Construction of power grid dispatching-oriented professional knowledge base and intelligent query method

Abstract

The application provides a power grid dispatch-oriented professional knowledge base construction and intelligent query method, relates to the technical field of data processing, and comprises the following steps: performing coordinate processing on a fault recording wave chart uploaded by a dispatch end, and identifying phase waveform profiles; dividing the phase waveform profiles into positive and negative half cycles based on the zero-value baselines corresponding to the phase waveform profiles, to obtain positive and negative half cycle division results; extracting matching parameters of each phase based on the positive and negative half cycle division results, matching the matching parameters with standard waveform parameter templates corresponding to each fault type in a knowledge base, and obtaining a target fault type; and calling corresponding knowledge units from the knowledge base according to the target fault type, and arranging the knowledge units into standardized answers and outputting the standardized answers to the dispatch end. The application improves the disposal decision efficiency of dispatch personnel when the fault type cannot be accurately described and the availability of knowledge base assisted decision.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for constructing a professional knowledge base and intelligent querying for power grid dispatching. Background Technology

[0002] Power grid dispatching is a core component in ensuring the safe and stable operation of the power system. Dispatchers need to make accurate handling decisions based on operating procedures within a short period of time after a fault occurs. To improve knowledge acquisition efficiency, existing technologies construct knowledge base systems to perform structured modeling and semantic vectorization of procedure texts, enabling dispatchers to query handling steps using natural language queries. However, all of the above solutions are based on the premise that "users can accurately describe the type of fault encountered in words," meaning the system assumes that the dispatcher has already completed the fault assessment and can convert it into a valid textual expression before triggering the semantic retrieval process.

[0003] However, in actual dispatching scenarios, after the system triggers fault waveform recording, dispatchers are first faced with voltage and current waveform images on the monitoring screen. For inexperienced dispatchers, the waveforms of fault types such as single-phase grounding, two-phase short circuits, and transformer inrush currents are visually similar, making judgment inherently difficult. If this leads to inaccurate textual descriptions, the knowledge base will return handling steps that do not match the actual fault based on incorrect semantics; if the fault type cannot be determined at all, dispatchers have to abandon the search and instead consult by phone, causing the knowledge base's decision-making support function to fail in the most critical moment when it is most needed.

[0004] Therefore, how to enable the knowledge base system to directly receive fault waveform images as query input and automatically complete fault identification and knowledge matching based on waveform geometric features and multi-channel parameters has become an urgent problem to be solved. Summary of the Invention

[0005] This application provides a professional knowledge base construction and intelligent query method for power grid dispatching, in order to solve the problems of existing knowledge base systems that use text semantic retrieval as the only entry point, cannot directly process fault waveform images, and cause the knowledge base's decision-making assistance function to fail when dispatchers cannot accurately describe the fault type.

[0006] The professional knowledge base construction and intelligent query method for power grid dispatching provided in this application includes: performing coordinate processing on the fault waveform diagram uploaded by the dispatching terminal to identify the waveform contours of each phase; The waveform profile of each phase is divided into positive and negative half cycles using the zero-value baseline corresponding to the waveform profile of each phase, and the positive and negative half cycle division results are obtained. Based on the positive and negative half-cycle segmentation results, the matching parameters of each phase are extracted, and the matching parameters are matched with the standard waveform parameter templates corresponding to each fault type in the knowledge base to obtain the target fault type. Based on the target fault type, the corresponding knowledge unit is retrieved from the knowledge base, organized into a standardized answer, and output to the scheduling end.

[0007] Optionally, in one possible implementation, the step of performing coordinate processing on the fault waveform uploaded by the scheduling terminal to identify the waveform contours of each phase includes: Coordinate processing was performed with the lower left corner of the fault waveform as the origin to determine the time coordinates and amplitude coordinates of each pixel. For each phase channel, the number of pixels corresponding to each amplitude coordinate is counted, and the amplitude coordinates in each phase channel where the number of pixels exceeds the preset density threshold are determined as the zero value baseline of each phase. The sequence of pixels that continuously deviate from the zero baseline in each phase channel is identified as the waveform profile of each phase, resulting in the waveform profile of phase A, phase B, and phase C.

[0008] Optionally, in one possible implementation, the step of counting the number of pixels corresponding to each amplitude coordinate for each phase channel, and determining the amplitude coordinates where the number of pixels in each phase channel exceeds a preset density threshold as the zero-value baseline for each phase, includes: The number of pixels corresponding to each amplitude coordinate in each phase channel is counted to obtain the pixel distribution of each phase amplitude. When it is determined that there are two amplitude coordinates in the amplitude pixel distribution of each phase where the number of pixels both exceed the preset dense threshold, the middle amplitude coordinate of the two amplitude coordinates is taken as the zero value baseline of each phase. When it is determined that there is only one amplitude coordinate in the amplitude pixel distribution of each phase where the number of pixels exceeds a preset dense threshold, the amplitude coordinate is used as the zero value baseline of each phase.

[0009] Optionally, in one possible implementation, extracting the matching parameters of each phase based on the positive and negative half-cycle segmentation results includes: Obtain all intersection points between the waveform profile of each phase and the zero baseline. Take the waveform profile segment above the zero baseline between two adjacent intersection points as the positive half-cycle profile and the waveform profile segment below the zero baseline as the negative half-cycle profile. When the waveform profile segment between two adjacent intersection points coincides with the zero baseline, the coincident segment is identified as a discontinuity segment. The time coordinate length of all discontinuities in the waveform profile of each phase is counted to obtain the discontinuity duration of each phase. Based on the positive half-cycle area enclosed by the positive half-cycle profile and the zero-value baseline, and the negative half-cycle area enclosed by the negative half-cycle profile and the zero-value baseline, combined with the theoretical area of ​​the whole cycle of each phase waveform profile, the normalized half-cycle area difference of each phase is obtained, where the theoretical area of ​​the whole cycle is determined by the time coordinate span and amplitude coordinate peak value of each phase waveform profile. The time coordinates of the first positive intersection point of the waveform profiles of phases A, B, and C with the zero baseline are subtracted pairwise to obtain the phase difference AB, phase difference BC, and phase difference AC. The phase difference AB, phase difference BC, and phase difference AC are then compared with the standard phase differences pre-stored in the knowledge base to obtain the phase difference sequence. The normalized half-cycle area difference of each phase, the discontinuity duration of each phase, and the phase difference sequence are used as matching parameters.

[0010] Optionally, in one possible implementation, the step of calculating the time coordinate lengths of all discontinuities in the waveform profiles of each phase to obtain the discontinuity duration of each phase includes: The number of intermediate segments in the waveform profile of each phase is counted to obtain the number of segments between each phase; When the number of discontinuities in any phase is greater than one, the time coordinate lengths of all discontinuities in each phase are summed to obtain the discontinuity duration of each phase. When the number of discontinuities in any phase is equal to one, the time coordinate length of the unique discontinuity in each phase is taken as the discontinuity duration of each phase.

[0011] Optionally, in one possible implementation, after counting the number of intermediate segments in the waveform profile of each phase to obtain the number of inter-phase segments, the process includes: By counting the number of discontinuous segments greater than zero, we obtain the set of discontinuous phases; When the set of discontinuous phases is determined to be empty, it is determined that there are no discontinuous segments in the waveform profile of each phase, and the normalized half-circle area difference and phase difference sequence of each phase are used as the matching basis. When determining that the discontinuous phase set is not empty, the discontinuity duration of each phase in the discontinuous phase set, the normalized half-circuit area difference of each phase, and the phase difference sequence are used together as the matching criteria.

[0012] Optionally, in one possible implementation, matching the matching parameters with standard waveform parameter templates corresponding to each fault type in the knowledge base to obtain the target fault type includes: Based on the normalized half-cycle area difference of each phase and the corresponding parameters in each standard waveform parameter template, the area deviation set of each fault type is obtained; based on the phase difference sequence and the corresponding parameters in each standard waveform parameter template, the phase deviation set of each fault type is obtained. When it is determined that the set of discontinuous phases is not empty, the discontinuity duration of each phase in the set of discontinuous phases and the corresponding discontinuity duration of each phase in each standard waveform parameter template are used to construct a set of discontinuity deviations. The set of area deviations, the set of phase deviations and the set of discontinuity deviations are merged to obtain the comprehensive deviation set of each fault type. When determining that the set of discontinuous phases is empty, the area deviation set and the phase deviation set are combined as the comprehensive deviation set for each fault type. The fault type with the smallest total set of comprehensive deviations is identified as the target fault type.

[0013] Optionally, in one possible implementation, determining the fault type with the smallest total aggregate deviation set as the target fault type includes: When the minimum value of the total set of comprehensive deviations is less than or equal to the preset deviation threshold, the corresponding fault type is determined as the target fault type. When the minimum value of the total set of comprehensive deviations is greater than the preset deviation threshold, the normalized half-cycle area difference, discontinuity duration and phase difference sequence of each phase are marked as parameters to be confirmed and sent to the scheduling expert terminal. The fault type confirmation result returned by the scheduling expert terminal is received and the fault type confirmation result is used as the target fault type.

