less than or equal to 0.3V and the source capability is greater than or equal to 5mA, and the high impedance means that the output leakage current is less than or equal to 1µA and the gate is only pulled to the source through a pull-down resistor of greater than or equal to 100kΩ. The fault is simulated based on the driving voltage to obtain the corresponding output waveform. The simulation is performed in simulation software, and the output waveform is represented as the output voltage waveform.

[0013] As a preferred embodiment of the big data-based electrical equipment automation fault analysis method of the present invention, wherein: the first data is classified according to the output waveform, and the classification means classifying each output waveform into a set fault type; Extracting relevant features from the first categorized data, where the first data feature is represented by the average of the extreme values, the method for extracting the relevant features includes: Select each output waveform under any fault type, extract the pole values ​​of any output waveform, number each pole value in chronological order, the number is a natural number, select any numbered pole value, calculate the average value of the numbered pole values ​​of each output waveform, traverse each pole value number, and obtain the average value of each pole value of the output waveform. Construct a first mapping relationship between relevant features and corresponding fault types. The expression for the first mapping relationship is: F: Fault type = {k1,k2,...k} L}; Where F is the mapping rule for the first mapping relationship, and k L The average value of the poles numbered L for the fault type; The current output waveform is obtained, and the current pole value of the current output waveform is extracted. Any fault type is selected, and the average value of each pole value is obtained according to the first mapping relationship. The difference between the current pole value and the average value of the corresponding pole value is calculated, and the average value of the difference between each current pole value and the average value of the corresponding pole value is calculated and recorded as the first offset value. 0.2 is set as the offset value threshold. The first offset value is compared with 0.2. When the first offset value is greater than 0.2, the first mapping relationship of the next fault type is skipped. When the first offset value is less than or equal to 0.2, the corresponding fault type is set as the current fault type. When all fault types have been skipped and all the first offset values ​​are greater than 0.2, the current fault type is set to none. When the current fault type is set to none, it means that the current working state of the inverter is normal.

[0014] Secondly, an automated fault analysis system for electrical equipment based on big data includes a screening module, a simulation module, and a construction module; The filtering module constructs an inverter knowledge graph based on historical operating data, and filters the historical operating data based on the inverter knowledge graph to obtain the first data; The simulation module sets the fault types, matches the corresponding driving voltage according to the fault types, simulates the faults according to the driving voltages, and obtains the corresponding output waveforms. The construction module classifies the first data according to the output waveform, extracts the relevant features of the classified first data, and constructs a first mapping relationship between the relevant features and the corresponding fault types.

[0015] The beneficial effects of this invention are as follows: By filtering historical data through a knowledge graph and automatically removing low-information-redundant samples, the amount of subsequent computation is significantly reduced, and the efficiency of big data processing is improved. Faults are injected into the HIL platform with one click using a table lookup method based on fault type and driving voltage. Various open-circuit conditions can be reproduced without disassembly or damage, achieving zero-loss, high-throughput fault simulation. The output waveform is used as an anchor point to complete automatic data classification and feature extraction, establishing an interpretable feature-fault mapping relationship, making the fault type clear at a glance and facilitating rapid on-site location. The mapping relationship is embedded in the knowledge graph in the form of triples, supporting online incremental updates. When the equipment model or topology changes, only local data needs to be supplemented, resulting in low maintenance costs. Attached Figure Description

[0016] Figure 1 A schematic diagram of the basic process of an automated fault analysis method for electrical equipment based on big data, provided as an embodiment of the present invention; Figure 2 Inverter main circuit diagram for a big data-based electrical equipment automation fault analysis method provided in one embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example, refer to Figure 1 As an embodiment of the present invention, an automated fault analysis method for electrical equipment based on big data is provided, comprising the following steps: Step S100: Construct an inverter knowledge graph based on historical operating data, and filter the historical operating data based on the inverter knowledge graph to obtain the first data; Step S200: Set the fault type, match the corresponding drive voltage according to the fault type, simulate the fault according to the drive voltage, and obtain the corresponding output waveform; Step S300: Classify the first data according to the output waveform, extract the relevant features of the classified first data, and construct the first mapping relationship between the relevant features and the corresponding fault types.

