Power system fault time automatic identification analysis method and system based on multi-source data fusion

Abstract

This invention proposes an automatic fault time identification and analysis method and system for power systems based on multi-source data fusion. It includes a three-level mechanism: a multi-dimensional data preprocessing module, a fault type adaptive identification module, a dynamic steady-state value calculation module, a fault entry time judgment module, a fault time correction module, and a standardized result output module. It constructs a fully automated analysis method encompassing "data preprocessing – fault type adaptive matching – steady-state feature extraction – multi-condition collaborative judgment – ​​intelligent time point correction." This invention improves the accuracy and robustness of fault time identification, automatically adapts to common faults such as low-voltage and high-voltage faults, and has high scenario coverage. It achieves end-to-end automated processing from data preprocessing to result output, thus improving efficiency.

Description

Technical Field

[0001] This invention relates to the field of power system simulation and analysis technology, specifically to power system fault identification and analysis technology. Background Technology

[0002] In the operation and analysis of dynamic simulation systems for renewable energy grid connection, accurate identification of fault times (including fault entry time and fault exit time) is a core prerequisite for subsequent data validity verification, system error assessment, and fault mechanism analysis, directly determining the application value of the simulation results. However, current mainstream fault time identification technologies in the industry still have significant technical bottlenecks, specifically as follows:

[0003] 1. Limitations of manual-dependent recognition: Traditional methods require engineers to manually observe simulation curves and judge fault times based on experience. This not only leads to poor consistency of recognition results due to differences in subjective cognition (the recognition error between different people on the same dataset can be more than 0.05s), but also requires analysis of each massive amount of simulation data, resulting in low processing efficiency (the average processing time for a single dataset exceeds 30 minutes), making it difficult to meet the needs of large-scale simulation tasks.

[0004] 2. Adaptability limitations of fixed threshold algorithms: Existing automatic identification schemes employ a "single fixed threshold + simple logical judgment" model, which can only adapt to specific fault scenarios (such as a single low-voltage fault). When the fault type changes (such as from low-voltage to high-voltage) or system parameters fluctuate, the threshold needs to be manually readjusted, making it unable to cope with complex and ever-changing power system simulation conditions. Furthermore, this type of scheme does not consider data noise interference, and when the simulation data fluctuates within ±5%, the identification accuracy drops sharply by more than 50%.

[0005] 3. Insufficient robustness and scenario coverage: Existing technologies lack effective mechanisms for handling scenarios with ambiguous fault boundaries (such as gradual data changes during the fault entry phase and transient fluctuations in the system after the fault exit). At the same time, a unified identification framework for multiple fault types has not been established, which makes it easy to misjudge or miss time points in mixed faults and complex transient processes, and fails to meet the stability and reliability requirements of engineering applications. Summary of the Invention

[0006] The purpose of this invention is to address the problems of heavy reliance on manual intervention, narrow applicability, and weak robustness in existing technologies. It provides an automatic fault time identification and analysis method for power systems based on multi-source data fusion, and proposes a corresponding system. This method achieves high-precision automatic location of fault entry and exit times in power system simulation data. Using the location results as a benchmark, it supports subsequent multi-dimensional data analysis and error quantification assessment, making it suitable for core scenarios such as dynamic simulation of renewable energy grid connection and power system stability testing.

[0007] The technical solution of this invention is as follows: An automatic fault time identification and analysis method for power systems based on multi-source data fusion. The method includes a three-level mechanism for multi-dimensional data preprocessing. The first level involves detecting outliers in the multi-source power system simulation data, marking and smoothing outliers that deviate from the normal data distribution by more than three standard deviations, and using adjacent data weighted interpolation to ensure data continuity. The second level aligns the time series of the multi-source power system simulation data to eliminate time deviations caused by acquisition delays from different data sources. The third level filters high-frequency noise in the multi-source power system simulation data, retaining effective signals related to fault characteristics and improving the anti-interference capability of subsequent processing.

[0008] Step 2: Adaptive fault type identification. Based on the input multi-source parameters of the power system, key feature vectors are extracted. A pre-trained lightweight classifier is used in conjunction with typical fault scenarios to automatically match the fault type and output the feature weight coefficients corresponding to the fault type, providing a basis for subsequent adaptive threshold adjustment.