[0014] Optionally, in one possible implementation, after receiving the fault type confirmation result returned by the scheduling expert and using the fault type confirmation result as the target fault type, the method further includes: When it is determined that a standard waveform parameter template corresponding to the fault type confirmation result exists in the knowledge base, the parameter to be confirmed is included in the historical waveform feature parameter set corresponding to the standard waveform parameter template. When the number of parameters in the historical waveform feature parameter set reaches the preset update threshold, the standard waveform parameter template is updated based on the historical waveform feature parameter set. If it is determined that there is no standard waveform parameter template in the knowledge base that corresponds to the fault type confirmation result, a new standard waveform parameter template is constructed with the parameter to be confirmed as the initial value, and then stored in the knowledge base after being bound to the corresponding knowledge unit.

[0015] Optionally, in one possible implementation, the step of retrieving the corresponding knowledge unit from the knowledge base according to the target fault type, organizing it into a standardized answer, and outputting it to the scheduling end includes: Based on the target fault type, retrieve the corresponding knowledge unit from the knowledge base, and add the target fault type, the total amount of the comprehensive deviation set, and the historical fault case number on which the matching is based to the standardized answer; When the total amount of the comprehensive deviation set is determined to be within the preset prompt range, the fault type with the second smallest total amount of the comprehensive deviation set is obtained. The handling steps of the knowledge unit corresponding to the second smallest fault type and the handling steps corresponding to the target fault type are integrated into the standardized answer. The total amount of the comprehensive deviation set and the difference description corresponding to each fault type are marked in the standardized answer, and a manual confirmation prompt is sent to the dispatcher. When the total set of comprehensive deviations is not within the preset prompt range, the standardized answer will be output to the scheduling terminal.

[0016] Optionally, in one possible implementation, the step of performing coordinate processing on the fault waveform uploaded by the scheduling terminal to identify the waveform contours of each phase further includes: When it is determined that the fault waveform diagram contains voltage and current channels, the waveform profiles of each phase corresponding to the voltage and current channels are identified respectively to obtain voltage waveform profile groups and current waveform profile groups. The voltage waveform profile group and the current waveform profile group are divided into positive and negative half-cycles, and the normalized half-cycle area difference, discontinuity duration and phase difference sequence are extracted to obtain the voltage matching parameter group and the current matching parameter group. Obtain the normalized positive half-cycle area ratio of the voltage waveform profile group and the current waveform profile group under the same phase. Based on the normalized positive half-cycle area ratio of the voltage waveform profile group and the normalized positive half-cycle area ratio of the current waveform profile group, obtain the voltage and current positive half-cycle area ratio deviation of each phase. Determine the phases with voltage and current positive half-cycle area ratio deviations greater than a preset deviation threshold as abnormal phases, and obtain the abnormal phase set. The voltage matching parameter group and the current matching parameter group are used to match the standard waveform parameter templates corresponding to each fault type in the knowledge base.

[0017] Optionally, in one possible implementation, after determining the phases with voltage-current positive half-cycle area ratio deviations greater than a preset deviation threshold as abnormal phases and obtaining the abnormal phase set, the method further includes: When it is determined that the abnormal phase set is not empty, the number of abnormal phases and the identifier of each abnormal phase in the abnormal phase set are used as additional parameters, which, together with the voltage matching parameter group and the current matching parameter group, participate in the matching with the standard waveform parameter templates corresponding to each fault type in the knowledge base. After the matching is completed, when it is confirmed that the number of abnormal phases and the identifier of each abnormal phase are consistent with the pre-stored standard waveform parameter template corresponding to the target fault type, the total amount of the comprehensive deviation set obtained by the matching is adjusted to the preset confidence interval to obtain the total amount of the comprehensive deviation set after confidence adjustment. The total set of comprehensive deviations after confidence adjustment is used to determine the target fault type; When the abnormal phase set is determined to be empty, only the voltage matching parameter group and the current matching parameter group are used to match the standard waveform parameter templates corresponding to each fault type in the knowledge base.

[0018] Optionally, in one possible implementation, the step of comparing the AB phase difference, BC phase difference, and AC phase difference with the pre-stored standard phase differences in the knowledge base to obtain a phase difference sequence includes: When any phase deviation exceeds a preset phase threshold, the corresponding phase is marked as an abnormal phase, and the abnormal phase identifier is included in the phase difference sequence as an additional dimension recorded in the phase difference sequence. When all phase deviations are determined to be within the preset phase threshold, the phase difference sequence is marked as symmetrical and normal, and no phase abnormality identifier is generated.

[0019] The method for constructing a professional knowledge base and intelligently querying information for power grid dispatching provided in this application has the following beneficial effects: 1. After the fault waveform diagram is processed into coordinates, this application extracts the normalized half-cycle area difference, discontinuity duration and phase difference sequence of each phase as matching parameters, and matches them with the standard waveform parameter templates stored in the knowledge base. The fault waveform diagram is directly used as the knowledge base query entry point, so that dispatchers can obtain the corresponding handling steps without completing the text description of the fault type. This solves the problem that the knowledge base auxiliary decision-making function of the existing solution is actually ineffective when dispatchers cannot determine the fault type. 2. This application introduces an intermittent phase set mechanism in the matching parameters. Only phases with intermittent segments are constructed into an intermittent deviation set and included in the comprehensive deviation set. Phases without intermittent segments are not included in the construction of intermittent deviations. This allows the unique feature of inrush current, such as the duration of the intermittent segment, to play a role in effectively distinguishing fault types and to automatically exit during interference matching. This avoids mismatches caused by uniformly processing all phases and improves the accuracy of waveform feature matching. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the method for constructing a professional knowledge base and intelligently querying information for power grid dispatching provided in an embodiment of this application. Figure 2 This is a schematic diagram of the power grid dispatching knowledge graph structure of the professional knowledge base construction and intelligent query method for power grid dispatching provided in the embodiments of this application; Figure 3 This is a diagram showing the structure of a multi-type knowledge resource metadata model for the construction and intelligent query method of a professional knowledge base for power grid dispatching provided in this application embodiment; Figure 4 This is a diagram of the power grid dispatching knowledge base structure for the power grid dispatching professional knowledge base construction and intelligent query method provided in this application embodiment; Figure 5 This is a diagram of the scheduling knowledge question-answering service architecture based on LLM and knowledge base, which is a method for constructing a professional knowledge base and intelligent querying for power grid scheduling provided in the embodiments of this application. Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0023] See Figure 1 This is a flowchart illustrating the method for constructing a professional knowledge base and performing intelligent queries for power grid dispatching, provided in an embodiment of this application. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not impose any limitations on this.

[0024] The method for constructing a professional knowledge base and intelligently querying power grid dispatching provided in this application includes steps S1 to S4, as follows: Step S1: Perform coordinate processing on the fault waveform diagram uploaded by the dispatch terminal to identify the waveform contours of each phase.

[0025] It should be noted that the substation waveform recording device automatically generates a fault waveform diagram and uploads it to the dispatching terminal after a fault occurs. The fault waveform diagram records the waveform changes of each phase voltage and current before and after the fault in image form, which is the core basis for dispatchers to determine the fault type. Fault waveform diagrams output by waveform recording devices from different manufacturers vary in image size, waveform line thickness, and channel arrangement. The images do not include a unified numerical coordinate system, and the waveforms are presented in pixel form within the image. In this case, if the image is directly scanned column by column extreme values, scattered pixels near the zero baseline due to printing noise will be mistakenly included in the waveform contour, leading to distortion of the subsequently extracted matching parameters. For example, in the waveform diagram of a single-phase ground fault on a 220 kV line, there are a large number of background noise pixels near the zero baseline of phase A. The traditional column-by-column extreme value method misjudges the noise points as part of the waveform contour of phase A, resulting in an underestimation of the positive half-cycle area. The matching result incorrectly points to a three-phase short-circuit fault, and the dispatchers retrieve the wrong handling knowledge unit based on this. Traditional image processing methods cannot reliably distinguish between real waveform contours and background noise without relying on manually labeled coordinates. Step S1 establishes a unified coordinate system and locates the zero-value baselines of each phase based on pixel density distribution, distinguishing the waveform contours from noise at the pixel level, thus laying the foundation for the accurate extraction of subsequent matching parameters.

[0026] In some embodiments, step S1 includes steps S11 to S13: Step S11: Perform coordinate processing with the lower left corner of the fault waveform as the origin to determine the time coordinates and amplitude coordinates of each pixel.

[0027] The time coordinate refers to the column number of a pixel on the horizontal axis of the image from the origin, while the amplitude coordinate refers to the row number of a pixel on the vertical axis of the image from the origin. The origin is set at the bottom left corner to align the coordinate directions with the physical meaning of the waveform's time axis (left-to-right) and amplitude axis (bottom-to-top). The specific operation is as follows: Read the pixel matrix of the fault waveform. Assume the image width is W columns and the height is H rows. For the pixel in the r-th row and c-th column, the time coordinate is set to c, and the amplitude coordinate is set to H minus r. Convert the image row numbering from top to bottom to amplitude direction numbering from bottom to top. After coordinate conversion, all pixels participate in subsequent analysis in the form of time and amplitude coordinates.

