A method and system for protecting a diesel generator from inter-turn short circuit faults based on multi-parameter features

Abstract

The application discloses a kind of diesel generator interturn short-circuit fault protection method and system based on multi-parameter feature, relating to the technical field of intelligent fault diagnosis of electric power equipment, including, feature extraction is carried out to operating state monitoring data, generates real-time feature sequence set according to time sequence integration, and real-time feature sequence set is encoded and fused with space-time feature, and generates multidimensional space-time feature map;Comprehensive matching score set is mapped into fault confidence according to linear relationship, obtains fault confidence sequence, and determines fault grade in combination with feature amplitude overrun condition, generates fault diagnosis result set;Based on fault diagnosis result set, hierarchical response is executed, and hierarchical protection action execution record is generated, and short-circuit fault protection report is generated in combination with fault diagnosis result set.The application realizes the quantitative similarity analysis of real-time feature and standard mode, accurately evaluates fault probability by multi-parameter weighted fusion mechanism, and improves the reliability and explainability of diagnosis result.

Description

Technical Field

[0001] This invention relates to the field of intelligent fault diagnosis technology for power equipment, and in particular to a method and system for protecting diesel generators from inter-turn short-circuit faults based on multi-parameter characteristics. Background Technology

[0002] As critical backup power equipment, the monitoring and protection against inter-turn short-circuit faults in the stator windings of diesel generators is a significant issue in the field of power equipment safety. Current methods primarily employ fault diagnosis based on electrical parameter analysis. These methods involve acquiring three-phase current signals using current transformers, extracting specific harmonic components using Fast Fourier Transform, and combining this with negative-sequence current detection principles for fault identification. Some solutions utilize vibration sensors to assist in diagnosis, improving identification reliability by monitoring changes in the twice-power-frequency component in the stator core vibration spectrum. These methods have formed standardized technical routes and have mature application practices in generator set protection.

[0003] Existing methods still have shortcomings in practical applications: the spatiotemporal correlation between multiple physical parameters has not been fully explored; electrical quantities, mechanical vibration and temperature parameters are usually analyzed independently, making it difficult to capture the collaborative evolution of cross-dimensional features; fixed threshold strategies lack the ability to adaptively adjust to generator operating conditions, which may affect the accuracy of judgment, especially in scenarios with load fluctuations and equipment aging. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a diesel generator inter-turn short-circuit fault protection method based on multi-parameter features to solve the problems of insufficient multi-parameter collaborative analysis and poor adaptability to operating conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for protecting diesel generator inter-turn short-circuit faults based on multi-parameter features, comprising: collecting and preprocessing operational status monitoring data; extracting features from the operational status monitoring data, integrating them into a real-time feature sequence set according to a time series, and performing spatiotemporal feature encoding and fusion on the real-time feature sequence set to generate a multi-dimensional spatiotemporal feature map; dynamically matching the multi-dimensional spatiotemporal feature map by calling a pre-stored fault feature library, calculating similarity scores, obtaining a similarity score set, and calculating a comprehensive matching score based on the similarity score set to generate a comprehensive matching score set; mapping the comprehensive matching score set to fault confidence levels according to a linear relationship, obtaining a fault confidence sequence, and determining the fault level based on the feature amplitude exceeding the limit to generate a fault diagnosis result set; executing a graded response based on the fault diagnosis result set, generating a graded protection action execution record, and generating a short-circuit fault protection report based on the fault diagnosis result set.

[0008] As a preferred embodiment of the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features described in this invention, the operating status monitoring data includes three-phase current waveforms, engine vibration signals, and winding temperature distribution data.

[0009] Preprocessing includes time alignment, normalization, band-stop / low-pass filtering, and physical quantity unit calibration.

[0010] As a preferred embodiment of the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features described in this invention, the steps for generating the real-time feature sequence set are as follows:

[0011] Based on the preprocessed operating status monitoring data, the energy entropy feature sequence of the three-phase current waveform data and the harmonic amplitude feature sequence of the body vibration signal data are extracted, and the temperature gradient feature sequence is obtained by differential calculation based on the winding temperature distribution data.

[0012] The energy entropy feature sequence, harmonic amplitude feature sequence, and temperature gradient feature sequence are integrated and recombined to generate a real-time feature sequence set.

[0013] As a preferred embodiment of the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features described in this invention, the steps for generating the multi-dimensional spatiotemporal feature map are as follows:

[0014] Spatiotemporal feature encoding is performed on the real-time feature sequence set, and sensor spatial location markers and timestamp sequence identifiers are added to form a feature encoding sequence;

[0015] Based on the feature encoding sequence, the weight distribution of different features is calculated using the attention algorithm, and then a weighted linear combination is performed to generate a unified feature vector;

[0016] Based on a unified feature vector, the weighted and fused current energy entropy feature value, vibration harmonic amplitude feature value, and temperature gradient feature value are rearranged according to the time dimension and sensor spatial dimension to generate a multidimensional spatiotemporal feature map.

[0017] As a preferred embodiment of the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features described in this invention, the steps for obtaining the similarity score set are as follows:

[0018] Three types of characteristic data of typical inter-turn short circuit modes are extracted from the pre-stored fault feature library to form a standard fault feature set;

[0019] The real-time feature sequence set in the multidimensional spatiotemporal feature map is dynamically time-warped and matched with the standard fault feature set, and the DTW distance between each real-time feature sequence and the corresponding template is calculated to generate a multi-parameter DTW distance set.