[0009] Step 3: Dynamic steady-state value calculation. A combination of sliding window weighted average and steady-state convergence judgment is used. Continuous data within a certain period before the fault is automatically extracted as the calculation window. Weights are assigned based on data reliability, and the steady-state benchmark value is calculated by weighted average. Convergence verification: If the data fluctuation within the calculation window is ≤2%, the steady-state value is confirmed to be valid; if the fluctuation exceeds the threshold, the calculation window is automatically expanded to 0.04 s and recalculated.

[0010] Step 4: Fault entry time determination. Based on the feature weight coefficients of the fault type adaptive output, an adaptive threshold + time continuity verification algorithm is used to dynamically adjust the threshold. For low-pass faults, dynamic compensation is based on high-low pass combined. When the threshold condition is met, the data change trend within 0.005 s before and after the threshold condition time point is further verified. The first moment that satisfies both the threshold condition and the data change trend is determined as the fault entry time point.

[0011] Step 5: Fault Time Correction, Bidirectional Search: Based on the initially locked time point, search forward for all candidate points within a 0.005 s range, selecting the point that meets the threshold condition and has the smoothest connection with subsequent fault stage data as the final entry time; search backward for candidate points within a 0.01 s range, selecting the point with the longest system stability duration as the final exit time; Global Reasonableness Verification: Calculate the fault duration. If the duration falls within the preset reasonable range, the correction is complete; if it exceeds the preset reasonable range, an anomaly marker is automatically triggered, and the anomaly reason is output, supporting manual review.

[0012] Step 6: Standardize the output results. The identified and corrected fault time data will be output in accordance with the industry standard format. The output content includes: fault type, fault entry time, fault exit time, fault duration, steady-state baseline value, and data reliability score.

[0013] Furthermore, a multi-condition collaborative judgment step for fault exit time is set up, and a three-dimensional collaborative verification model is constructed, including dimension 1, electrical parameter steady-state recovery verification: detecting whether the voltage and current fall within the steady-state value ±5% after fault exit, and the duration is ≥0.02s;

[0014] Dimension 2, System stability verification: By calculating the voltage fluctuation coefficient ΔU / U≤2% and the current change rate ΔI / Δt≤5A / s, it is determined whether the system has entered a stable operating state;

[0015] Dimension 3, Data Continuity Verification: Compare the data curves 0.01 seconds before and after the fault exit time point to ensure there are no sudden changes or breaks, in order to eliminate false recovery signals caused by abnormal data acquisition;

[0016] Decision logic: The moment is locked as the failure exit time point only when all three dimensions meet the verification conditions.

[0017] As a preferred approach, the first-level outlier detection in step one employs an improved isolated forest algorithm; the second-level time series alignment employs a dynamic time warping algorithm; and the third-level high-frequency noise filtering employs a wavelet threshold denoising algorithm.

[0018] Preferably, in step two, the multi-source parameters of the power system include grid-connected voltage level, renewable energy type, and fault triggering conditions. The lightweight classifier is optimized based on support vector machine (SVM), and the training set contains more than 1,000 typical fault scenarios.

[0019] Preferably, the weighting in step three is as follows: the weight of data at the center of the window is 1.2, and the weight of data at the edge is 0.8.

[0020] Preferably, in step four, when dynamically adjusting the threshold, the following approach is adopted: for low-penetration faults, dynamic compensation is applied based on a combination of high and low penetration.

[0021] Preferably, the reasonable range for step five is 0.05 s-10 s.

[0022] Preferably, the industry standard format for step six adopts IEEE 1547-2018, with fault entry time accurate to 0.001s and fault exit time accurate to 0.001s. The data reliability score is calculated by combining noise intensity and verification pass rate, with a score of 0-100.

[0023] An automatic fault time identification and analysis system for power systems based on multi-source data fusion includes:

[0024] The three-level multi-dimensional data preprocessing module performs three levels of multi-dimensional data preprocessing: Level 1: Detects outliers in the multi-source power system simulation data, marks and smooths outliers that deviate from the normal data distribution by more than three standard deviations, and uses adjacent data weighted interpolation to repair them, ensuring data continuity; Level 2: Aligns the time series of the multi-source power system simulation data to eliminate time deviations caused by acquisition delays from different data sources; Level 3: Filters high-frequency noise in the multi-source power system simulation data, retaining effective signals related to fault characteristics and improving the anti-interference capability of subsequent processing;

[0025] The fault type adaptive identification module is used to identify fault types: based on the input multi-source parameters of the power system, it extracts the key feature vectors, and automatically matches the fault type by using a pre-trained lightweight classifier in conjunction with typical fault scenarios, and outputs the feature weight coefficients corresponding to the fault type.