[0028] It should be noted that the coordinate processing transforms the dimensionless pixel position into coordinate values ​​with a physical correspondence between time and amplitude, so that the operations in steps S12 to S13 are all completed within a unified coordinate system without relying on image metadata or external scale annotations. This adapts to the differences in image formats of different recording devices and avoids zero-baseline positioning deviations caused by inconsistent image formats.

[0029] Step S12: Count the number of pixels corresponding to each amplitude coordinate for each phase channel, and determine the amplitude coordinates where the number of pixels in each phase channel exceeds the preset density threshold as the zero value baseline of each phase.

[0030] Each phase channel refers to the image region divided by phase in the fault waveform recording. Phase A, phase B, and phase C each occupy a segment of the image's vertical direction. The number of pixels corresponding to each amplitude coordinate is the count of all pixels with the same amplitude coordinate value within a certain phase channel. The preset density threshold is used to distinguish pixel clusters near zero values ​​from ordinary waveform pixels. Those skilled in the art can select this threshold based on the actual resolution of the waveform recording and the thickness of the waveform lines. When the image resolution is high and the lines are thin, the preset density threshold can be appropriately reduced; when there is a lot of image noise, the preset density threshold should be appropriately increased to eliminate interference from discrete noise points. In a normal three-phase AC waveform, the contour lines of both the positive and negative half-cycles pass near the zero value, resulting in a much higher number of pixels accumulated at the amplitude coordinate where the zero baseline is located compared to other amplitude coordinate positions. Using this pattern, the zero baseline of each phase can be located without relying on external annotations.

[0031] It should be noted that the zero-value baseline is located based on the pixel density distribution, so that the determination of the zero-value baseline is independent of the fixed row number of the image or the estimation of the image center. It can still be accurately located when there are differences in the image layout generated by different recording devices, and provides a reliable reference line for the positive and negative half-cycle segmentation in step S2.

[0032] In some embodiments, step S12 includes steps S121 to S123: Step S121: Count the number of pixels corresponding to each amplitude coordinate in each phase channel to obtain the pixel distribution of each phase amplitude.

[0033] The amplitude pixel distribution of each phase is a distribution relationship formed by using amplitude coordinates as indices and the corresponding number of pixels as values, reflecting the aggregation of pixels in the amplitude direction within each phase channel. The specific operation is as follows: all pixels in a certain phase channel are grouped and counted according to their amplitude coordinates. A key-value pair sequence is formed by combining the amplitude coordinates with the corresponding number of pixels. Phases A, B, and C are counted independently to obtain three sets of amplitude pixel distributions for each phase.

[0034] It should be noted that the amplitude pixel distribution of each phase channel is independently statistically analyzed to avoid mutual interference between waveform contour pixels of different phases in the same statistical analysis. This ensures that the positioning of the zero-value baseline of each phase depends only on the pixel distribution pattern within the phase channel, providing input for the classification judgment in steps S122 and S123.

[0035] Step S122: When it is determined that there are two amplitude coordinates in the amplitude pixel distribution of each phase where the number of pixels exceeds the preset dense threshold, the middle amplitude coordinate of the two amplitude coordinates is taken as the zero value baseline of each phase.

[0036] The amplitude coordinates where the number of pixels exceeds the preset density threshold are caused by the waveform outline being drawn with lines of a certain width. Lines near the zero baseline appear as two parallel edge lines in the image, each accumulating a large number of pixels at its corresponding amplitude coordinate, forming two dense peaks. Taking the median amplitude coordinate between the two, i.e., the geometric center of the two dense peaks, corresponds to the actual centerline position of the waveform line.

[0037] It should be noted that if the amplitude coordinates corresponding to any dense peak are directly taken as the zero baseline, a systematic deviation of half the line width will be introduced, resulting in a fixed offset in the calculation of the positive and negative half-circle areas in step S313. Taking the intermediate amplitude coordinates will eliminate this deviation and ensure the symmetry of the calculation of the positive and negative half-circle areas.

[0038] Step S123: When it is determined that there is only one amplitude coordinate in the amplitude pixel distribution of each phase with a number of pixels exceeding the preset dense threshold, the amplitude coordinate of the only one amplitude coordinate with a number of pixels exceeding the preset dense threshold is taken as the zero value baseline of each phase.

[0039] There is only one dense peak. When the waveform outline is drawn with a single pixel thin line in the waveform record, all pixels near the zero baseline are concentrated on a single amplitude coordinate. There is no double edge effect. The amplitude coordinate corresponding to the only dense peak can be directly used as the zero baseline of each phase.

[0040] It should be noted that steps S122 and S123 together constitute a complete coverage of the number of dense peaks in the amplitude pixel distribution of each phase. When there are two dense peaks, step S122 is performed, and when there is only one dense peak, step S123 is performed. Under different waveform image quality and line width conditions, the zero value baseline of each phase can be accurately located, and there is no situation where it cannot be assigned to any branch.

[0041] Preferably, after determining the zero-value baselines for each phase, the amplitude coordinates of the zero-value baselines for phase A, phase B, and phase C can be further compared with the midpoint coordinates of the longitudinal range of each phase channel. If the zero-value baseline of a certain phase deviates from the midpoint coordinates of the phase channel by more than the allowable offset determined by those skilled in the art according to the waveform chart format specification, the zero-value baseline of that phase is marked as pending review and a re-statistical analysis of that phase channel is triggered to detect the positioning anomalies caused by severe noise in the waveform chart.

[0042] Step S13: Identify the sequence of pixels that continuously deviate from the zero baseline in each phase channel as the waveform profile of each phase, and obtain the waveform profile of phase A, phase B and phase C.

[0043] A sequence of pixels continuously deviating from the zero baseline refers to a set of pixels whose amplitude coordinates are not equal to the amplitude coordinates of the zero baselines of each phase channel and are continuously arranged in the time coordinate direction. The identification method is as follows: For all pixels in each phase channel, pixels whose amplitude coordinates are equal to the amplitude coordinates of the zero baselines of each phase are removed. The remaining pixels are arranged in ascending order of time coordinates. Pixels whose adjacent time coordinate differences do not exceed the continuity tolerance are grouped into the same continuous sequence. The continuous sequence with the largest number of pixels is taken as the waveform profile of each phase. The continuity tolerance is selected based on the maximum allowable breakpoint width of the waveform lines in the time direction in the waveform recording. Those skilled in the art can adjust this based on actual waveform recording samples.

[0044] It should be noted that step S13 excludes background pixels and noise pixels near the zero baseline, retaining only the deviated pixel sequence that truly reflects the waveform change process, so that the pixel composition of the A-phase waveform contour, B-phase waveform contour and C-phase waveform contour is consistent with the actual physical waveform shape, ensuring the clean boundary of the positive and negative half-cycle segmentation in step S2, and also ensuring that the calculation of each matching parameter in step S3 is not affected by background noise.

[0045] In some embodiments, step S1 further includes steps S141 to S144: Step S141: When it is determined that the fault waveform diagram contains voltage and current channels, the waveform profiles of each phase corresponding to the voltage and current channels are identified respectively to obtain voltage waveform profile groups and current waveform profile groups.

[0046] The voltage channel refers to the image area recording the waveforms of phase A voltage, phase B voltage, and phase C voltage. The current channel refers to the image area recording the waveforms of phase A current, phase B current, and phase C current. The voltage waveform profile group is a set of three waveform profiles obtained by performing steps S11 to S13 on each of the three phases within the voltage channel. The current waveform profile group is a set of three waveform profiles obtained by performing steps S11 to S13 on each of the three phases within the current channel. Step S141 is performed only when the fault waveform recorder contains both voltage and current channels. If the fault waveform recorder contains only one type of channel, waveform profile recognition is performed only for the existing channel type, and steps S142 to S144 are not performed.

[0047] It should be noted that waveform parameters from a single channel cannot fully reflect the characteristics of the fault type. For example, in a single-phase ground fault, the voltage of the faulty phase drops while the current surges, and the waveform changes in opposite directions. Relying solely on current waveform parameters can lead to confusion between single-phase ground faults and phase-to-phase short circuits under certain operating conditions. Extracting voltage waveform profiles and current waveform profiles separately provides the input basis for simultaneously extracting two sets of matching parameters in step S142.

[0048] Step S142: Perform positive and negative half-cycle segmentation on the voltage waveform profile group and the current waveform profile group respectively, and extract the normalized half-cycle area difference, discontinuity duration and phase difference sequence to obtain the voltage matching parameter group and the current matching parameter group.

[0049] The methods for segmenting the positive and negative half-cycles and extracting the normalized half-cycle area difference, discontinuity duration, and phase difference sequences are consistent with the descriptions in steps S311 to S315, the only difference being that the input objects are voltage waveform profile groups and current waveform profile groups, respectively. The voltage matching parameter group is a set of normalized half-cycle area difference, discontinuity duration, and phase difference sequences extracted from the voltage waveform profile group, and the current matching parameter group is a set of normalized half-cycle area difference, discontinuity duration, and phase difference sequences extracted from the current waveform profile group.

[0050] It should be noted that the voltage matching parameter group and the current matching parameter group reflect the morphological characteristics of the voltage waveform and the current waveform during the fault process, respectively. The two participate in the knowledge base matching in step S144, which expands the parameter dimension on which the matching is based from a single channel to two channels, voltage and current, reducing the probability of misjudging different fault types as the same type when the voltage or current single channel parameters are similar.