[0020] The multi-parameter DTW distance set is normalized and converted into a similarity score to obtain a similarity score set.

[0021] As a preferred embodiment of the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features described in this invention, the generation of the comprehensive matching score set refers to performing weighted fusion calculation on the similarity score set based on preset weighting coefficient rules to obtain the comprehensive matching score, and integrating the comprehensive matching scores according to the time series to generate the comprehensive matching score set.

[0022] As a preferred embodiment of the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features described in this invention, the step of obtaining the fault confidence sequence refers to converting each comprehensive matching score in the comprehensive matching score set into a fault confidence percentage through linear relationship calculation to obtain the fault confidence sequence.

[0023] As a preferred embodiment of the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features described in this invention, the steps for generating the fault diagnosis result set are as follows:

[0024] The amplitude data of each feature component in the unified feature vector is compared with the pre-stored safety threshold library, and the multiple by which the amplitude of each feature component exceeds the corresponding safety threshold is calculated to generate a set of feature amplitude over-limit multiples.

[0025] Align the fault confidence sequence with the set of feature amplitude exceeding the limit multiple by timestamp, perform logical judgment according to the preset judgment rules, and output the fault level identifier;

[0026] By integrating the fault confidence sequence, the set of feature amplitude exceeding the limit multiple, and the fault level identifier, a fault diagnosis result set is generated.

[0027] As a preferred embodiment of the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features described in this invention, the steps for generating a short-circuit fault protection report are as follows:

[0028] Based on the fault diagnosis result set, control instructions of the corresponding level are triggered according to the preset hierarchical protection strategy library to generate a set of protection action execution instructions;

[0029] According to the protection action execution instruction set, the generator is triggered to perform fault protection operation, and the execution status, execution time point and generator feedback signal are recorded in real time to generate a protection action execution record set;

[0030] The protection action execution record set and the fault diagnosis result set are correlated and aligned along the time axis to generate a short-circuit fault protection report.

[0031] Secondly, the present invention provides a diesel generator inter-turn short-circuit fault protection system based on multi-parameter features, including a data acquisition module, a feature fusion module, a dynamic matching module, a fault diagnosis module, and a graded response module;

[0032] The data acquisition module is used to collect operational status monitoring data and perform preprocessing.

[0033] The feature fusion module is used to extract features from the operation status monitoring data, integrate them according to the time series to generate a real-time feature sequence set, and perform spatiotemporal feature encoding and fusion on the real-time feature sequence set to generate a multi-dimensional spatiotemporal feature map.

[0034] The dynamic matching module is used to call the pre-stored fault feature library to dynamically match the multi-dimensional spatiotemporal feature map, calculate the similarity score, obtain the similarity score set, and calculate the comprehensive matching score based on the similarity score set to generate the comprehensive matching score set.

[0035] The fault diagnosis module is used to map the comprehensive matching score set to fault confidence according to a linear relationship, obtain the fault confidence sequence, and determine the fault level in combination with the feature amplitude exceeding the limit, and generate a fault diagnosis result set.

[0036] The graded response module is used to execute graded responses based on the fault diagnosis result set, generate graded protection action execution records, and generate short-circuit fault protection reports in combination with the fault diagnosis result set.

[0037] The beneficial effects of this invention are as follows: by extracting features from operational status monitoring data and generating multi-dimensional spatiotemporal feature maps, the electrical, vibration, and thermal parameters are fused and characterized in time and space dimensions, fully depicting the multi-physics coupling relationship in the fault evolution process, and improving the comprehensiveness and sensitivity of fault feature extraction; by calling the fault feature library for dynamic matching to generate a comprehensive matching score set, the quantitative similarity analysis between real-time features and standard patterns is realized; and the fault probability is accurately assessed through a multi-parameter weighted fusion mechanism, improving the reliability and interpretability of diagnostic results. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 The flowchart shows a diesel generator inter-turn short-circuit fault protection method based on multi-parameter characteristics.

[0040] Figure 2 This is a schematic diagram of a diesel generator inter-turn short-circuit fault protection system based on multi-parameter characteristics.

[0041] Figure 3 A flowchart for generating real-time feature sequence sets and multidimensional spatiotemporal feature maps;

[0042] Figure 4 A flowchart for generating dynamic time-warped matching and comprehensive matching scores. Detailed Implementation

[0043] 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.

[0044] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0045] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0046] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for protecting diesel generators from inter-turn short-circuit faults based on multi-parameter features, comprising the following steps:

[0047] S1. Collect and preprocess operational status monitoring data;

[0048] Operational status monitoring data includes three-phase current waveforms, machine body vibration signals, and winding temperature distribution data;

[0049] It should be noted that three-phase current waveform data refers to the instantaneous value sequence of current in the three-phase windings of the generator that changes with time, which is converted and acquired in real time by a high-precision current transformer installed in the output circuit; body vibration signal data refers to the physical quantity of periodic or non-periodic oscillations in the mechanical structure of the generator during operation due to electromagnetic force and rotating parts, which is monitored and acquired in real time by vibration acceleration sensors installed on the bearings and the base; winding temperature distribution data refers to the temperature measurement values ​​at different spatial locations of the generator stator windings, which is monitored and acquired in real time by distributed temperature sensors (such as fiber optic temperature sensors) embedded in the winding layers or ends.