[0019] This invention filters historical data through a knowledge graph, automatically removing low-information-redundant samples, significantly reducing subsequent computational load and improving big data processing efficiency. It uses a fault type and drive voltage lookup table to inject faults with one click on the HIL platform, reproducing various open-circuit conditions without disassembly or damage, achieving zero-loss, high-throughput fault simulation. Using the output waveform as an anchor point, it automatically classifies and extracts features, establishing an interpretable feature-fault mapping relationship, making fault types clear at a glance and facilitating rapid on-site location. The mapping relationship is embedded in the knowledge graph in the form of triples, supporting online incremental updates. When the equipment model or topology changes, only local data needs to be supplemented, resulting in low maintenance costs.

[0020] Historical operating data includes historical input three-phase voltage, historical neutral point voltage, historical interval duration, and historical inverter temperature. The historical interval duration is the interval duration during which the inverter performs state switching. It is calculated to include the first duration, the second duration, and the third duration. The historical inverter temperature is the historical temperature after the historical interval duration. The three-phase voltages are denoted as the first-phase voltage, the second-phase voltage, and the third-phase voltage, and the three phases are labeled as phase u, phase v, and phase w, respectively. The neutral point is the reference point for the voltage in the main circuit of the three-phase inverter, denoted as N. The topology of the main circuit of the three-phase inverter is a fully controlled bridge topology. The bridge arms corresponding to phases u, v, and w are denoted as bridge arm a, bridge arm b, and bridge arm c. The neutral point is represented as a three-phase three-wire neutral point. According to the voltage law of the fully controlled bridge topology, the voltage change law and flow direction law of the bridge arm corresponding to the first phase voltage, the second phase voltage and the third phase voltage are the same. The first set of data is obtained by filtering historical operating data based on historical neutral point voltage.

[0021] In practice, historical operating data is simultaneously incorporated into three-phase voltage, neutral point voltage, state switching interval duration, and temperature after the interval ends, forming a multi-domain coupled original profile of "electrical-temporal-thermal" data. This provides a high-dimensional, low-redundancy information source for subsequent knowledge graphs. The "interval duration" is subdivided into first, second, and third durations, which can accurately characterize the dynamic residence time of the inverter in different commutation stages, allowing potential aging or drive anomalies to be exposed in advance. Using the three-phase three-wire neutral point N as a unified reference, the inherent voltage law of the fully controlled bridge topology is utilized to normalize the voltage / current variation law of the bridge arms of phases u, v, and w, eliminating phase sequence differences and achieving cross-phase data alignment and direct comparison. By using the neutral point voltage to pre-screen historical operating data, transient disturbances or measurement outliers in the power grid can be automatically filtered out, retaining the effective fragments that are truly related to device behavior, significantly improving the signal-to-noise ratio and computational efficiency of subsequent fault modeling.

[0022] Methods for constructing inverter knowledge graphs include: Set the switch transistor number. After the switch transistor number is set, set the switch state according to the voltage difference between any bridge arm and the neutral point voltage. Mark the circuit operation state according to the switch state. Set the inverter knowledge spectrum according to the marked circuit operation state. The switch states include the first state, the second state, and the third state.

[0023] The methods for numbering switching transistors include: The switching transistor connecting the U-phase input AC voltage and the a-arm bridge is denoted as S in the direction of the positive input current. u1 S u2 S u3 and S u4 The switching transistor connecting the V-phase input AC voltage and the B-arm bridge is denoted as S in the direction of the positive input current. v1 S v2 S v3 and S v4 The switching transistor connecting the W-phase input AC voltage and the C-bridge arm is denoted as S in the direction of the positive input current. w1 S w2 S w3 and S w4 ; The methods for setting the switch status include: Select the circuit corresponding to bridge arm a, and input a positive voltage from the voltage input port. The positive voltage is half of the total voltage, denoted as the first voltage, S. u1 and S u2 The call was connected, S u3 and S u4 When disconnected, the neutral point voltage is detected as 0, the equipotential of bridge arm a is the first voltage, the first difference between the first voltage and 0 is calculated, and the switch state corresponding to the first difference is set as the first state. A negative voltage is input from the voltage input port. The absolute value of the negative voltage is half of the total voltage, denoted as the second voltage, S. u1 and S u2 Disconnected, S u3 and S u4 When the circuit is switched on, the neutral point voltage is detected to be 0, the equipotential of the a-arm is the second voltage, the first difference between the second voltage and 0 is calculated, and the switch state corresponding to the second difference is set to the second state. A positive voltage and a negative voltage are input from the voltage input port, respectively. The positive voltage is half of the total voltage, and the absolute value of the negative voltage is half of the total voltage. These are denoted as the first voltage and the second voltage, respectively. u2 and S u3 The call was connected, S u1 and S u4When the circuit is switched on, the neutral point voltage is detected as the sum of the first voltage and the second voltage. The equipotential of the a-arm is the sum of the first voltage and the second voltage. The sum of the first voltage and the second voltage is recorded as the first sum. The third difference between the first sum and the first sum is calculated. The switch state corresponding to the third difference is set as the third state. The switch state is marked as the circuit operating state, which includes the first state, the second state, and the third state.