[0026] The dynamic steady-state value calculation module is used to calculate steady-state values. It employs a combination of sliding window weighted average and steady-state convergence judgment, automatically extracting continuous data within a certain period before the fault as the calculation window, assigning weights based on data reliability, and calculating the steady-state benchmark value through weighted average. Convergence verification: if the data fluctuation within the calculation window is ≤2%, the steady-state value is confirmed to be valid; if the fluctuation exceeds the threshold, the calculation window is automatically expanded to 0.04 s and recalculated.

[0027] The fault entry time judgment module is used to determine the fault entry time. Based on the feature weight coefficients of the fault type adaptive output, the adaptive threshold + time continuity verification algorithm is used to dynamically adjust the threshold. For low-pass faults, dynamic compensation is based on high-low pass combined. When the threshold condition is met, the data change trend within 0.005 s before and after the threshold condition time point is further verified. The first moment that satisfies both the threshold condition and the data change trend is determined as the fault entry time point.

[0028] The fault time correction module is used to correct fault time: Bidirectional search: Based on the initially locked time point, it searches forward for all candidate points within a 0.005 s range, selecting the point that meets the threshold condition and has the smoothest transition with subsequent fault stage data as the final entry time; it also searches backward for candidate points within a 0.01 s range, selecting the point with the longest system stability duration as the final exit time; Global rationality verification: It calculates the fault duration. If the duration falls within a preset reasonable range, the correction is complete; if it exceeds the preset reasonable range, an anomaly marker is automatically triggered, and the anomaly reason is output, supporting manual review.

[0029] The standardized result output module is used to output the fault time determination results: it outputs the identified and corrected fault time data in accordance with the industry standard format. The output content includes: fault type, fault entry time, fault exit time, fault duration, steady-state benchmark value, and data reliability score.

[0030] The system also includes a multi-condition collaborative judgment module for fault exit time, which performs verification in three dimensions: electrical parameter steady-state recovery verification, system stability verification, and data continuity verification. When all three dimensions meet the verification conditions, the time is locked as the fault exit time point.

[0031] The beneficial effects of this invention are: (1) a breakthrough improvement in recognition accuracy:

[0032] Relying on the "adaptive threshold algorithm + multi-condition collaborative judgment", the location error of fault entry time and exit time is ≤0.002s, which improves the accuracy by 90% compared with traditional manual identification and by 70% compared with fixed threshold algorithm. In a dataset containing 10% noise, the identification accuracy is still maintained above 95%, and the robustness is significantly better than the existing solution.

[0033] (2) High adaptability across all scenarios: Through adaptive fault type identification, it can automatically adapt to more than 95% of common fault scenarios such as low-voltage and high-voltage, without the need for manual parameter adjustment; at the same time, it is compatible with simulation data of different voltage levels (10kV-500kV) and different renewable energy types (photovoltaic, wind power), and the scenario coverage is 60% higher than that of existing technologies.

[0034] (3) Full-process automation and efficiency optimization:

[0035] It achieves end-to-end automated processing from data preprocessing to result output, reducing the processing time of a single simulation dataset to less than 1 minute, which is 30 times more efficient than manual processing and 5 times more efficient than existing semi-automatic solutions, and can meet the batch processing needs of large-scale simulation tasks.

[0036] (4) Significant engineering application value:

[0037] The standardized fault time information output can be directly connected to power system simulation analysis platforms and error assessment tools, providing a reliable time reference for fault impact range analysis, protection device action logic verification, and system stability optimization, reducing data preprocessing costs in engineering applications, and promoting the intelligent upgrade of power system simulation analysis. Attached Figure Description

[0038] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0039] An automatic fault time identification and analysis method for power systems based on multi-source data fusion is proposed. The method involves a three-stage multi-dimensional data preprocessing mechanism. The first stage involves detecting outliers in the multi-source power system simulation data, marking and smoothing outliers that deviate from the normal data distribution by more than three standard deviations, and using adjacent data weighted interpolation to ensure data continuity. The second stage involves aligning the time series of the multi-source power system simulation data to eliminate time deviations caused by acquisition delays from different data sources. The third stage involves filtering high-frequency noise in the multi-source power system simulation data, retaining effective signals related to fault characteristics, and improving the anti-interference capability of subsequent processing.