[0050] Preprocessing includes time alignment, normalization, band-stop / low-pass filtering, and physical quantity unit calibration;

[0051] It should be noted that time alignment refers to using an interpolation algorithm to unify the timestamps of the operational status monitoring data, eliminating time series deviations between data from different sensors; normalization refers to converting the operational status monitoring data into a uniform numerical range, eliminating the impact of amplitude differences on the analysis; band-stop / low-pass filtering refers to using a band-stop filter to eliminate power frequency interference for the three-phase current waveform and machine vibration signal, and using a low-pass filter to suppress high-frequency noise while retaining the effective fault characteristic frequency band; physical quantity unit calibration refers to converting the three-phase current waveform into standard physical units to ensure that the data is consistent with the actual physical quantities.

[0052] S2. Extract features from the operation status monitoring data, integrate them according to the time series to generate a real-time feature sequence set, and perform spatiotemporal feature encoding and fusion on the real-time feature sequence set to generate a multi-dimensional spatiotemporal feature map.

[0053] Based on the preprocessed operating status monitoring data, the energy entropy feature sequence of the three-phase current waveform data and the harmonic amplitude feature sequence of the body vibration signal data are extracted, and the temperature gradient feature sequence is obtained by differential calculation based on the winding temperature distribution data.

[0054] Furthermore, based on the preprocessed operational status monitoring data, wavelet packet decomposition is performed on the three-phase current waveform data to decompose the three-phase current waveform data into different frequency bands. A high-frequency sub-band reflecting the inter-turn short-circuit characteristics is selected, and the ratio of the signal energy of the high-frequency sub-band to the total energy is calculated using the wavelet energy entropy calculation method. Simultaneously, the logarithm of this ratio is taken to obtain the current energy entropy characteristic value, forming a current energy entropy characteristic sequence. Fixed-length sampling is performed on the machine vibration signal data, and it is converted to a frequency domain spectrum representation using fast Fourier transform. The amplitude of specific frequency components related to the fault is extracted as the vibration harmonic amplitude characteristic value, generating a vibration harmonic amplitude characteristic sequence. Based on the winding temperature distribution data, the absolute value of the instantaneous temperature difference between each adjacent temperature measurement point is calculated using the adjacent point difference method, resulting in a temperature gradient characteristic value, generating a temperature gradient characteristic sequence.

[0055] The expression for calculating the current energy entropy of the high-frequency subband is:

[0056]

[0057] Where H is the current energy entropy value of the high-frequency subband, characterizing the degree of disorder in the energy distribution of the high-frequency subband; N is the total number of high-frequency subbands; i is the subband index; p i It is the energy percentage of the i-th subband; lnp i It is the natural logarithm of the energy percentage;

[0058] It should be noted that a specific frequency component refers to a vibration frequency component that is strongly correlated with inter-turn short-circuit faults, as determined by spectral analysis (e.g., a 100Hz double power frequency component).

[0059] Integrate the energy entropy feature sequence, harmonic amplitude feature sequence, and temperature gradient feature sequence, and align and recombine them to generate a real-time feature sequence set;

[0060] Furthermore, the current energy entropy feature sequence, vibration harmonic amplitude feature sequence, and temperature gradient feature sequence are timestamped according to a unified time base, and the feature values ​​of missing time points are filled in to generate three types of time-synchronized feature sequences. The three types of time-synchronized feature sequences are then recombined according to millisecond-level time windows, and the current energy entropy feature values, vibration harmonic amplitude feature values, and temperature gradient feature values ​​within each time window are combined into multi-dimensional data points to generate a joint feature matrix. A timestamp index and sensor spatial location identifier are added to each multi-dimensional data point in the joint feature matrix to form a real-time feature sequence set with complete spatiotemporal markings.

[0061] Spatiotemporal feature encoding is performed on the real-time feature sequence set, and sensor spatial location markers and timestamp sequence identifiers are added to form a feature encoding sequence;

[0062] Furthermore, current energy entropy feature values, vibration harmonic amplitude feature values, and temperature gradient feature values ​​are extracted from the real-time feature sequence set in chronological order for each sampling point to form three-dimensional data points. The physical installation coordinates of the sensors embedded in the generator stator windings are obtained through generator design drawings and installation records. These sensor coordinates are normalized to generate standardized spatial coordinate vectors, which are then bound as metadata to the corresponding sensor feature data points, generating spatial coordinate codes containing sensor installation location markers. The spatial coordinate codes of each three-dimensional data point are bound to their corresponding feature values ​​to form spatially marked feature data points. A high-precision timestamp sequence identifier is added to each spatially marked feature data point to ensure strict synchronization between the time dimension and the original data acquisition timeline. All spatiotemporally marked feature data points are reorganized in chronological order to generate a feature coding sequence with complete spatiotemporal coding information.