[0024] In practice, a unified numbering rule of "Su1~Su4, Sv1~Sv4, Sw1~Sw4" is adopted to ensure that the switch numbers of the three-phase four-tube bridge arms correspond one-to-one with the positive direction of the current, eliminating the misalignment of graph nodes caused by naming confusion, and ensuring that cross-phase data can be directly spliced. Using only the single-dimensional criterion of "bridge arm voltage difference - neutral point voltage", the complex circuit at any time can be abstracted into three discrete states: the first, the second, and the third, significantly reducing the node size of the knowledge graph while retaining all commutation information. The three states cover positive level, negative level, and zero level (or clamping). Under basic operating conditions, the voltage trajectory of the fully controlled bridge topology can be completely depicted without additional sensors, providing a unified syntax for subsequent automatic marking, data filtering, and fault injection. The state definition corresponds one-to-one with the driving voltage logic and can be directly embedded into FPGA or DSP firmware to achieve "run-mark-store-while-operating". The graph construction process requires zero manual intervention and has low field deployment costs. Since the state mark is strictly bound to the device number, when a switch tube is open-circuited or has a driving abnormality, the corresponding state sequence will immediately show an illegal transition. The graph can locate the specific tube number in seconds, realizing interpretable fault tracing.

[0025] According to the marking logic of circuit operating status, historical operating data is filtered. The filtering method for historical operating data includes: calculating the corresponding first difference, second difference, and third difference; calculating the difference between the first difference and the first voltage; calculating the difference between the second difference and the second voltage; and calculating the difference between the third difference and the third voltage, which are respectively denoted as the first error, the second error, and the third error. 0.5 is set as the error threshold. The first error, the second error, and the third error are compared with 0.5. When there is a first error, the second error, or the third error greater than 0.5, the corresponding historical operating data is deleted. When all the first errors, the second errors, and the third errors are greater than 0.5, the corresponding historical operating data is retained. Select the retained historical operation data, and in chronological order, select the adjacent historical operation data corresponding to the transition from the first state to the second state. Obtain the historical interval duration of the adjacent historical operation data. Iterate through each historical interval duration and calculate the first average of the historical interval duration. Record the first average as the first duration. Select the adjacent historical operation data corresponding to the transition from the second state to the third state, and obtain the historical interval duration of the adjacent historical operation data. Iterate through each historical interval duration and calculate the second average of the historical interval duration. Record the second average as the second duration. Take the adjacent historical operation data corresponding to the transition from the first state to the third state, obtain the historical interval duration of the adjacent historical operation data, iterate through each historical interval duration, calculate the third average of the historical interval duration, and record the third average as the third duration. The first duration is set as the interval between switching from the first state to the second state, the second duration is set as the interval between switching from the second state to the third state, and the third duration is set as the interval between switching from the first state to the third state. Here, switching means that the on switch corresponding to the previous state turns off at the same time, and after the corresponding interval, the off switch corresponding to the next state turns on at the same time.

[0026] In practice, a second-order error calculation using "difference-further difference" separates measurement noise, sampling drift, and actual state transitions. Errors exceeding the threshold are discarded, automatically removing bad points and retaining high-confidence segments to improve the purity of the spectral data. Only when "the first, second, and third errors are simultaneously greater than the threshold" are they retained. Conversely, the "all wrong is right" principle is used to instantly identify rare but real extreme conditions, providing rare cases for the subsequent abnormal sample library and preventing the false negatives of valid information. The first, second, and third time intervals are obtained by segmenting the statistical intervals according to the "state transition direction," naturally aligning with... Based on the actual commutation dead zone and turn-on delay of the device, the "commutation rhythm" can be calibrated online without the need for an additional oscilloscope to capture waveforms, establishing a time reference for aging assessment. The three durations directly correspond to the "off → on" switching cycle of the drive sequence, and can be compared with the PWM firmware parameters in a closed loop to achieve bidirectional verification between the "pattern and controller". Once drift is detected, it immediately prompts an abnormality in the driver board or gate-level circuit, completing an early warning of faults. The entire screening and duration extraction process only relies on the existing voltage sampling channel and does not increase hardware costs. The algorithm is lightweight and can be completed in a single cycle in a DSP or edge FPGA, making it suitable for batch deployment.