[0051] Step S143: Obtain the normalized positive half-cycle area ratio of the voltage waveform profile group and the current waveform profile group under the same phase. Based on the normalized positive half-cycle area ratio of the voltage waveform profile group and the normalized positive half-cycle area ratio of the current waveform profile group, obtain the voltage and current positive half-cycle area ratio deviation of each phase. Determine the phases with voltage and current positive half-cycle area ratio deviations greater than the preset deviation threshold as abnormal phases, and obtain the abnormal phase set.

[0052] The normalized positive half-cycle area ratio refers to the ratio of the positive half-cycle area of ​​a phase waveform profile to the sum of the positive and negative half-cycle areas. The calculation method is the same as in step S313. The voltage-current positive half-cycle area ratio deviation refers to the absolute value of the difference between the normalized positive half-cycle area ratio of the voltage waveform profile group and the normalized positive half-cycle area ratio of the current waveform profile group under the same phase. The preset deviation threshold is used to distinguish between the normal state where the voltage and current positive half-cycle area ratios are consistent and the abnormal state where they deviate. Those skilled in the art can set this threshold based on the deviation range of the voltage and current normalized positive half-cycle area ratios under normal three-phase AC operating conditions. Under normal operating conditions, the deviation is usually no more than 0.05. This range can be used as a reference to select the preset deviation threshold in conjunction with actual waveform recording samples. The specific operation is as follows: calculate the normalized positive half-cycle area ratio of the voltage waveform profile group and the normalized positive half-cycle area ratio of the current waveform profile group for phases A, B, and C respectively. Take the absolute value of the difference between the two ratios of the same phase and compare them one by one with the preset deviation threshold. Include the phases with deviations exceeding the preset deviation threshold into the abnormal phase set.

[0053] It should be noted that the voltage and current waveforms of the faulty phase will exhibit asymmetrical changes when a fault occurs, causing the positive half-cycle area ratio of the faulty phase voltage and current to deviate, while the non-faulty phases remain consistent under normal operating conditions. By pre-locating the phase affected by the fault through the deviation of the positive half-cycle area ratio of voltage and current, a basis is provided for incorporating abnormal phase information as an additional parameter into the matching in step S1431, thereby improving the sensitivity of the knowledge base matching to the faulty phase.

[0054] In some embodiments, after determining the phases with voltage-current positive half-cycle area ratio deviations greater than a preset deviation threshold as abnormal phases in step S143 and obtaining the abnormal phase set, steps S1431 to S1434 are included: Step S1431: When it is determined that the abnormal phase set is not empty, the number of abnormal phases and the identifier of each abnormal phase in the abnormal phase set are used as additional parameters, and together with the voltage matching parameter group and the current matching parameter group, they participate in the matching with the standard waveform parameter templates corresponding to each fault type in the knowledge base.

[0055] The number of abnormal phases refers to the number of phase identifiers in the abnormal phase set, with a value of 1, 2, or 3. Each abnormal phase identifier refers to the number of each phase in the abnormal phase set, which can be phase A, phase B, or phase C. Using the number of abnormal phases and each abnormal phase identifier as additional parameters means that when initiating matching with the standard waveform parameter templates corresponding to each fault type in the knowledge base, in addition to the voltage matching parameter group and the current matching parameter group, the number of abnormal phases and each abnormal phase identifier are also sent to the matching calculation. The standard waveform parameter templates corresponding to each fault type in the knowledge base pre-store the corresponding standard values ​​for the number of abnormal phases and the standard values ​​for the abnormal phase identifiers. During matching, the additional parameters are compared with the standard values. It is easy to understand that the number of abnormal phases for a single-phase ground fault is 1, and the number of abnormal phases for a three-phase short-circuit fault is 3. The introduction of additional parameters makes the matching process sensitive to the phase distribution characteristics of different fault types.

[0056] It should be noted that step S1431 is only executed when the abnormal phase set is not empty. The abnormal phase information is used as an auxiliary dimension to participate in the matching rather than directly determining the result. In step S1432, confidence adjustment is used to provide biased support for the case where the additional parameters match, so as to avoid the abnormal phase identification error having a veto effect on the matching conclusion when the waveform quality is unstable.

[0057] Step S1432: After the matching is completed, when it is determined that the number of abnormal phases and the identifier of each abnormal phase are consistent with the pre-stored standard waveform parameter template corresponding to the target fault type, the total amount of the comprehensive deviation set obtained by matching is adjusted to the preset confidence interval to obtain the total amount of the comprehensive deviation set after confidence adjustment.

[0058] The total comprehensive deviation set is the sum of the comprehensive deviation sets of each fault type calculated in steps S321 to S323. The smaller the total, the higher the degree of matching. The preset confidence interval is a numerical range of the total comprehensive deviation set that represents the degree of confidence of the matching result. Those skilled in the art can set it according to the parameter dispersion of the standard waveform parameter templates for each fault type in the knowledge base. Adjusting the total comprehensive deviation set obtained from the matching to the preset confidence interval means that when the original value of the total comprehensive deviation set is outside the preset confidence interval, but the additional parameters of the abnormal phase are consistent with the standard value, the total comprehensive deviation set is adjusted to the endpoint value of the preset confidence interval.

[0059] It should be noted that, while retaining the multidimensional deviation information of the voltage matching parameter group and the current matching parameter group, the confidence adjustment incorporates the consistency of the additional parameters of the abnormal phase as an additional confidence basis. This makes the determination of the target fault type in step S324 subject to the constraints of both waveform parameters and fault phase, which is of direct significance for dispatchers to retrieve the correct handling knowledge unit.

[0060] Step S1433: Use the total amount of the comprehensive deviation set after confidence adjustment to participate in the determination of the target fault type.

[0061] The total amount of the comprehensive deviation set after confidence adjustment replaces the total amount of the original comprehensive deviation set in steps S321 to S323 and is sent to step S324 to participate in the determination of the target fault type. The fault type with the smallest total amount of the comprehensive deviation set after confidence adjustment is taken as the target fault type. The specific judgment logic is the same as that in step S324.

[0062] It should be noted that step S1433 seamlessly connects the total amount of the comprehensive deviation set after confidence adjustment to the judgment process of step S324, so that the matching result of the additional parameters of the abnormal phase is integrated into the determination of the target fault type in a quantifiable way, rather than being processed separately as an independent judgment rule, thereby reducing the decision complexity when the two types of information conflict.

[0063] Step S1434: When the abnormal phase set is determined to be empty, only the voltage matching parameter group and the current matching parameter group are used to match the standard waveform parameter templates corresponding to each fault type in the knowledge base.

[0064] An empty abnormal phase set means that the voltage and current positive half-cycle area ratio deviations of phases A, B, and C do not exceed the preset deviation threshold. This indicates that no signal affecting a specific phase due to a fault has been detected, and there are no available additional parameters for the abnormal phase. The matching process relies only on the voltage matching parameter set and the current matching parameter set, and is executed according to the standard procedure of steps S321 to S324, without triggering the confidence adjustment in steps S1431 to S1433. Steps S1431 and S1434 together constitute a complete coverage of the state judgment of whether the abnormal phase set is empty, and there is no situation where it cannot be classified into any path.

[0065] Preferably, after step S143 is completed, the voltage and current positive half-cycle area ratio deviation values ​​of each phase in the abnormal phase set can be further sorted, the phase with the largest deviation value is marked as the main abnormal phase, the main abnormal phase is listed separately in the additional parameters, and compared with the pre-stored standard value of the main abnormal phase in the standard waveform parameter template of each fault type in the knowledge base, so as to provide more granular additional parameter support in the case of complex fault scenarios such as phase-to-phase short circuit and single-phase grounding accompanied by disturbance.

[0066] Step S144: Input the voltage matching parameter group and the current matching parameter group together into the matching step with the standard waveform parameter template corresponding to each fault type in the knowledge base.

[0067] The normalized half-cycle area difference, discontinuity duration, and phase difference sequences from the voltage matching parameter group and the current matching parameter group are combined into a single input. Following the matching process from steps S321 to S324, the deviation is calculated against the standard waveform parameter templates corresponding to each fault type in the knowledge base. This yields the total comprehensive deviation set for each fault type, thereby determining the target fault type. The standard waveform parameter templates corresponding to each fault type in the knowledge base contain standard parameters for both voltage and current channels during construction, corresponding one-to-one with the input structure of step S144.

[0068] It should be noted that step S144 is the convergence point of steps S141 to S143, where the results of dual-channel waveform contour recognition and abnormal phase detection are uniformly sent to the knowledge base matching process. The matching results are accompanied by procedure source information, and the dispatcher can trace the matching basis after obtaining the target fault type, ensuring the verifiability of the handling knowledge unit.

[0069] Preferably, after step S1 is completed, the waveform profiles of phase A, phase B, and phase C, the amplitude coordinates of the zero-value baselines of each phase, as well as the voltage waveform profile group, current waveform profile group, voltage matching parameter group, current matching parameter group, and abnormal phase set obtained during steps S141 to S144, can be uniformly cached in the intermediate result storage area of ​​this fault waveform analysis task for use in steps S2 to S4. This avoids repeatedly performing coordinate processing and waveform profile recognition on the same fault waveform image, and reduces the resource consumption of repeated calculations when processing multiple fault waveform images simultaneously at the scheduling end.