[0040] Step 2: Adaptive fault type identification. Based on the input multi-source parameters of the power system, key feature vectors are extracted. A pre-trained lightweight classifier is used in conjunction with typical fault scenarios to automatically match the fault type and output the feature weight coefficients corresponding to the fault type, providing a basis for subsequent adaptive threshold adjustment.

[0041] Step 3: Dynamic steady-state value calculation. A combination of sliding window weighted average and steady-state convergence judgment is used. Continuous data within a certain period before the fault is automatically extracted as the calculation window. Weights are assigned based on data reliability, and the steady-state benchmark value is calculated by weighted average. Convergence verification: If the data fluctuation within the calculation window is ≤2%, the steady-state value is confirmed to be valid; if the fluctuation exceeds the threshold, the calculation window is automatically expanded to 0.04 s and recalculated.

[0042] Step 4: Fault entry time determination. Based on the feature weight coefficients of the fault type adaptive output, an adaptive threshold + time continuity verification algorithm is used to dynamically adjust the threshold. For low-pass faults, dynamic compensation is based on high-low pass combined. When the threshold condition is met, the data change trend within 0.005 s before and after the threshold condition time point is further verified. The first moment that satisfies both the threshold condition and the data change trend is determined as the fault entry time point.

[0043] Step 5: Fault Time Correction, Bidirectional Search: Based on the initially locked time point, search forward for all candidate points within a 0.005 s range, selecting the point that meets the threshold condition and has the smoothest connection with subsequent fault stage data as the final entry time; search backward for candidate points within a 0.01 s range, selecting the point with the longest system stability duration as the final exit time; Global Reasonableness Verification: Calculate the fault duration. If the duration falls within the preset reasonable range, the correction is complete; if it exceeds the preset reasonable range, an anomaly marker is automatically triggered, and the anomaly reason is output, supporting manual review.

[0044] Step 6: Standardize the output results. The identified and corrected fault time data will be output in accordance with the industry standard format. The output content includes: fault type, fault entry time, fault exit time, fault duration, steady-state baseline value, and data reliability score.

[0045] Between steps four and five, there is a multi-condition collaborative judgment step for fault exit time. A three-dimensional collaborative verification model is constructed, including dimension 1, electrical parameter steady-state recovery verification; dimension 2, system stability verification; and dimension 3, data continuity verification. Only when all three dimensions meet the verification conditions is the moment locked as the fault exit time point.

[0046] Simultaneously, an automatic fault time identification and analysis system for power systems based on multi-source data fusion is proposed, and its components are as follows:

[0047] Multi-dimensional data preprocessing module

[0048] To address potential issues in power system simulation data, such as missing values, abnormal pulses, and time axis misalignment, a three-level preprocessing mechanism is employed:

[0049] Level 1: Based on the improved Isolation Forest algorithm, outliers in multi-source power system simulation data are detected. Outliers that deviate from the normal data distribution by more than 3 times the standard deviation are marked and smoothed (using adjacent data weighted interpolation method to repair) to ensure data continuity.

[0050] The second level: The Dynamic Time Warping (DTW) algorithm is used to align the time series of multi-source power system simulation data (such as voltage, current, power, etc.) and eliminate the time deviation caused by the acquisition delay of different data sources.

[0051] The third stage employs wavelet threshold denoising technology to filter high-frequency noise (such as fluctuations caused by electromagnetic interference) in multi-source power system simulation data, retaining effective signals related to fault characteristics and improving the anti-interference capability of subsequent algorithms.

[0052] Fault type adaptive identification module

[0053] Based on the input multi-source parameters of the power system (such as grid-connected voltage level, renewable energy type, and fault triggering conditions), a "fault type feature matching model" is constructed:

[0054] Extract key feature vectors (such as pre-fault voltage change rate, initial reactive current amplitude, and power fluctuation trend) from the input power system multi-source parameters.