[0027] Obtain the corresponding historical inverter temperature, set the first value as the temperature threshold, and set the second value as the change amount. The change amount and the duration have the same dimension. Compare the historical inverter temperature with the first value. When the historical inverter temperature is greater than the first value, continuously calculate the corresponding interval duration and the change amount by weighting, and continuously obtain the new interval duration. Under the new interval duration, continuously re-filter the historical operating data and continuously obtain the new historical inverter temperature until the new historical inverter temperature is less than or equal to the first value. Stop continuously calculating the corresponding interval duration and the change amount by weighting, and set the interval duration at this time as the comprehensive interval duration when the corresponding circuit operating state switches. Establish the correspondence between the switching combinations of circuit operating states and the corresponding interval durations, and set the filtering method and correspondence of historical operating data as the inverter knowledge spectrum.

[0028] Methods for obtaining primary data by filtering historical operating data based on the inverter knowledge spectrum include: Based on the method of filtering historical operating data, the retained historical operating data is obtained; Based on the corresponding relationship, the comprehensive interval duration corresponding to the switching between different circuit operating states is obtained; Set the retained historical operation data and comprehensive interval duration as the first data.

[0029] In practice, "temperature exceeding the threshold" is used as the iteration trigger condition. The interval duration is compressed in real time using a continuously weighted method, allowing for pre-cooling at the data level. This avoids offline calibration of the thermal accumulation model, truly achieving online closed-loop screening of the "thermal-electrical" coupling. The "comprehensive interval duration" obtained when the iteration stops implicitly contains the device's thermal characteristics, naturally becoming a "thermal-state" edge attribute in the knowledge graph. Subsequent fault injection or lifetime prediction does not require re-measuring the thermal impedance curve. The "state switching combination ↔ comprehensive interval duration" is solidified into a graph template, allowing for easy replacement in the field. For power modules or heat dissipation systems, new edge weights can be automatically generated with just one iteration. Old models can be reused with zero code modification, and maintenance costs are close to zero. The first set of filtered data contains both "high-fidelity electrical segments" and "time sequence beats after thermal convergence," which can be directly used to train temperature-sensitive fault diagnosis models. This solves the problem of false high-temperature alarms caused by the "electricity without heat" of traditional big data. The entire process relies only on existing voltage and temperature sampling channels without adding sensors. The algorithm is lightweight and can be completed at the edge gateway. The iteration convergence time is less than 1 minute, making it suitable for batch unit synchronous upgrades.

[0030] The fault types are set, including: any open circuit fault of any switch in any arm of the same phase; one switch in the upper half and one switch in the lower half of the same phase arm are open circuit simultaneously; two switches in the upper half of the different phase arm are open circuit simultaneously; two switches in the lower half of the different phase arm are open circuit simultaneously; two switches in positions 1 and 4 of the different phase cross arm are open circuit simultaneously; two switches in positions 2 and 3 of the different phase cross arm are open circuit simultaneously; two switches in positions 1 and 3 of the different phase cross arm are open circuit simultaneously; and two switches in positions 2 and 4 of the different phase cross arm are open circuit simultaneously. Match the corresponding drive voltage according to the type of fault. The drive voltage matching methods include: For any open-circuit fault of any switch in any bridge arm of the same phase: set the drive voltage signal of the switch to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of one switching device in the upper half and one in the lower half of the same phase bridge arm: set the drive voltage signal of the corresponding switching device in the upper half and the lower half to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices in the upper half of the bridge arm in different phases: set the drive voltage signal of the switching device in the upper half of the bridge in the two different phases to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices in the lower half of the bridge arm in different phases: set the drive voltage signal of the switching device in the lower half of the bridge in the two different phases to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices at positions 1 and 4 on different phase cross-bridge arms: set the drive voltage signal of the switching devices at positions 1 and 4 to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices at positions 2 and 3 on different phase cross-bridge arms: set the drive voltage signal of the switching devices at positions 2 and 3 to a continuous low level or a high impedance state; For a simultaneous open-circuit fault of two switching devices at positions 1 and 3 on different phase cross-bridge arms: set the drive voltage signal of the switching devices at positions 1 and 3 to a continuous low level or a high impedance state. For a simultaneous open-circuit fault of two switching devices at positions 2 and 4 on different phase cross-bridge arms: set the drive voltage signal of the switching devices at positions 2 and 4 to a continuous low level or a high impedance state. Low voltage means the driver output voltage is less than or equal to 0.3V and the source capability is greater than or equal to 5mA. High impedance means the output leakage current is less than or equal to 1µA and the gate is only connected to the source through a pull-down resistor of greater than or equal to 100kΩ. The fault is simulated based on the driving voltage to obtain the corresponding output waveform. The simulation is performed in simulation software, and the output waveform is represented as the output voltage waveform.