[0063] Based on the feature encoding sequence, the weight distribution of different features is calculated using the attention algorithm, and then a weighted linear combination is performed to generate a unified feature vector;

[0064] Furthermore, spatiotemporally labeled current energy entropy feature values, vibration harmonic amplitude feature values, and temperature gradient feature values ​​are extracted from the feature encoding sequence to generate a feature dataset to be fused. An attention algorithm is used to calculate the weight values ​​of the current energy entropy feature, vibration harmonic amplitude feature, and temperature gradient feature, respectively, generating a feature weight distribution set. Based on the weight values ​​in the feature weight distribution set, the current energy entropy feature values, vibration harmonic amplitude feature values, and temperature gradient feature values ​​are weighted and calculated to generate a weighted feature value set. The feature values ​​in the weighted feature value set are then fused according to their weight ratios and weighted summation is performed to generate a unified feature vector.

[0065] Based on the unified feature vector, the weighted fusion of current energy entropy feature value, vibration harmonic amplitude feature value and temperature gradient feature value are rearranged according to the time dimension and sensor space dimension to generate a multi-dimensional spatiotemporal feature map.

[0066] Furthermore, based on a unified feature vector, all weighted current energy entropy feature values, vibration harmonic amplitude feature values, and temperature gradient feature values ​​are extracted to generate a feature data set to be reconstructed. The feature data set to be reconstructed is arranged in the order of the original acquisition timestamps to form a time series dataset distributed along the time dimension. Each feature value in the time series dataset is redistributed in the spatial dimension according to the corresponding sensor spatial location marker to form a three-dimensional data structure that simultaneously contains the time dimension, spatial dimension, and feature dimension. The three-dimensional data structure is then transformed into a graph: where the X-axis represents the time series, the Y-axis represents the sensor spatial location marker, and the Z-axis represents each weighted feature value, generating a multi-dimensional spatiotemporal feature graph.

[0067] S3. Call the pre-stored fault feature library to dynamically match the multi-dimensional spatiotemporal feature map, calculate the similarity score, obtain the similarity score set, and calculate the comprehensive matching score based on the similarity score set to generate the comprehensive matching score set.

[0068] Three types of characteristic data of typical inter-turn short circuit modes are extracted from the pre-stored fault feature library to form a standard fault feature set;

[0069] Furthermore, based on the pre-stored fault feature database, the data storage area of ​​typical inter-turn short circuit modes is retrieved according to the fault type identifier to obtain the original feature data set. From the original feature data set, three types of feature data, namely current energy entropy feature value, vibration harmonic amplitude feature value, and temperature gradient feature value, are selected to generate a preliminary classification feature dataset. The preliminary classification feature dataset is then standardized in terms of data format, unifying the timestamp format and numerical range to generate a standardized feature dataset. The three types of feature data, namely current energy entropy feature value, vibration harmonic amplitude feature value, and temperature gradient feature value, in the standardized feature dataset are classified and organized according to the fault severity to form the final standard fault feature set.

[0070] It should be noted that the pre-stored fault feature database is a standardized feature library built from historical fault case data and digital twin simulation data accumulated over a long period of time. It contains three types of standardized time-series data sets that label the fault location, severity, and timestamp: current energy entropy feature value, vibration harmonic amplitude feature value, and temperature gradient feature value. The fault type identifier is obtained by analyzing the feature patterns of historical fault data and combining them with classification codes predefined in accordance with international EE requirements.

[0071] The real-time feature sequence set in the multidimensional spatiotemporal feature map is dynamically time-warped and matched with the standard fault feature set, and the DTW distance between each real-time feature sequence and the corresponding template is calculated to generate a multi-parameter DTW distance set.

[0072] Furthermore, the corresponding current energy entropy feature template, vibration harmonic amplitude feature template, and temperature gradient feature template are obtained from the standard fault feature set to generate a standard feature template set. The current energy entropy feature sequence, vibration harmonic amplitude feature sequence, and temperature gradient feature sequence in the real-time feature sequence set are dynamically time-warped and matched with the corresponding current energy entropy feature template, vibration harmonic amplitude feature template, and temperature gradient feature template in the standard feature template set, respectively. The DTW (minimum path) distance between each real-time feature sequence and the corresponding template is calculated using the DTW algorithm, and the distances are recorded and integrated to generate a multi-parameter DTW distance set.

[0073] The expression for calculating the DTW distance is:

[0074]

[0075] Where D(X,Y) is the DTW distance between the real-time feature sequence X (such as the current energy entropy feature sequence) and the corresponding standard feature template Y (such as the current energy entropy feature template); d(x h ,y j ) is the real-time data point x h With template data point y j Euclidean distance between them; x h It is the specific value of the real-time feature sequence X at the h-th position; y j π is the specific value of the standard feature template Y at the j-th position; π is the alignment path, which is the mapping set of points connecting the real-time feature sequence X and the corresponding standard feature template Y; h is the index of the real-time feature sequence X; j is the index of the corresponding standard feature template Y;

[0076] The multi-parameter DTW distance set is normalized and converted into a similarity score to obtain a similarity score set;

[0077] Furthermore, DTW distance values ​​corresponding to the current energy entropy feature sequence, vibration harmonic amplitude feature sequence, and temperature gradient feature sequence are extracted from the multi-parameter DTW distance set to form the original distance dataset. The original distance dataset is then normalized to its maximum value, and normalized distance values ​​are obtained based on the ratio of each DTW distance value to the preset maximum reference distance. The normalized distance values ​​are linearly mapped to a percentage range using a scaling conversion method, converting them into similarity scores and generating a standardized similarity score set. The current energy entropy feature similarity scores, vibration harmonic amplitude feature similarity scores, and temperature gradient feature similarity scores are extracted from the standardized similarity score set and categorized and integrated according to feature type to form a similarity score set containing timestamps and feature identifiers.