[0070] Step S2: Divide the waveform profile of each phase into positive and negative half cycles using the zero-value baseline corresponding to the waveform profile of each phase to obtain the positive and negative half cycle division results.

[0071] It should be noted that the positive and negative half-cycle segmentation is a prerequisite for subsequent matching parameter extraction, and its accuracy directly determines the calculation quality of the normalized half-cycle area difference and discontinuity duration of each phase in step S3. In actual waveform recordings, the waveform profile is not a strict sine curve. After a fault occurs, the waveform may be distorted, truncated, or discontinuous. If a fixed image center line is used as the segmentation reference, the actual offset zero value position will be misjudged as the zero axis, resulting in the calculation reference of the positive and negative half-cycle areas not matching the true waveform. Using the zero value baseline of each phase determined in step S12 as the segmentation reference ensures that the reference for positive and negative half-cycle segmentation strictly corresponds to the true zero value position of each phase, avoiding systematic errors introduced by reference offset.

[0072] The specific operation of positive and negative half-cycle segmentation is as follows: For the waveform contours of phase A, phase B, and phase C, the amplitude coordinates of the zero-value baseline of each phase are used as the dividing boundary. The sequence of pixels in each phase waveform contour whose amplitude coordinates are greater than the amplitude coordinates of the zero-value baseline of each phase are defined as the positive half-circle contour, the sequence of pixels whose amplitude coordinates are less than the amplitude coordinates of the zero-value baseline of each phase are defined as the negative half-circle contour, and the sequence of pixels whose amplitude coordinates are equal to the amplitude coordinates of the zero-value baseline of each phase are recorded as zero-value coincident segments for use in step S312 to identify discontinuities.

[0073] The positive and negative half-cycle segmentation results include the positive half-cycle contour, negative half-cycle contour, and zero-value coincidence segment of each of phases A, B, and C, which serve as the input for step S3.

[0074] Preferably, after the positive and negative half-cycles are segmented, the number of pixels in the positive and negative half-cycle contours of each phase can be checked. If the number of pixels in the positive or negative half-cycle contour of a certain phase is lower than the minimum number of pixels determined by those skilled in the art based on the waveform recording resolution and the number of normal waveform cycles, then that phase is marked as an incomplete waveform phase. In the matching process of step S3, the matching parameters for incomplete waveform phases are assigned a lower reference weight to reduce the interference of missing waveform data on the matching results. The minimum number of pixels is selected based on the product of the number of complete cycles of the normal three-phase AC waveform within the recording time period and the waveform recording resolution, which can be determined by those skilled in the art based on commonly used waveform recording samples.

[0075] Step S3: Extract the matching parameters of each phase based on the positive and negative half-cycle segmentation results, and match the matching parameters with the standard waveform parameter templates corresponding to each fault type in the knowledge base to obtain the target fault type.

[0076] It should be noted that fault type identification in power grid dispatching operations has historically relied on the accumulated experience of dispatchers. Different shift workers may arrive at different conclusions when faced with the same fault waveform chart due to differences in experience. The risk of misjudgment is even higher under fatigue conditions such as night shifts. The knowledge base stores standard waveform parameter templates corresponding to each fault type. It is a structured accumulation of historical fault waveform data, covering the waveform morphology characteristics of common fault types such as line tripping, bus faults, and protection malfunctions. For the question-and-answer service processing flow, please refer to [link / reference]. Figure 5 The Chinese natural language questions input by dispatchers are converted into question text vectors by a vector embedding model. These vector vectors are then matched against pre-stored knowledge base text vectors in the FAISS vector database, and the matching text vectors are returned by a text vector retrieval machine. Text from the power grid dispatch knowledge base is loaded and segmented, then synchronously loaded into the vector embedding model to generate text vectors, which are then stored in the FAISS vector database. The Chat large language model, based on the prompt template, associates the retrieved knowledge materials to generate answers and references, which are then returned to the dispatching end. The deviation calculation in step S3 between the matching parameters and the standard waveform parameter templates corresponding to each fault type in the knowledge base corresponds to... Figure 5 In the similarity retrieval step, the determination of the target fault type corresponds to... Figure 5 The reasoning output stage in the process.

[0077] Step S3 transforms the positive and negative half-cycle segmentation results obtained in step S2 into matching parameters that can be directly compared with the standard waveform parameter template. Then, by minimizing the total amount of the comprehensive deviation set, the target fault type is located, so that fault judgment is transformed from relying on personal experience to objective quantitative comparison based on waveform characteristics. This provides a traceable basis for dispatchers to retrieve the correct handling knowledge unit in step S4.

[0078] In some embodiments, step S3 includes steps S31 and S32: Step S31: Extract the matching parameters of each phase based on the positive and negative half-cycle segmentation results.

[0079] Matching parameters refer to the set of parameters extracted from the positive and negative half-cycle segmentation results that can quantitatively characterize the waveform morphology of each phase. These parameters include the normalized half-cycle area difference of each phase, the discontinuity duration of each phase, and the phase difference sequence. These three types of parameters describe the fault recorded waveform from three dimensions: waveform area symmetry, waveform discontinuity characteristics, and three-phase phase relationship. Together, they form the input for comparison with the standard waveform parameter templates corresponding to each fault type in the knowledge base.

[0080] It should be noted that step S31 compresses the two-dimensional waveform contour image information into three types of quantifiable parameters, enabling the matching calculation in the subsequent step S32 to be completed in a low-dimensional parameter space. This avoids the instability of comparison results caused by image noise and resolution differences when directly comparing the similarity of waveform image pixels. Each of the three types of parameters captures different dimensional features of the waveform: the normalized half-cycle area difference reflects whether the energy distribution of the positive and negative half-cycles is symmetrical; the discontinuity duration reflects whether the waveform has current or voltage interruptions; and the phase difference sequence reflects whether the three-phase timing relationship has shifted. The combination of these three parameters can distinguish different fault types with similar waveform features in the knowledge base matching process.

[0081] In some embodiments, step S31 includes steps S311 to S315: Step S311: Obtain all intersection points of each phase waveform profile with the zero baseline, take the waveform profile segment above the zero baseline between two adjacent intersection points as the positive half-cycle profile, and take the waveform profile segment below the zero baseline as the negative half-cycle profile.

[0082] The intersection points of each phase waveform profile and the zero-value baseline refer to the pixels in the pixel sequence of each phase waveform profile whose amplitude coordinates are equal to the amplitude coordinates of the zero-value baseline of each phase. These intersection points are then arranged in ascending order of time coordinates to obtain the complete sequence of intersection points. A waveform profile segment between two adjacent intersection points refers to a subsequence of pixels located between two adjacent intersection points on the time coordinate. For each waveform profile segment, the average amplitude coordinates of all pixels within the subsequence are taken. If the average value is greater than the amplitude coordinates of the zero-value baseline of each phase, the segment is designated as a positive half-cycle profile; if the average value is less than the amplitude coordinates of the zero-value baseline of each phase, the segment is designated as a negative half-cycle profile; and if the average value is equal to the amplitude coordinates of the zero-value baseline of each phase, the segment is sent to step S312 for identification as a discontinuity segment.

[0083] It should be noted that the positive and negative half-cycle profiles are defined by the intersection of the waveform profile of each phase with the zero-value baseline. This ensures that the division of the positive and negative half-cycle profiles strictly corresponds to the physical process of the actual waveform crossing the zero value, avoiding the half-cycle misalignment problem that occurs in waveform distortion scenarios when the half-cycle is divided by a fixed time window. This establishes accurate geometric boundaries for the calculation of the positive and negative half-cycle areas in step S313.

[0084] Step S312: When the waveform profile segment between two adjacent intersection points coincides with the zero baseline, the coincident segment is identified as an intermittent segment. The time coordinate length of all intermittent segments in the waveform profile of each phase is counted to obtain the intermittent duration of each phase.

[0085] When a waveform profile segment coincides with the zero-value baseline, it means that the amplitude coordinates of all pixels within the waveform profile segment between two adjacent intersection points are equal to the amplitude coordinates of the zero-value baseline for each phase. This indicates that the waveform remains at a zero value position for that time period, corresponding to a physical interruption in current or voltage. The time coordinate length of the discontinuity segment refers to the difference between the maximum and minimum time coordinates within the coincident segment. The discontinuity duration for each phase is the summed value obtained by statistically analyzing the time coordinate lengths of all discontinuities in the waveform profile of each phase.

[0086] It should be noted that the duration of the interruption is a key parameter for distinguishing between fault types that have waveform interruption characteristics and those that do not. For example, a wire break fault will produce a noticeable current waveform interruption in the broken phase, while a short circuit fault usually does not produce a current interruption. If the duration of the interruption is not extracted and only the area difference and phase difference are relied upon, misjudgment will occur when the area difference and phase difference values ​​of the two types of faults are similar.

[0087] In some embodiments, step S312 involves calculating the time coordinate lengths of all discontinuities in the waveform profiles of each phase to obtain the discontinuity duration of each phase, including steps S3121 to S3123: Step S3121: Count the number of intermediate segments in the waveform profile of each phase to obtain the number of segments between each phase.