[0055] By using a pre-trained lightweight classifier (optimized based on support vector machine SVM, with a training set covering more than 1,000 typical fault scenarios), the fault type (low-penetration fault, high-penetration fault, etc.) is automatically matched, and the "feature weight coefficient" corresponding to the fault type is output, providing a basis for subsequent adaptive adjustment of the threshold.

[0056] Dynamic steady-state value calculation module

[0057] Breaking through the limitations of the traditional "fixed window averaging", a combination of sliding window weighted averaging and steady-state convergence judgment is adopted:

[0058] Window selection: Automatically extracts continuous data within a certain period before the fault as the calculation window to ensure that the data within the window is in a stable operating state;

[0059] Weighted calculation: Weights are assigned based on data credibility (the weight of data in the center of the window is 1.2, and the weight of data at the edge is 0.8), and steady-state reference values ​​(such as the steady-state value of reactive current and the steady-state value of voltage) are calculated by weighted average.

[0060] Convergence verification: If the data fluctuation within the window is ≤2%, the steady-state value is confirmed to be valid; if the fluctuation exceeds the threshold, the window is automatically expanded to 0.04 s and recalculated to ensure the accuracy of the steady-state value (calculation error ≤0.5%).

[0061] Fault Entry Time Judgment Module

[0062] Based on the feature weight coefficients adaptively output by fault type, an algorithm of "adaptive threshold + time continuity verification" is adopted:

[0063] Threshold dynamic adjustment: For low-penetration faults, dynamic compensation is applied based on a combination of high and low penetration rates;

[0064] Dual verification logic: When the threshold condition is met, further verify the data change trend within 0.005 s before and after the time point (e.g., for low-voltage faults, the current must show an upward trend) to avoid misjudgment caused by instantaneous noise.

[0065] Time point locking: The first moment that meets the "threshold condition + data change trend condition" is determined as the fault entry time point, with a positioning accuracy of up to 0.0001 s.

[0066] Fault exit time multi-condition collaborative judgment module

[0067] Construct a "three-dimensional collaborative verification model" to solve the problem of misjudgment of time points caused by transient fluctuations after fault exit:

[0068] Dimension 1: Verification of steady-state recovery of electrical parameters: Check whether the voltage and current fall within the range of "steady-state value ±5%" after the fault exits, and the duration is ≥0.02 s (to ensure non-instantaneous recovery);

[0069] Dimension 2: System stability verification: By calculating the voltage fluctuation coefficient (ΔU / U≤2%) and current change rate (ΔI / Δt≤5A / s), it is determined whether the system has entered a stable operating state; ΔU is the voltage change value, U is the stable voltage value, ΔI is the current change value, and Δt is the time elapsed for the current change value;

[0070] Dimension 3: Data continuity verification: Compare the data curves 0.01 seconds before and after the fault exit time point to ensure there are no sudden changes or breaks, in order to eliminate false recovery signals caused by abnormal data acquisition;

[0071] Decision logic: Only when all three dimensions meet the verification conditions will the time be locked as the fault exit time point, effectively avoiding the problem of "transient recovery being mistaken for stable exit".

[0072] Intelligent correction module for fault time points

[0073] A correction strategy of "bidirectional search + rationality check" is adopted to further optimize the accuracy of time points:

[0074] Fault entry time correction: Based on the initial locked time point, search all candidate points within a range of 0.005s forward, and select the point that "meets the threshold condition and has the smoothest connection with the subsequent fault stage data" as the final entry time.

[0075] Fault exit time correction: Based on the initial locked time point, search for candidate points within a range of 0.01s backward, and select the point with the longest "system stability duration" as the final exit time;

[0076] Global reasonableness check: Calculate the fault duration (exit time - entry time). If the duration falls within the preset reasonable range (e.g., 0.05s-10s), the correction is complete. If it exceeds the preset reasonable range, an anomaly flag is automatically triggered, and the cause of the anomaly is output, such as "the duration is too short, which may be due to noise interference". Manual review is supported.

[0077] Standardized Result Output Module

[0078] The identified and corrected fault time data is used to construct a "fault time information structured model" according to industry standard formats (such as IEEE 1547-2018). The output includes: fault type, fault entry time (accurate to 0.001s), fault exit time (accurate to 0.001s), fault duration, steady-state baseline value, and data credibility score (0-100 points, calculated based on noise intensity and verification pass rate), providing standardized input for subsequent data analysis and error assessment.