[0078] It should be noted that the preset maximum reference distance is the upper limit of the DTW distance obtained by statistically calculating the characteristic data of the most severe inter-turn short circuit state in the historical fault case database.

[0079] Based on the preset weighting coefficient rules, the similarity score set is weighted and fused to obtain the comprehensive matching score, and the comprehensive matching score is integrated according to the time series to generate a comprehensive matching score set;

[0080] Furthermore, similarity scores for current energy entropy, vibration harmonic amplitude, and temperature gradient are extracted from the similarity score set to generate a feature score set to be weighted. Based on a preset weighting coefficient rule, the similarity scores of each feature in the feature score set to be weighted are weighted to generate a weighted feature score set. The weighted scores of each feature in the weighted feature score set are summed to obtain a comprehensive matching score. The comprehensive matching scores calculated within a continuous time window are arranged and integrated according to timestamp order, and corresponding timestamp identifiers are added to generate a comprehensive matching score set.

[0081] It should be noted that the preset weighting coefficient rule is a weighting scheme determined by analyzing historical fault data and evaluating the parameter importance classification requirements in the generator industry. It includes specific values ​​for the current energy entropy characteristic weighting coefficient, the vibration harmonic amplitude characteristic weighting coefficient, and the temperature gradient characteristic weighting coefficient.

[0082] S4. Map the comprehensive matching score set to fault confidence according to a linear relationship, obtain the fault confidence sequence, and determine the fault level in combination with the feature amplitude exceeding the limit, and generate a fault diagnosis result set.

[0083] By calculating the linear relationship, each comprehensive matching score in the comprehensive matching score set is converted into a fault confidence percentage, thus obtaining a fault confidence sequence;

[0084] Furthermore, the comprehensive matching score corresponding to each timestamp is extracted from the comprehensive matching score set to generate a score dataset to be transformed. Then, each comprehensive matching score in the score dataset to be transformed is mapped to a fault confidence percentage interval through a linear transformation relationship to obtain the corresponding confidence percentage for each score. Simultaneously, the corresponding confidence percentages for each score are integrated to generate preliminary confidence data. The preliminary confidence data is then smoothed and filtered to eliminate random fluctuations, generating stable fault confidence percentage data. This stable fault confidence percentage data is then reordered according to the original timestamps, and generator numbers, sensor spatial location markers, and feature type markers are appended to generate a fault confidence sequence.

[0085] The amplitude data of each feature component in the unified feature vector is compared with the pre-stored safety threshold library, and the multiple by which the amplitude of each feature component exceeds the corresponding safety threshold is calculated to generate a set of feature amplitude over-limit multiples.

[0086] Furthermore, current energy entropy characteristic component amplitude data, vibration harmonic amplitude characteristic component amplitude data, and temperature gradient characteristic component amplitude data are extracted from the unified feature vector to generate a feature amplitude dataset to be compared. The corresponding current energy entropy safety threshold, vibration harmonic amplitude safety threshold, and temperature gradient safety threshold are retrieved from the pre-stored safety threshold library to generate a safety threshold set. The current energy entropy characteristic component amplitude data, vibration harmonic amplitude characteristic component amplitude data, and temperature gradient characteristic component amplitude data in the feature amplitude dataset to be compared are compared point-by-point with the corresponding current energy entropy safety threshold, vibration harmonic amplitude safety threshold, and temperature gradient safety threshold in the safety threshold set. The specific multiple by which the amplitude of each characteristic component exceeds the corresponding safety threshold is calculated to generate preliminary exceedance multiple data. The preliminary exceedance multiple data are classified and integrated according to feature type, and timestamp identifiers and feature type tags are added to generate a standardized feature amplitude exceedance multiple set.

[0087] It should be noted that the pre-stored safety threshold library is a set of boundary values ​​determined by analyzing historical normal operation data of generators, simulation experimental data and industry safety regulations. It includes the safety threshold for current energy entropy, the safety threshold for vibration harmonic amplitude and the safety threshold for temperature gradient.