[0088] The number of discontinuities in each phase is the result of counting all discontinuities identified in step S312 by phase. Phases A, B, and C are counted independently to obtain the number of discontinuities in phase A, phase B, and phase C. The statistical method is as follows: for all waveform contour segments in each phase that coincide with the zero baseline, adjacent overlapping segments are merged according to the continuity of the time coordinate and then counted to obtain the number of discontinuities in each phase.

[0089] It should be noted that the number of discontinuities in each phase is a prerequisite for selecting the matching basis in steps A1 to A3, and is also the basis for selecting the discontinuity duration calculation method in steps S3122 and S3123. Counting the number of discontinuities in each phase in advance ensures that the judgments of the two subsequent branches have clear numerical inputs, avoiding logical gaps caused by directly performing accumulation or single-selection operations when the number of discontinuities is unknown.

[0090] In some embodiments, after counting the number of intermediate segments in the waveform profile of each phase in step S3121 to obtain the number of interphase segments, steps A1 to A3 are included: Step A1: Count the number of discontinuous segments with a value greater than zero to obtain the set of discontinuous segments.

[0091] The discontinuous phase identification set refers to the set of phases whose numbers of discontinuities in phase A, phase B, and phase C are greater than zero. Each element in the set is a phase identifier, which can be an A-phase identifier, a B-phase identifier, or a C-phase identifier. The set can be empty. The specific operation is as follows: Check each of the discontinuities in phase A, phase B, and phase C one by one to see if it is greater than zero, and add the phase identifiers that meet the condition to the discontinuous phase identification set.

[0092] It should be noted that the discontinuity phase set identifies the phases with waveform interruption characteristics, providing input for the branch judgment in steps A2 and A3, and also providing a basis for clearly including the phase range in the calculation when constructing the discontinuity deviation set in step S322. This ensures that the discontinuity duration only participates in matching on phases with actual waveform interruption, avoiding the incorrect inclusion of zero discontinuity duration in the deviation calculation for phases without discontinuity.

[0093] Step A2: When the set of discontinuous phases is determined to be empty, it is determined that there are no discontinuous segments in the waveform profile of each phase, and the normalized half-circuit area difference and phase difference sequence of each phase are used as the matching basis.

[0094] An empty set of discontinuous phases means that the number of discontinuous segments in phase A, phase B, and phase C are all zero. This indicates that no waveform profile segment coincides with the zero baseline in any of the three phase waveform profiles, and the fault type does not contain waveform interruption characteristics. In this case, the discontinuity duration is zero for all fault types, and including it in the matching will not increase the distinguishing ability. Therefore, only the normalized half-cycle area difference and phase difference sequence of each phase are used as the matching basis, and matching is performed according to steps S321 and S323.

[0095] It should be noted that step A2 avoids forcibly including the zero discontinuity duration in the deviation calculation when there is no discontinuity feature, and prevents the zero discontinuity duration from interfering with the matching results dominated by the area deviation set and the phase deviation set, so that the matching basis strictly corresponds to the actual waveform characteristics of the current waveform record.

[0096] Step A3: When it is determined that the discontinuous phase set is not empty, the discontinuity duration of each phase in the discontinuous phase set, the normalized half-circuit area difference of each phase, and the phase difference sequence are used as the matching criteria.

[0097] A non-empty set of discontinuous phases means that at least one phase has a discontinuity segment count greater than zero, indicating that a fault caused waveform interruption in at least one phase. In this case, the discontinuity duration of each phase in the discontinuous phase set is merged with the normalized half-cycle area difference and phase difference sequence of each phase, and matching is performed according to steps S321 and S322, so that the discontinuity duration is used as a third-dimensional parameter to distinguish the fault type. Steps A2 and A3 together constitute a complete coverage of whether the discontinuous phase set is an empty set, and there is no situation where it cannot be classified into any branch.

[0098] It should be noted that the duration of the interruption is only included in the matching when the set of interruption phases is not empty. On the one hand, this avoids introducing invalid parameter dimensions in faults without interruption features, and on the other hand, it ensures that waveform interruption information is fully utilized in faults with interruption features, so that the number of dimensions for matching is adapted to the fault scenario according to the actual characteristics of the waveform rather than being fixed.

[0099] Preferably, after step S3121 is completed, the start and end time coordinates of the interruption segments of each phase in the interruption phase set can be further recorded. The start and end time coordinates of each phase interruption segment are stored as additional fields in the matching parameters for the scheduling expert to review in the manual confirmation stage of step S3242. This allows the scheduling expert to intuitively locate the time position of each phase waveform interruption when reviewing the parameters to be confirmed, thereby improving the efficiency of manual confirmation.

[0100] Step S3122: When the number of discontinuities in any phase is greater than one, sum the time coordinate lengths of all discontinuities in each phase to obtain the discontinuity duration of each phase.

[0101] The number of discontinuities in any phase is greater than one, meaning that at least one of the discontinuities in phase A, phase B, or phase C has a value greater than 1. In this case, if a phase has multiple discontinuities, the time coordinate length of each discontinuity needs to be included in the calculation. The sum of the time coordinate lengths of all discontinuities in each phase is taken as the discontinuity duration of each phase, reflecting the cumulative waveform interruption duration of that phase during the entire waveform recording period.

[0102] It should be noted that the time coordinate lengths of multiple discontinuities are summed instead of taking the maximum or average value because the standard values ​​of discontinuity durations pre-stored in the standard waveform parameter templates corresponding to each fault type in the knowledge base reflect the statistical regularity of the cumulative waveform interruption duration in historical fault recordings. This is consistent with the summation result in physical meaning, thus ensuring that the construction of the discontinuity deviation set in step S322 matches the standard value in terms of parameter caliber.

[0103] Step S3123: When the number of discontinuities in any phase is equal to one, the time coordinate length of the unique discontinuity in each phase is taken as the discontinuity duration of each phase.

[0104] If the number of discontinuities in any phase is equal to one, it means that the phase with a value of 1 among the discontinuity numbers of phase A, phase B, and phase C corresponds to a unique discontinuity. The time coordinate length of the unique discontinuity is directly used as the discontinuity duration of each phase, without the need for accumulation. Steps S3122 and S3123 together cover all cases where the number of discontinuities is greater than or equal to one. For phases with a number of discontinuities equal to zero, the discontinuity duration is zero and is not included in the construction of the discontinuity deviation set in step S322.

[0105] It should be noted that distinguishing between the two cases of the number of interruptions being greater than one and equal to one is to avoid performing meaningless accumulation operations in single interruption scenarios, while ensuring that the calculation path strictly corresponds to the actual number of interruptions, and guaranteeing that the interruption duration calculation results are consistent with the statistical caliber of the knowledge base standard value in different waveform recording scenarios.

[0106] Preferably, after step S312 is completed, the intermittent duration of each phase can be further compared with the maximum reasonable intermittent duration determined by those skilled in the art based on the rated frequency and recording time window. If the intermittent duration of a certain phase exceeds the maximum reasonable intermittent duration, the intermittent duration of that phase is marked as an abnormal value and a review is triggered to eliminate the situation of falsely high intermittent duration caused by the zero-value coincidence segment identification error in step S2.

[0107] Step S313: Based on the positive half-cycle area enclosed by the positive half-cycle profile and the zero-value baseline and the negative half-cycle area enclosed by the negative half-cycle profile and the zero-value baseline, and combined with the theoretical area of ​​the whole cycle of each phase waveform profile, the normalized half-cycle area difference of each phase is obtained, wherein the theoretical area of ​​the whole cycle is determined by the time coordinate span and amplitude coordinate peak value of each phase waveform profile.

[0108] The positive half-cycle area refers to the pixel area enclosed by the sequence of pixels on the positive half-cycle contour and the zero-value baseline in the coordinate plane formed by the time and amplitude coordinates, and is measured by the number of pixels. The negative half-cycle area refers to the pixel area enclosed by the sequence of pixels on the negative half-cycle contour and the zero-value baseline, and is measured by the number of pixels after taking the absolute value of the amplitude coordinate. The theoretical area of ​​the entire cycle refers to the theoretical area corresponding to the positive or negative half-cycle within a complete cycle under the conditions of the time coordinate span and amplitude coordinate peak value of each phase waveform contour, assuming the waveform is a standard sine wave. It is estimated by dividing the time coordinate span by the number of complete cycles of the standard sine wave and then multiplying it by the peak value of the amplitude coordinate. The peak value of the amplitude coordinate is the maximum absolute value of the amplitude coordinate in the pixel sequence of each phase waveform contour. The normalized half-cycle area difference of each phase is calculated by dividing the absolute value of the difference between the positive and negative half-cycle areas by the theoretical area of ​​the entire cycle to obtain the normalized area difference ratio.

[0109] It should be noted that normalizing the area difference between the positive and negative half-cycles using the theoretical area of ​​the whole cycle eliminates the difference in absolute pixel area caused by different image resolutions in different waveform recordings. This ensures that fault waveform recordings generated by different recording devices are comparable in terms of the calculated normalized half-cycle area difference, guaranteeing that the area deviation calculation in step S321 with the standard waveform parameter template in the knowledge base is based on a unified dimension. The normalized half-cycle area difference reflects the degree of asymmetry in the energy distribution of the positive and negative half-cycles of the waveform. It has different typical value ranges in different fault types such as single-phase grounding and phase-to-phase short circuits, and is one of the core parameters for distinguishing fault types.