[0031] In practice, a single "fault type ↔ drive voltage" lookup table can cover all open-circuit combinations of single transistors, in-phase, out-of-phase, and cross-arm bridges, without requiring physical disassembly, achieving "zero hardware loss" fault injection. Low-level and high-impedance states provide quantified electrical boundaries, ensuring reliable turn-off of the switching transistors while avoiding false triggering of the drive line due to floating. The simulation results are repeatable and reproducible, with the simulation terminal directly outputting voltage waveforms that are naturally isomorphic to the subsequent "state-waveform" nodes in the knowledge graph, eliminating the need for actual sensor wiring and isolation. The modeling cycle for a single device is shortened from "days" to "minutes," for example, injecting "u-phase-Su2 open circuit" → Su2 drive... The system is set to a high impedance state (leakage current 0.8µA, gate 100kΩ pull-down), and "cross 1,4 dual open" is injected → Su1 and Sv4 are simultaneously driven to low level (0.25V, source capability 8mA). The simulation runs for 0.2s, the line voltage Uuv waveform is acquired, the waveform is automatically uploaded to the edge node, the notch depth and harmonic cluster are extracted, and compared with the "waveform fingerprint" node in the knowledge graph. The fault location is completed in 0.3s and the triplet is written back. All eight types of faults of the whole machine are completed in less than 15 minutes. No hardware modification is required. The field engineer can complete the "fault injection-waveform acquisition-graph update" closed loop by simply clicking the mouse.

[0032] The first data is categorized based on the output waveform. Categorization means classifying each output waveform into a pre-defined fault category. Extract relevant features from the first set of categorized data. These features are represented as the average of the extreme values. Methods for extracting relevant features include: Select each output waveform under any fault type, extract the pole values ​​of any output waveform, number each pole value in chronological order, and select any numbered pole value to calculate the average value of the numbered pole values ​​of each output waveform. Iterate through each pole value number to obtain the average value of each pole value of the output waveform. Construct a first mapping relationship between relevant features and corresponding fault types. The expression for the first mapping relationship is: F: Fault type = {k1,k2,...k} L}; Where F is the mapping rule for the first mapping relationship, and k L This represents the average value of the poles numbered L for each fault type. Obtain the current output waveform, extract the current pole value of the current output waveform, select any fault type, obtain the average value of each pole value according to the first mapping relationship, calculate the difference between the current pole value and the average value of the corresponding pole value, calculate the average of the differences between each current pole value and the average value of the corresponding pole value, and record it as the first offset value. Set 0.2 as the offset value threshold, compare the first offset value with 0.2. When the first offset value is greater than 0.2, jump to the first mapping relationship of the next fault type. When the first offset value is less than or equal to 0.2, set the corresponding fault type as the current fault type. When all fault types have been jumped and all the first offset values ​​are greater than 0.2, set the current fault type to none. When the current fault type is set to none, it means that the current inverter is in normal working state.

[0033] In practical implementation, the "average value of poles" is used as the sole feature. It has extremely low dimensionality and is naturally immune to waveform translation and amplitude scaling, eliminating the large amount of computation required by traditional Fourier / wavelet methods. Real-time comparison can be completed at the microcontroller-level edge nodes. The first mapping relationship directly expresses the "fault type" as a pole sequence template with clear physical meaning. Field engineers can understand, modify, or add new fault types without a deep learning background. After point-by-point subtraction, only the "first offset value" is retained as a scalar. The category can be determined by comparing it with the threshold of 0.2. The algorithm has few branches and a definite execution path, meeting the "zero unpredictable delay" requirement of industrial controllers. When all template comparisons fail, it is automatically classified as "no fault," avoiding the forced labeling of normal operating conditions, significantly reducing the false alarm rate, and solving the downtime losses caused by "overdiagnosis."