[0079] This invention constructs a fully automated system encompassing "data preprocessing - adaptive fault type matching - steady-state feature extraction - multi-condition collaborative judgment - intelligent time point correction," enabling high-precision identification of fault times under different fault types (such as low-voltage and high-voltage) and different data quality (with and without noise). This provides a reliable time reference for power system simulation analysis while improving processing efficiency and engineering adaptability.

Claims

1. An automatic fault time identification and analysis method for power systems based on multi-source data fusion, characterized by: Step 1: Multi-dimensional data preprocessing with a three-level mechanism. Level 1: Detect outliers in multi-source power system simulation data, mark and smooth outliers that deviate from the normal data distribution by more than 3 times the standard deviation, and use adjacent data weighted interpolation to repair them to ensure data continuity. The second stage: Aligning the time series of multi-source power system simulation data to eliminate time deviations caused by acquisition delays from different data sources; The third level: Filter high-frequency noise in multi-source power system simulation data, retain effective signals related to fault characteristics, and improve the anti-interference capability of subsequent processing; Step 2: Adaptive fault type identification. Based on the input multi-source parameters of the power system, key feature vectors are extracted. A pre-trained lightweight classifier is used in conjunction with typical fault scenarios to automatically match the fault type and output the feature weight coefficients corresponding to the fault type, providing a basis for subsequent adaptive threshold adjustment. Step 3: Calculation of dynamic steady-state value. A combination of sliding window weighted average and steady-state convergence judgment is used. Continuous data within a certain period before the fault is automatically extracted as the calculation window. Weights are assigned based on data reliability, and the steady-state benchmark value is calculated by weighted average. Convergence verification: If the data fluctuation amplitude within the calculation window is ≤2%, the steady-state value is confirmed to be valid. If the fluctuation exceeds the threshold, the calculation window is automatically expanded to 0.04 s and the calculation is recalculated. Step 4: Fault entry time determination. Based on the feature weight coefficients of the fault type adaptive output, an adaptive threshold + time continuity verification algorithm is used to dynamically adjust the threshold. For low-pass faults, dynamic compensation is based on high-low pass combined. When the threshold condition is met, the data change trend within 0.005 s before and after the threshold condition time point is further verified. The first moment that satisfies both the threshold condition and the data change trend is determined as the fault entry time point. Step 5, Fault Time Correction, Two-Way Search: Based on the initially locked time point, search forward for all candidate points within a range of 0.005 s, and select the point that meets the threshold condition and has the smoothest connection with the subsequent fault stage data as the final entry time. Search backward for candidate points within a range of 0.01 s, and select the point with the longest stable duration of the system as the final exit time; Global rationality verification: Calculate the fault duration. If the duration falls within the preset reasonable range, the correction is completed; if it exceeds the preset reasonable range, an anomaly marker is automatically triggered, and the cause of the anomaly is output, supporting manual review; Step 6: Standardize the output results. The identified and corrected fault time data will be output in accordance with the industry standard format. The output content includes: fault type, fault entry time, fault exit time, fault duration, steady-state baseline value, and data reliability score.

2. The automatic identification and analysis method for power system fault time based on multi-source data fusion according to claim 1, characterized in that: A multi-condition collaborative judgment step for fault exit time is set up, and a three-dimensional collaborative verification model is constructed, including dimension 1, electrical parameter steady-state recovery verification: detecting whether the voltage and current fall within the steady-state value ±5% after fault exit, and the duration is ≥0.02s; Dimension 2, System stability verification: By calculating the voltage fluctuation coefficient ΔU / U≤2% and the current change rate ΔI / Δt≤5A / s, it is determined whether the system has entered a stable operating state; Dimension 3, Data Continuity Verification: Compare the data curves 0.01 seconds before and after the fault exit time point to ensure there are no sudden changes or breaks, in order to eliminate false recovery signals caused by abnormal data acquisition; Decision logic: When all three dimensions meet the verification conditions, that moment is locked as the fault exit time point.

3. The automatic identification and analysis method for power system fault time based on multi-source data fusion according to claim 1, characterized in that: Step 1 employs an improved isolated forest algorithm for first-level outlier detection; the second-level time series alignment uses a dynamic time warping algorithm; and the third-level high-frequency noise filtering uses a wavelet thresholding denoising algorithm.