[0088] It should be noted that the current energy entropy safety threshold is determined by statistically analyzing the fluctuation range of the high-frequency energy entropy of the current during normal operation, taking the mean plus three times the standard deviation as the upper limit of the threshold, with an exemplary range of 0.8 to 1.2; the vibration harmonic amplitude safety threshold is the vibration limit of the diesel generator, set in conjunction with the rated speed of the unit, with an exemplary range of 2.5 to 4.0 mm / s; the temperature gradient safety threshold is a critical value determined based on the heat resistance grade of the generator insulation material and the efficiency of the cooling system, usually set as the maximum allowable temperature difference between adjacent temperature measurement points, with an exemplary range of 8 to 15℃ / min;

[0089] Align the fault confidence sequence with the set of feature amplitude exceeding the limit multiple by timestamp, perform logical judgment according to the preset judgment rules, and output the fault level identifier;

[0090] Furthermore, fault confidence data and feature amplitude exceeding multiple data at the same timestamp are extracted from the fault confidence sequence and feature amplitude exceeding multiple set to generate a time-aligned comparison dataset. A preset fault level determination rule base is invoked to perform logical judgment on the time-aligned comparison dataset: when the fault confidence is greater than the emergency fault confidence threshold, and the current energy entropy exceeding multiple exceeds the current energy entropy safety threshold, or the temperature gradient exceeding multiple exceeds the temperature gradient safety threshold, it is determined to be an L3 emergency fault; when the fault confidence is greater than the alarm fault confidence threshold but less than the emergency fault confidence threshold, and the vibration harmonic amplitude exceeding multiple... When the vibration harmonic amplitude exceeds the safe threshold, it is determined to be an L2 alarm fault; when the fault confidence is greater than the warning fault confidence threshold but less than the alarm fault confidence threshold, and any (current / vibration / temperature) feature exceeds the corresponding safe threshold by multiple, it is determined to be an L1 warning fault; when the fault confidence is less than the alarm fault confidence threshold and the emergency fault confidence threshold, and all feature exceedance multiples do not exceed the corresponding safe thresholds, it is determined to be an L0 normal state, and a preliminary fault level identifier is generated; the preliminary fault level identifier is checked for timing consistency, and a timestamp is added to it, and the final fault level identifier is output;

[0091] It should be noted that the preset fault level judgment rule base is a standardized set of judgment rules developed by comprehensively analyzing historical fault data, industry regulations, and expert experience. It includes confidence thresholds for each level of fault, safety thresholds for each characteristic quantity, and multi-parameter combination logic conditions. The emergency fault confidence threshold (95%) is set based on the statistical confidence lower limit of historical serious fault cases to ensure that the highest level of protection is triggered only when the fault characteristics are highly significant. The alarm fault confidence threshold (80%) is set based on the reliability statistics of early fault signals. It is lower than the emergency fault confidence threshold but can effectively identify potential faults. The early warning fault confidence threshold (60%) is set by statistically analyzing the confidence distribution pattern of early fault signals and combining it with experts' experience assessment of the reliability of minor fault identification. It is used to capture potential fault trends at the earliest stage.

[0092] It should be noted that the current energy entropy safety threshold is determined by analyzing the fluctuation range of the high-frequency energy entropy of the current during normal operation, using a statistical method of mean plus three times the standard deviation, and is used to capture abnormal current distortion. An exemplary value range is 0.8 to 1.2. The temperature gradient safety threshold is set according to the maximum allowable temperature rise rate of the stator winding insulation material, and the critical value is determined by analyzing the historical temperature difference data of adjacent temperature measurement points in the winding thermal field distribution. An exemplary value range is 8 to 15℃ / min. The vibration harmonic amplitude safety threshold is set based on the unit's structural resonance characteristics and the statistical distribution of vibration data during historical normal operation. The safety boundary is determined by monitoring the vibration amplitude variation range of the bearings and stator frame. An exemplary value range is 2.5 to 4.0 mm / s.

[0093] Integrate the fault confidence sequence, the set of feature amplitude exceeding the limit multiple, and the fault level identifier to generate a fault diagnosis result set;

[0094] Furthermore, timestamped fault confidence data is extracted from the fault confidence sequence, data on the multiples of each feature exceeding the limit are extracted from the feature amplitude exceeding the limit set, and fault level data is extracted from the fault level identifier to generate the original data set to be integrated. The fault confidence data, feature exceeding the limit multiple data, and fault level data in the original data set are precisely aligned according to the timestamp, and the correlation of data at the same time is ensured by the time series matching algorithm. Equipment number and sensor spatial location markers are added to generate a fault diagnosis result set.

[0095] S5. Execute graded responses based on the fault diagnosis result set, generate graded protection action execution records, and generate short-circuit fault protection reports in combination with the fault diagnosis result set.

[0096] Based on the fault diagnosis result set, control instructions of the corresponding level are triggered according to the preset hierarchical protection strategy library to generate a set of protection action execution instructions;

[0097] Furthermore, fault level identifiers are extracted from the fault diagnosis results set, and a preset hierarchical protection strategy library is invoked. The corresponding control instruction template is matched according to the fault level identifier: an emergency trip instruction is generated when the fault level identifier is L3, a load reduction alarm instruction is generated when the fault level identifier is L2, an early warning monitoring instruction is generated when the fault level identifier is L1, and a normal operation monitoring instruction is generated when the fault level identifier is L0. The emergency trip instruction, load reduction alarm instruction, early warning monitoring instruction, and normal operation monitoring instruction are integrated to output a set of protection action execution instructions.

[0098] It should be noted that the preset graded protection strategy library is a standardized set of strategies developed by analyzing historical fault data, industry safety regulations and expert experience. It includes control instruction templates corresponding to four fault levels (L0 normal state / L1 early warning fault / L2 alarm fault / L3 emergency fault), execution parameter thresholds for each level and action execution effect verification logic.