[0110] Step S314: Subtract the time coordinates of the first positive intersection point of the waveform profiles of phase A, phase B, and phase C with the zero baseline pairwise to obtain the phase difference of AB, phase difference of BC, and phase difference of AC. Compare the phase difference of AB, phase difference of BC, and phase difference of AC with the standard phase difference stored in the knowledge base item by item to obtain the phase difference sequence.

[0111] The first positive intersection point refers to the first pixel in the sequence of pixels of each phase waveform contour where the amplitude coordinate changes from less than the amplitude coordinate of the zero baseline of each phase to greater than the amplitude coordinate of the zero baseline of each phase. The time coordinate of the first positive intersection point of each phase is taken as the phase time coordinate of each phase. The AB phase difference is the difference between the phase time coordinate of phase A and phase B, the BC phase difference is the difference between the phase time coordinate of phase B and phase C, and the AC phase difference is the difference between the phase time coordinate of phase A and phase C. The standard phase difference pre-stored in the knowledge base is the standard value corresponding to the normal phase timing relationship of the three phases under each fault type. The AB phase difference, BC phase difference, and AC phase difference are subtracted from the standard phase difference item by item to obtain three phase deviations, which together constitute the phase difference sequence.

[0112] It should be noted that during normal operation of the three phases, the AB phase difference, BC phase difference, and AC phase difference have fixed standard values. When a fault occurs, the time coordinates of the first positive intersection point of the phases affected by the fault will shift, causing the phase difference to deviate from the standard value. By capturing the shift characteristics of the three-phase timing relationship through the phase difference sequence, additional distinguishing factors can be provided between fault types with similar area difference parameters.

[0113] In some embodiments, step S314, which compares the AB phase difference, BC phase difference, and AC phase difference with the pre-stored standard phase differences in the knowledge base to obtain a phase difference sequence, includes steps B1 and B2: Step B1: When any phase deviation exceeds the preset phase threshold, mark the corresponding phase as an abnormal phase and include the abnormal phase identifier in the phase difference sequence as an additional dimension in the phase difference sequence.

[0114] Any phase deviation exceeding the preset phase threshold means that the absolute value of at least one of the three phase deviations (AB phase difference, BC phase difference, and AC phase difference) obtained by subtracting each item from the pre-stored standard phase difference in the knowledge base exceeds the preset phase threshold. The preset phase threshold is used to distinguish between normal and abnormal deviations in the three-phase timing relationship. Those skilled in the art can set it based on the phase jitter range of a normal three-phase AC system at its rated frequency. An abnormal phase refers to the phase corresponding to a phase difference whose phase deviation exceeds the preset phase threshold. For example, when the AB phase difference exceeds the preset phase threshold, both phases A and B are marked as candidates for abnormal phases. The actual abnormal phase is determined by combining the deviations of the BC and AC phase differences, and the abnormal phase identifier is added as an additional dimension to the phase difference sequence.

[0115] It should be noted that by incorporating the phase anomaly identification as an additional dimension into the phase difference sequence, the phase difference sequence carries both phase deviation numerical information and anomaly phase location information. In the phase deviation set construction in step S321, both types of information can be utilized simultaneously, thereby improving the matching accuracy of fault types that are sensitive to phase characteristics.

[0116] Step B2: When all phase deviations do not exceed the preset phase threshold, mark the phase difference sequence as symmetrical and normal, and do not generate a phase abnormality identifier.

[0117] If all phase deviations do not exceed the preset phase threshold, it means that the absolute values ​​of the three phase deviations (AB phase difference, BC phase difference, and AC phase difference) obtained by subtracting each from the standard phase difference are all within the preset phase threshold, indicating that there is no abnormal shift in the three-phase timing relationship. In this case, the phase difference sequence only contains the three phase deviation values, is marked as symmetrical and normal, and no phase abnormality identifier is added. It participates in the construction of the phase deviation set according to the standard procedure of step S321. Steps B1 and B2 together constitute a complete coverage of the state judgment of whether all phase deviations exceed the preset phase threshold.

[0118] It should be noted that when the phase relationship is normal, no phase anomaly identifier is generated, which avoids the interference of null value anomaly identifiers on matching calculation in fault scenarios without phase anomaly features, and maintains the consistency of the phase difference sequence structure under different fault scenarios.

[0119] Preferably, after step S314 is completed, the AB phase difference, BC phase difference and AC phase difference can be further cross-validated with the standard three-phase phase difference at the rated frequency, and the consistency deviation of the three phase differences can be calculated. If the consistency deviation exceeds the allowable range determined by those skilled in the art based on the rated frequency tolerance, the phase difference sequence is marked as low confidence. In step S321, a lower reference weight is assigned to the phase deviation set to reduce the misleading effect of the phase difference sequence on the matching result when the waveform is severely distorted and the first positive intersection point is identified as wrong.

[0120] Step S315: Use the normalized half-cycle area difference of each phase, the discontinuity duration of each phase, and the phase difference sequence as matching parameters.

[0121] The normalized half-cycle area difference of each phase is calculated by step S313, the discontinuity duration of each phase is calculated by step S3122 or step S3123, and the phase difference sequence is calculated by step S314. The three are combined into matching parameters, which are used as input for step S32.

[0122] It should be noted that step S315 integrates the three types of heterogeneous parameters into matching parameters, so that the matching calculation in step S32 is completed under a single input structure. The source and type of the parameters have been clarified in steps S311 to S314, and there is no need to reconfirm the physical meaning of the parameters in step S32, which reduces the complexity of the matching process.

[0123] Preferably, after step S31 is completed, the integrity of the normalized half-cycle area difference, the discontinuity duration and phase difference sequence of each phase can be checked. If any parameter cannot be calculated due to insufficient waveform data of the corresponding phase, the uncalculated parameter is marked as missing. In the matching of step S32, the parameter dimension containing the missing parameter is not included in the deviation set, so as to ensure that the matching calculation can still be performed normally when the parameters are incomplete.

[0124] Step S32: Match the matching parameters with the standard waveform parameter templates corresponding to each fault type in the knowledge base to obtain the target fault type.

[0125] The standard waveform parameter templates corresponding to each fault type in the knowledge base are parameter benchmarks stored after structured modeling of historical fault waveform data. Each fault type corresponds to one standard waveform parameter template, which includes the standard values ​​of the normalized half-cycle area difference of each phase, the standard values ​​of the intermittent duration of each phase, and the standard values ​​of the phase difference sequence for that fault type. The matching process involves calculating the deviation between the matching parameters obtained in step S315 and the standard waveform parameter templates corresponding to all fault types in the knowledge base, and determining the target fault type by the magnitude of the total set of comprehensive deviations. For the overall organizational structure of the knowledge base, please refer to [link to relevant documentation]. Figure 2The leftmost layer is the Knowledge Domain layer, containing N knowledge domain nodes from Domain 1 to Domain N. The middle layer is the Complex layer, where each domain node is mapped to several complex knowledge nodes (Complex Knowledge 1 to Complex Knowledge K) through inclusion / submission relationships. The rightmost layer is the Meta layer, where each complex knowledge node is further subdivided into meta-knowledge points (Meta-knowledge point 1 to Meta-knowledge point S). Dashed arrows indicate prerequisite dependencies between meta-knowledge points, meaning that mastering one meta-knowledge point presupposes mastering other meta-knowledge points. Through this three-layer structure, the standard waveform parameter template can establish a traceable association path with the corresponding knowledge unit.

[0126] It should be noted that step S32 is the core judgment step of this application, and its output, the target fault type, directly determines the content of the handling knowledge unit received by the dispatcher in step S4. Based on the matching method of the total comprehensive deviation set, the fault type judgment is transformed into a distance metric problem in the parameter space, so that the judgment result has a quantifiable confidence basis. After obtaining the target fault type, the dispatcher can evaluate the reliability of the matching result through the numerical value of the total comprehensive deviation set, rather than relying solely on a single conclusion given by the system.

[0127] In some embodiments, step S32 includes steps S321 to S324: Step S321: Based on the normalized half-cycle area difference of each phase and the corresponding parameters in each standard waveform parameter template, obtain the area deviation set of each fault type; based on the phase difference sequence and the corresponding parameters in each standard waveform parameter template, obtain the phase deviation set of each fault type.

[0128] The area deviation set for each fault type refers to the set of area deviations for phase A, phase B, and phase C obtained by subtracting the normalized half-cycle area difference of each phase from the standard value of the normalized half-cycle area difference of each phase in the standard waveform parameter template for the corresponding fault type in the knowledge base. The phase deviation set for each fault type refers to the set of three phase deviations obtained by subtracting the AB, BC, and AC phase difference deviations in the phase difference sequence from the standard value of the phase difference sequence in the standard waveform parameter template for the corresponding fault type. The construction of the area deviation set and the phase deviation set is performed sequentially on the standard waveform parameter templates corresponding to all fault types in the knowledge base, resulting in one area deviation set and one phase deviation set for each fault type.