4. The automatic identification and analysis method for power system fault time based on multi-source data fusion according to claim 1, characterized in that: Step 2: The power system's multi-source parameters include grid-connected voltage level, renewable energy type, and fault triggering conditions. The lightweight classifier is optimized based on support vector machine (SVM), and the training set contains more than 1,000 typical fault scenarios.

5. The automatic identification and analysis method for power system fault time based on multi-source data fusion according to claim 1, characterized in that: The weights assigned in step three are as follows: the weight of data at the center of the window is 1.2, and the weight of data at the edge is 0.

8.

6. The automatic identification and analysis method for power system fault time based on multi-source data fusion according to claim 1, characterized in that: Step four, when dynamically adjusting the threshold, adopts the following approach: for low-passage faults, dynamic compensation is applied based on a combination of high and low passage.

7. The automatic identification and analysis method for power system fault time based on multi-source data fusion according to claim 1, characterized in that: Step 5 sets the reasonable range to 0.05 s-10 s.

8. The automatic identification and analysis method for power system fault time based on multi-source data fusion according to claim 1, characterized in that: Step six adopts the industry standard format of IEEE 1547-2018, with fault entry time accurate to 0.001s and fault exit time accurate to 0.001s. The data reliability score is calculated by noise intensity and verification pass rate, with a score of 0-100.

9. An automatic fault time identification and analysis system for power systems based on multi-source data fusion, characterized in that: include: The three-level multi-dimensional data preprocessing module performs three levels of multi-dimensional data preprocessing: Level 1: Detects outliers in the multi-source power system simulation data, marks and smooths outliers that deviate from the normal data distribution by more than three standard deviations, and uses adjacent data weighted interpolation to repair them, ensuring data continuity; Level 2: Aligns the time series of the multi-source power system simulation data to eliminate time deviations caused by acquisition delays from different data sources; Level 3: Filters high-frequency noise in the multi-source power system simulation data, retaining effective signals related to fault characteristics and improving the anti-interference capability of subsequent processing; The fault type adaptive identification module is used to identify fault types: based on the input multi-source parameters of the power system, it extracts the key feature vectors, and automatically matches the fault type by using a pre-trained lightweight classifier in conjunction with typical fault scenarios, and outputs the feature weight coefficients corresponding to the fault type. The dynamic steady-state value calculation module is used to calculate steady-state values. It employs a combination of sliding window weighted average and steady-state convergence judgment, automatically extracting continuous data within a certain period before the fault as the calculation window, assigning weights based on data reliability, and calculating the steady-state benchmark value through weighted average. Convergence verification: if the data fluctuation within the calculation window is ≤2%, the steady-state value is confirmed to be valid; if the fluctuation exceeds the threshold, the calculation window is automatically expanded to 0.04 s and recalculated. The fault entry time judgment module is used to determine the fault entry time. Based on the feature weight coefficients of the fault type adaptive output, the adaptive threshold + time continuity verification algorithm is used to dynamically adjust the threshold. For low-pass faults, dynamic compensation is based on high-low pass combined. When the threshold condition is met, the data change trend within 0.005 s before and after the threshold condition time point is further verified. The first moment that satisfies both the threshold condition and the data change trend is determined as the fault entry time point. The fault time correction module is used to correct fault time: Bidirectional search: Based on the initially locked time point, it searches forward for all candidate points within a 0.005 s range, selecting the point that meets the threshold condition and has the smoothest transition with subsequent fault stage data as the final entry time; it also searches backward for candidate points within a 0.01 s range, selecting the point with the longest system stability duration as the final exit time; Global rationality verification: It calculates the fault duration. If the duration falls within a preset reasonable range, the correction is complete; if it exceeds the preset reasonable range, an anomaly marker is automatically triggered, and the anomaly reason is output, supporting manual review. The standardized result output module is used to output the fault time determination results: it outputs the identified and corrected fault time data in accordance with the industry standard format. The output content includes: fault type, fault entry time, fault exit time, fault duration, steady-state benchmark value, and data reliability score.

10. The automatic power system fault time identification and analysis system based on multi-source data fusion according to claim 9, characterized in that: it also... include: The fault exit time multi-condition collaborative judgment module performs verification in three dimensions: electrical parameter steady-state recovery verification, system stability verification, and data continuity verification. When all three dimensions meet the verification conditions, the moment is locked as the fault exit time point.

Images