[0099] According to the protection action execution instruction set, the generator is triggered to perform fault protection operation, and the execution status, execution time point and generator feedback signal are recorded in real time to generate a protection action execution record set;

[0100] Furthermore, the control command content in the protection action execution command set is analyzed, the command type (emergency trip / load reduction alarm / early warning monitoring / normal operation monitoring) and associated target parameters are identified, and an executable command queue is generated. This executable command queue is then sent to the generator control interface: when the command type is emergency trip, the circuit breaker is triggered to perform a tripping operation and the tripping time is recorded; when the command type is load reduction alarm, the load regulator is driven to reduce the output power to the target value and the actual load value is recorded; when the command type is early warning monitoring, the audible and visual alarm is activated and the early warning trigger time is recorded; when the command type is normal operation monitoring, the generator is maintained under normal operating conditions. Before operation, a preliminary execution status record is generated, including operation type, execution time, target parameters, and actual parameters. The opening and closing status is monitored through the circuit breaker auxiliary contacts, the real-time power value is read through the power sensor, and the alarm trigger status is obtained through the alarm feedback circuit. These are then integrated to form a generator feedback signal. The generator feedback signal is compared item by item with the expected effect of the command (e.g., the contacts should be open after the circuit breaker is opened, and the power should be equal to the target value after load reduction) to generate execution effect verification data. The preliminary execution status record and execution effect verification data are integrated, and a precise execution timestamp and equipment response parameters are added to generate a protection action execution record set.

[0101] It should be noted that the target parameters are specific execution indicators extracted from the fault diagnosis results set, including load reduction target value and early warning fault confidence threshold, which are obtained directly by parsing fault level identifier and feature over-limit multiple data;

[0102] The protection action execution record set and the fault diagnosis result set are correlated and aligned along the time axis to generate a short-circuit fault protection report;

[0103] Furthermore, data records with the same timestamp in the protection action execution record set and the fault diagnosis result set are associated and aligned along the time axis using a time series matching algorithm to generate an associated data set. Device numbers are added to the associated data set, and the data is converted into a standard report format using a standardized template to generate a short-circuit fault protection report containing the fault evolution process, protection action execution effect and characteristic data.

[0104] It should be noted that the standardized template is a unified reporting framework developed through structured design by integrating the report format requirements in industry regulations with the structure of historical fault data records.

[0105] This embodiment also provides a diesel generator inter-turn short-circuit fault protection system based on multi-parameter features, including: a data acquisition module, a feature fusion module, a dynamic matching module, a fault diagnosis module, and a graded response module;

[0106] The data acquisition module is used to collect operational status monitoring data and perform preprocessing.

[0107] The feature fusion module is used to extract features from the operation status monitoring data, integrate them according to the time series to generate a real-time feature sequence set, and perform spatiotemporal feature encoding and fusion on the real-time feature sequence set to generate a multi-dimensional spatiotemporal feature map.

[0108] The dynamic matching module is used to call the pre-stored fault feature library to dynamically match the multi-dimensional spatiotemporal feature map, calculate the similarity score, obtain the similarity score set, and calculate the comprehensive matching score based on the similarity score set to generate the comprehensive matching score set.

[0109] The fault diagnosis module is used to map the comprehensive matching score set to fault confidence according to a linear relationship, obtain the fault confidence sequence, and determine the fault level in combination with the feature amplitude exceeding the limit, and generate a fault diagnosis result set.

[0110] The graded response module is used to execute graded responses based on the fault diagnosis result set, generate graded protection action execution records, and generate short-circuit fault protection reports in combination with the fault diagnosis result set.

[0111] This embodiment also provides a computer device applicable to the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the diesel generator inter-turn short-circuit fault protection method based on multi-parameter features proposed in the above embodiment.

[0112] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0113] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for protecting diesel generator inter-turn short-circuit faults based on multi-parameter features as proposed in the above embodiments. The storage medium can be 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.

[0114] In summary, this invention achieves a fusion representation of electrical, vibration, and thermal parameters in both time and space dimensions by extracting features from operational status monitoring data and generating a multi-dimensional spatiotemporal feature map. This comprehensively depicts the multi-physics coupling relationship during fault evolution, enhancing the comprehensiveness and sensitivity of fault feature extraction. Furthermore, by dynamically matching fault feature libraries to generate a comprehensive matching score set, it enables quantitative similarity analysis between real-time features and standard patterns. Finally, through a multi-parameter weighted fusion mechanism, it accurately assesses fault probability, improving the reliability and interpretability of diagnostic results.