[0034] This invention filters historical data through a knowledge graph, automatically removing low-information-redundant samples, significantly reducing subsequent computational load and improving big data processing efficiency. It uses a fault type and drive voltage lookup table to inject faults with one click on the HIL platform, reproducing various open-circuit conditions without disassembly or damage, achieving zero-loss, high-throughput fault simulation. Using the output waveform as an anchor point, it automatically classifies and extracts features, establishing an interpretable feature-fault mapping relationship, making fault types clear at a glance and facilitating rapid on-site location. The mapping relationship is embedded in the knowledge graph in the form of triples, supporting online incremental updates. When the equipment model or topology changes, only local data needs to be supplemented, resulting in low maintenance costs.

[0035] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium is implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0036] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention should all be covered within the scope of the claims of the present invention.

Claims

1. A method for electrical equipment automated fault analysis based on big data, characterized in that, The method comprises the following steps: Step S100, constructing an inverter knowledge spectrum according to historical operation data, screening the historical operation data according to the inverter knowledge spectrum, and obtaining first data; Step S200, setting a fault type, matching a corresponding drive voltage according to the fault type, simulating the fault according to the drive voltage, and obtaining a corresponding output waveform; Step S300, classifying the first data according to the output waveform, extracting relevant features of the classified first data, and constructing a first mapping relationship between the relevant features and the corresponding fault type.

2. The big data based electrical equipment automation fault analysis method as claimed in claim 1, wherein: The historical operation data includes historical input three-phase voltage, historical neutral point voltage, historical interval duration and historical inverter temperature, the historical interval duration is represented as the interval duration of the inverter state switching, and is obtained by calculation, the historical interval duration includes a first duration, a second duration and a third duration, and the historical inverter temperature is represented as a historical temperature after the historical interval duration; The three-phase voltage is respectively denoted as a first-phase voltage, a second-phase voltage and a third-phase voltage, and the three phases are respectively marked as a u phase, a v phase and a w phase; The neutral point is a reference point of voltage in a main circuit of a preset three-phase inverter, denoted as N, and the topology structure of the three-phase inverter main circuit is a full-bridge topology, the bridge arms corresponding to the u phase, the v phase and the w phase are denoted as a bridge arm, a bridge arm and a bridge arm; The neutral point is represented as a three-phase three-wire neutral point, according to the voltage law of the full-bridge topology, the voltage variation law and the flow direction law of the bridge arm working corresponding to the first-phase voltage, the second-phase voltage and the third-phase voltage are obtained. The historical operation data is screened according to the historical neutral point voltage, and the first data is obtained.

3. The big data based electrical equipment automation fault analysis method of claim 1, wherein: The construction method of the inverter knowledge spectrum comprises: After the switch tube number is set, the switch state is set according to the voltage difference value of any bridge arm and the voltage difference value of the neutral point, and the circuit operating state is marked according to the switch state, and the inverter knowledge spectrum is set according to the marked circuit operating state; The switch state includes a first state, a second state and a third state.

4. The big data based electrical equipment automation fault analysis method of claim 3, wherein: The method of setting the switch state comprises: Switches connected between the u-phase input AC voltage and the a-leg are denoted as S u1 , S u2 , S u3 , and S u4 Switches connected between the v-phase input AC voltage and the b-leg are denoted as S v1 , S v2 , S v3 , and S v4 Switches connected between the w-phase input AC voltage and the c-leg are denoted as S w1 , S w2 , S w3 , and S w4 ; The switch state is marked as the circuit operating state, that is, the circuit operating state includes a first state, a second state and a third state. Select the circuit corresponding to the a bridge arm, input the forward voltage from the voltage input port, the forward voltage is half of the total voltage, denoted as the first voltage, S u1 and S u2 be turned on, S u3 and S u4 be turned off, the neutral point voltage is 0, the equipotential of the a bridge arm is the first voltage, calculate the first difference between the first voltage and 0, and set the first difference to the first state. A negative voltage is input from the voltage input port, the absolute value of the negative voltage is half of the total voltage, denoted as a second voltage, S u1 and S u2 are turned off, S u3 and S u4 are turned on, the neutral point voltage is 0, the equipotential of the bridge arm is the second voltage, a first difference value between the second voltage and 0 is calculated, and the second difference value is set as a second state. A positive voltage and a negative voltage are input from the voltage input port, respectively. The positive voltage is half of the total voltage, and the absolute value of the negative voltage is half of the total voltage. These are denoted as the first voltage and the second voltage, respectively. u2 and S u3 The call was connected, S u1 and S u4 When the circuit is switched on, the neutral point voltage is detected as the sum of the first voltage and the second voltage. The equipotential of the a-arm is the sum of the first voltage and the second voltage. The sum of the first voltage and the second voltage is recorded as the first sum. The third difference between the first sum and the first sum is calculated. The switch state corresponding to the third difference is set as the third state. According to the marking logic of the circuit operating state, the historical operation data is screened, and the screening method of the historical operation data comprises: calculating a corresponding first difference, a second difference and a third difference, calculating the difference between the first difference and a first voltage, calculating the difference between the second difference and a second voltage, calculating the difference between the third difference and a third voltage, respectively denoted as a first error, a second error and a third error, setting 0.5 as an error threshold, respectively comparing the first error, the second error and the third error with 0.5, when the first error or the second error or the third error is greater than 0.5, the corresponding historical operation data is deleted, and when all the first error, the second error and the third error are greater than 0.5, the corresponding historical operation data is retained.

5. The big data based electrical equipment automation fault analysis method of claim 4, wherein: ​ The selected reserved historical running data is selected in chronological order, corresponding adjacent historical running data of the first state to the second state transition is selected, the historical interval length of the adjacent historical running data is obtained, each historical interval length is traversed, the first average value of the historical interval length is calculated, the first average value is recorded as the first length, corresponding adjacent historical running data of the second state to the third state transition is selected, the historical interval length of the adjacent historical running data is obtained, each historical interval length is traversed, the second average value of the historical interval length is calculated, the second average value is recorded as the second length, corresponding adjacent historical running data of the first state to the third state transition is selected, the historical interval length of the adjacent historical running data is obtained, each historical interval length is traversed, the third average value of the historical interval length is calculated, and the third average value is recorded as the third length; The first length is set as the interval length of the first state switching the second state, the second length is set as the interval length of the second state switching the third state, and the third length is set as the interval length of the first state switching the third state, wherein the switching means that the on switch corresponding to the previous state is closed at the same time point, and the off switch corresponding to the next state is opened at the same time point after the corresponding interval length.

6. The big data based electrical equipment automation fault analysis method of claim 5, wherein: The corresponding historical inverter temperature is obtained, the first value is set as the temperature threshold, and the second value is set as the change amount, the change amount and the length are the same dimension, the historical inverter temperature is compared with the first value, when the historical inverter temperature is greater than the first value, the corresponding interval length is continuously weighted with the change amount, and a new interval length is continuously obtained, under the new interval length, the historical running data is continuously re-screened, and a new historical inverter temperature is continuously obtained, until the new historical inverter temperature is less than or equal to the first value, the continuous weighted calculation of the corresponding interval length and the change amount is stopped, and the interval length at this time is set as the comprehensive interval length when the circuit running state is switched; The switching combination of the circuit running state and the corresponding interval length are set to establish a corresponding relationship, and the historical running data screening method and the corresponding relationship are set as an inverter knowledge spectrum.

7. The big data based electrical equipment automation fault analysis method of claim 6, wherein: The method for screening the historical running data according to the inverter knowledge spectrum comprises the following steps: According to the method for screening the historical running data, the reserved historical running data is obtained; According to the corresponding relationship, the comprehensive interval length corresponding to the switching between different circuit running states is obtained; The reserved historical running data and the comprehensive interval length are set as the first data.