[0115] 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 can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for protecting diesel generators from inter-turn short-circuit faults based on multi-parameter characteristics, characterized in that: include, Collect and preprocess operational status monitoring data, including: Operational status monitoring data includes three-phase current waveforms, machine body vibration signals, and winding temperature distribution data; Preprocessing includes time alignment, normalization, band-stop / low-pass filtering, and physical quantity unit calibration; Feature extraction is performed on the operational status monitoring data, and the data is integrated according to time series to generate a real-time feature sequence set. This real-time feature sequence set is then spatiotemporally encoded and fused to generate a multi-dimensional spatiotemporal feature map, including: Based on the preprocessed operating status monitoring data, the energy entropy feature sequence of the three-phase current waveform data and the harmonic amplitude feature sequence of the body vibration signal data are extracted, and the temperature gradient feature sequence is obtained by differential calculation based on the winding temperature distribution data. Integrate the energy entropy feature sequence, harmonic amplitude feature sequence, and temperature gradient feature sequence, and align and recombine them to generate a real-time feature sequence set; Spatiotemporal feature encoding is performed on the real-time feature sequence set, and sensor spatial location markers and timestamp sequence identifiers are added to form a feature encoding sequence; Based on the feature encoding sequence, the weight distribution of different features is calculated using the attention algorithm, and then a weighted linear combination is performed to generate a unified feature vector; Based on the unified feature vector, the weighted fusion of current energy entropy feature value, vibration harmonic amplitude feature value and temperature gradient feature value are rearranged according to the time dimension and sensor space dimension to generate a multi-dimensional spatiotemporal feature map. The system calls a pre-stored fault feature library to dynamically match the multidimensional spatiotemporal feature map, calculates the similarity score, obtains a set of similarity scores, and calculates the comprehensive matching score based on the set of similarity scores to generate a comprehensive matching score set. The comprehensive matching score set is mapped to fault confidence scores according to a linear relationship, a fault confidence score sequence is obtained, and the fault level is determined by combining the feature amplitude exceeding the limit, generating a fault diagnosis result set, including: The amplitude data of each feature component in the unified feature vector is compared with the pre-stored safety threshold library, and the multiple by which the amplitude of each feature component exceeds the corresponding safety threshold is calculated to generate a set of feature amplitude over-limit multiples. Align the fault confidence sequence with the set of feature amplitude exceeding the limit multiple by timestamp, perform logical judgment according to the preset judgment rules, and output the fault level identifier; Integrate the fault confidence sequence, the set of feature amplitude exceeding the limit multiple, and the fault level identifier to generate a fault diagnosis result set; Based on the fault diagnosis result set, a graded response is executed, a graded protection action execution record is generated, and a short-circuit fault protection report is generated by combining the fault diagnosis result set.

2. The diesel generator inter-turn short-circuit fault protection method based on multi-parameter characteristics according to claim 1, characterized in that: The steps for obtaining the similarity score set are as follows: Three types of characteristic data of typical inter-turn short circuit modes are extracted from the pre-stored fault feature library to form a standard fault feature set; The real-time feature sequence set in the multidimensional spatiotemporal feature map is dynamically time-warped and matched with the standard fault feature set, and the DTW distance between each real-time feature sequence and the corresponding template is calculated to generate a multi-parameter DTW distance set. The multi-parameter DTW distance set is normalized and converted into a similarity score to obtain a similarity score set.

3. The diesel generator inter-turn short-circuit fault protection method based on multi-parameter characteristics according to claim 1, characterized in that: The process of generating a comprehensive matching score set refers to performing a weighted fusion calculation on the similarity score set based on a preset weighting coefficient rule to obtain a comprehensive matching score, and then integrating the comprehensive matching scores according to a time series to generate a comprehensive matching score set.

4. The diesel generator inter-turn short-circuit fault protection method based on multi-parameter characteristics according to claim 1, characterized in that: The process of obtaining the fault confidence sequence refers to converting each comprehensive matching score in the comprehensive matching score set into a fault confidence percentage through linear relationship calculation, thereby obtaining the fault confidence sequence.

5. The diesel generator inter-turn short-circuit fault protection method based on multi-parameter characteristics according to claim 1, characterized in that: The steps for generating a short-circuit fault protection report are as follows: Based on the fault diagnosis result set, control instructions of the corresponding level are triggered according to the preset hierarchical protection strategy library to generate a set of protection action execution instructions; According to the protection action execution instruction set, the generator is triggered to perform fault protection operation, and the execution status, execution time point and generator feedback signal are recorded in real time to generate a protection action execution record set; The protection action execution record set and the fault diagnosis result set are correlated and aligned along the time axis to generate a short-circuit fault protection report.

6. A diesel generator inter-turn short-circuit fault protection system based on multi-parameter characteristics, based on the diesel generator inter-turn short-circuit fault protection method based on multi-parameter characteristics according to any one of claims 1 to 5, characterized in that: It includes a data acquisition module, a feature fusion module, a dynamic matching module, a fault diagnosis module, and a graded response module; The data acquisition module is used to collect operational status monitoring data and perform preprocessing. The feature fusion module is used to extract features from the operation status monitoring data, integrate them according to the time series to generate a real-time feature sequence set, and perform spatiotemporal feature encoding and fusion on the real-time feature sequence set to generate a multi-dimensional spatiotemporal feature map. The dynamic matching module is used to call the pre-stored fault feature library to dynamically match the multi-dimensional spatiotemporal feature map, calculate the similarity score, obtain the similarity score set, and calculate the comprehensive matching score based on the similarity score set to generate the comprehensive matching score set. The fault diagnosis module is used to map the comprehensive matching score set to fault confidence according to a linear relationship, obtain the fault confidence sequence, and determine the fault level in combination with the feature amplitude exceeding the limit, and generate a fault diagnosis result set. The graded response module is used to execute graded responses based on the fault diagnosis result set, generate graded protection action execution records, and generate short-circuit fault protection reports in combination with the fault diagnosis result set.

Images