8. The big data based electrical equipment automation fault analysis method of claim 1, wherein: Set fault categories, the fault categories include: any switch tube open circuit fault of any bridge arm of the same phase, the upper half and the lower half of the same phase bridge arm each has one switch tube device open circuit fault at the same time, the upper half of the different phase bridge arm has two switch tube devices open circuit fault at the same time, the lower half of the different phase bridge arm has two switch tube devices open circuit fault at the same time, the 1, 4 position of the different phase cross bridge arm has two switch tube devices open circuit fault at the same time, the 2, 3 position of the different phase cross bridge arm has two switch tube devices open circuit fault at the same time, the 1, 3 position of the different phase cross bridge arm has two switch tube devices open circuit fault at the same time, the 2, 4 position of the different phase cross bridge arm has two switch tube devices open circuit fault at the same time; According to the fault category matching corresponding drive voltage, the matching method of the drive voltage includes: For any switch tube open circuit fault of any bridge arm of the same phase: the drive voltage signal of the switch tube is set to continuous low level or high resistance state; For the upper half and the lower half of the same phase bridge arm each has one switch tube device open circuit fault at the same time: the drive voltage signal of the corresponding switch tube of the upper half and the lower half is set to continuous low level or high resistance state; For the upper half of the different phase bridge arm has two switch tube devices open circuit fault at the same time: the drive voltage signal of the switch tube located in the upper half bridge of the two different phases is set to continuous low level or high resistance state; For the lower half of the different phase bridge arm has two switch tube devices open circuit fault at the same time: the drive voltage signal of the switch tube located in the lower half bridge of the two different phases is set to continuous low level or high resistance state; For the 1, 4 position of the different phase cross bridge arm has two switch tube devices open circuit fault at the same time: the drive voltage signal of the switch tube located in the 1 and 4 position is set to continuous low level or high resistance state; For the 2, 3 position of the different phase cross bridge arm has two switch tube devices open circuit fault at the same time: the drive voltage signal of the switch tube located in the 2 and 3 position is set to continuous low level or high resistance state; For the 1, 3 position of the different phase cross bridge arm has two switch tube devices open circuit fault at the same time: the drive voltage signal of the switch tube located in the 1 and 3 position is set to continuous low level or high resistance state; For the 2, 4 position of the different phase cross bridge arm has two switch tube devices open circuit fault at the same time: the drive voltage signal of the switch tube located in the 2 and 4 position is set to continuous low level or high resistance state; Wherein, the low electricity means that the driver output voltage is less than or equal to 0.3V and the source ability is greater than or equal to 5mA, the high resistance state means that the output leakage current is less than or equal to 1µA, and the gate is only through the pull-down resistance greater than or equal to 100kΩ to the source; According to the drive voltage, the corresponding output waveform is obtained by simulating the fault, the simulation is carried out in the simulation software, and the output waveform is represented as the output voltage waveform.

9. The big data based electrical equipment automation fault analysis method of claim 8, wherein: According to the output waveform, the first data is classified, and the classification is represented as classifying each output waveform into the set fault category. Extracting relevant features of the classified first data, the first data features are represented as average values of pole values, and the method for extracting the relevant features comprises: Selecting each output waveform under any fault category, extracting the pole values of any output waveform, numbering each pole value in time sequence, the numbering is natural number, selecting any numbered pole value, calculating the average value of the numbered pole value of each output waveform, traversing each pole value number to obtain the average value of each pole value of the output waveform; Constructing a first mapping relationship between the relevant features and the corresponding fault categories, the expression of the first mapping relationship is: F: fault category = {k1, k2,... k L}; wherein F is a mapping rule of the first mapping relationship, k L is an average value of pole values numbered L of the failure category. Obtaining a current output waveform, extracting a current pole value of the current output waveform, selecting any fault category, obtaining the average value of each pole value according to the first mapping relationship, calculating the difference between the current pole value and the average value of the corresponding pole value, calculating the average value of the difference between each current pole value and the average value of the corresponding pole value, denoted as a first offset value, setting 0.2 as an offset value threshold, comparing the first offset value with 0.2, when the first offset value is greater than 0.2, jumping to the first mapping relationship of the next fault category, when the first offset value is less than or equal to 0.2, setting the corresponding fault category as the current fault category, when jumping through all fault categories and all first offset values are greater than 0.2, setting the current fault type as none, when the current fault type is set as none, indicating that the working state of the current inverter is normal state.

10. A big data based electrical equipment automation fault analysis system for performing the big data based electrical equipment automation fault analysis method as claimed in claim 1, characterized in that, It comprises a screening module, a simulation module and a construction module; The screening module constructs an inverter knowledge spectrum according to historical operation data, screens the historical operation data according to the inverter knowledge spectrum to obtain first data; The simulation module sets a fault category, matches a corresponding driving voltage according to the fault category, simulates the fault according to the driving voltage to obtain a corresponding output waveform; The construction module classifies the first data according to the output waveform, extracts relevant features of the classified first data, and constructs a first mapping relationship between the relevant features and the corresponding fault categories.

Images