[0129] It should be noted that the area deviation and phase deviation are constructed as independent sets instead of being directly merged, so that the composition of the final comprehensive deviation set can be flexibly determined in steps S322 and S323 based on whether the discontinuous phase deviation set is an empty set. When there are no discontinuous features, the combination of the area deviation set and the phase deviation set is directly used as the comprehensive deviation set to avoid introducing invalid parameter dimensions.

[0130] Step S322: When it is determined that the discontinuity phase set is not an empty set, construct the discontinuity duration of each phase in the discontinuity phase set and the corresponding discontinuity duration of each phase in each standard waveform parameter template as a discontinuity deviation set. Merge the area deviation set, phase deviation set and discontinuity deviation set to obtain the comprehensive deviation set of each fault type.

[0131] The discontinuity deviation set refers to the set of discontinuity duration deviations for each phase obtained by subtracting the discontinuity duration of each phase in the discontinuity phase set from the standard value of the corresponding phase discontinuity duration in the standard waveform parameter template for the corresponding fault type in the knowledge base. Merging the area deviation set, phase deviation set, and discontinuity deviation set involves summing all deviation values ​​from the three sets into a single set, which serves as the comprehensive deviation set for each fault type.

[0132] It should be noted that the discontinuity deviation set only includes the discontinuity duration deviation of phases with actual discontinuities in the discontinuity phase set. The discontinuity duration of non-discontinuous phases is zero and does not contribute to the matching result. This ensures that the composition of the comprehensive deviation set strictly corresponds to the actual waveform characteristics of the current fault waveform recording, and guarantees that the combined result of the three types of deviation sets can fully reflect the overall gap between the current waveform recording parameters and the standard parameters of each fault type.

[0133] Step S323: When the set of discontinuous phases is determined to be empty, the set of area deviation and the set of phase deviation are combined as the comprehensive deviation set of each fault type.

[0134] When the set of discontinuous phases is empty, none of the three phases exhibit waveform interruption characteristics, and the duration of the discontinuity has no ability to distinguish between all fault types. Therefore, no set of discontinuity deviations is constructed; instead, the combined set of area deviations and phase deviations is used as the comprehensive deviation set for each fault type. Steps S322 and S323 together constitute a complete coverage of the state judgment regarding whether the set of discontinuous phases is empty.

[0135] It should be noted that when there are no discontinuity features, the construction of the discontinuity deviation set is omitted, so that the number of dimensions of the comprehensive deviation set is consistent with the number of effective parameters of the current waveform. This avoids the dilution of the proportion of area deviation and phase deviation in the total amount of the comprehensive deviation set by zero-value discontinuity duration deviation, thereby ensuring that the determination of the target fault type is mainly driven by parameters with actual distinguishing ability.

[0136] Step S324: Determine the fault type with the smallest total total deviation set as the target fault type.

[0137] The total comprehensive deviation set is the sum of the absolute values ​​of all deviations in the comprehensive deviation sets for each fault type. It reflects the overall distance between the current waveform recording parameters and the standard waveform parameter template for a certain fault type. The smaller the total, the closer the current waveform recording parameters are to the standard parameters for that fault type. The fault type with the smallest total comprehensive deviation set is determined as the target fault type. This means selecting the fault type in the parameter space that is closest to the current waveform recording parameters as the matching result.

[0138] It should be noted that using the fault type corresponding to the minimum total value of the comprehensive deviation set as the target fault type provides objective numerical support for the matching results. Dispatchers can assess the degree of differentiation between candidate fault types by checking the total comprehensive deviation set of each fault type, rather than simply accepting the system's single conclusion. This improves the traceability of fault judgment results and is consistent with the traceability positioning of the procedure source information in the knowledge base.

[0139] In some embodiments, determining the fault type with the smallest total comprehensive deviation set as the target fault type in step S324 includes steps S3241 and S3242: Step S3241: When the minimum value of the total set of comprehensive deviations is less than or equal to the preset deviation threshold, the corresponding fault type is determined as the target fault type.

[0140] The preset deviation threshold is a critical value for the total amount of comprehensive deviations used to distinguish between reliable and questionable matching results. Those skilled in the art can set this threshold based on the parameter dispersion and historical matching accuracy of the standard waveform parameter templates for each fault type in the knowledge base. A smaller preset deviation threshold requires higher reliability of the matching results and increases the probability of triggering manual confirmation in step S3242. A larger preset deviation threshold covers a wider range of automatic matching, but also increases the risk of misjudgment between fault types with similar parameters. When the minimum value of the total amount of comprehensive deviations is less than or equal to the preset deviation threshold, it indicates that the overall distance between the current waveform parameters and the standard parameters of the corresponding fault type is within an acceptable range. The corresponding fault type is then directly identified as the target fault type and sent to step S4 to retrieve the corresponding knowledge unit.

[0141] It should be noted that step S3241 establishes a credibility threshold for the automatic matching results, so that the matching results are directly output as the target fault type only when the total deviation meets the standard. When the parameter matching quality is insufficient, the manual confirmation process in step S3242 is triggered, realizing a reasonable boundary division between automatic judgment and human-machine collaboration, and avoiding the output of low-credibility automatic judgment conclusions to the scheduling end in the case of poor waveform quality or new fault scenarios.

[0142] Step S3242: When the minimum value of the total comprehensive deviation set is greater than the preset deviation threshold, mark the normalized half-cycle area difference, discontinuity duration and phase difference sequence of each phase as parameters to be confirmed and send them to the scheduling expert terminal. Receive the fault type confirmation result returned by the scheduling expert terminal and use the fault type confirmation result as the target fault type.

[0143] If the minimum total value of the comprehensive deviation set exceeds the preset deviation threshold, it indicates that the overall distance between the current waveform parameters and the standard parameters for all fault types in the knowledge base exceeds the reliable range, and automatic matching cannot provide a reliable conclusion. Marking the normalized half-cycle area difference, discontinuity duration, and phase difference sequence of each phase as parameters to be confirmed involves packaging all parameter values ​​involved in the matching and sending them to the dispatch expert. This allows the dispatch expert to directly view the specific values ​​of the parameters to be confirmed and, combined with field operational experience, provide a fault type confirmation result. The parameters to be confirmed also include a ranking of the total comprehensive deviation set for each fault type, for the dispatch expert to refer to the tendencies of the automatic matching conclusion.

[0144] It should be noted that step S3242 embodies the core design of human-machine collaboration in this application. Issues with waveform quality, novel fault scenarios, or fault types not yet covered by the knowledge base may all lead to automatic matching failure. Handling such cases with the dispatch expert end, rather than forcibly outputting low-reliability automatic conclusions, ensures that the target fault type received by the dispatch end has a clearly defined responsible party in any scenario, meeting the requirements of traceability for decision-making in power grid dispatch operations. Steps S3241 and S3242 together constitute a complete coverage of the relationship between the minimum total value of the comprehensive deviation set and the preset deviation threshold.

[0145] In some embodiments, after receiving the fault type confirmation result returned by the scheduling expert in step S3242 and using the fault type confirmation result as the target fault type, the method further includes steps C1 and C2: Step C1: When it is determined that there is a standard waveform parameter template in the knowledge base that corresponds to the fault type confirmation result, the parameter to be confirmed is included in the historical waveform feature parameter set corresponding to the standard waveform parameter template. When the number of parameters in the historical waveform feature parameter set reaches the preset update threshold, the standard waveform parameter template is updated based on the historical waveform feature parameter set.

[0146] The historical waveform feature parameter set is a cumulative record of historical parameters maintained separately for each fault type in the knowledge base. Each time a scheduling expert confirms a parameter, the parameter to be confirmed is added to the historical waveform feature parameter set for the corresponding fault type. The preset update threshold is the minimum number of parameters in the historical waveform feature parameter set required to trigger an update of the standard waveform parameter template. Those skilled in the art can set this threshold based on the sample size required to statistically form a stable parameter estimate. When the number of parameters in the historical waveform feature parameter set reaches the preset update threshold, the standard value in the corresponding fault type's standard waveform parameter template is updated using the statistical mean of each parameter in the historical waveform feature parameter set, allowing the standard waveform parameter template to be continuously corrected with accumulated experience.

[0147] It should be noted that step C1 establishes an automatic update mechanism for the standard waveform parameter template of the knowledge base, enabling the knowledge base to continuously accumulate with the experience confirmation results from the scheduling expert end, gradually expanding the coverage of the standard waveform parameter template for actual waveform recording scenarios, reducing the frequency of manual confirmation in step S3242 due to deviations between the knowledge base parameters and the actual waveform, and reflecting the design goal of continuously expanding the knowledge base with the accumulation of experience.

[0148] Step C2: If the knowledge base does not contain a standard waveform parameter template corresponding to the fault type confirmation result, construct a new standard waveform parameter template with the parameter to be confirmed as the initial value, bind it with the corresponding knowledge unit, and store it in the knowledge base.

[0149] The absence of a standard waveform parameter template corresponding to the fault type confirmation result in the knowledge base means that the fault type confirmation result given by the scheduling expert has no corresponding entry in the existing fault type list of the knowledge base, indicating that the current fault type belongs to a new fault scenario not yet covered by the knowledge base. Constructing a new standard waveform parameter template using the parameters to be confirmed as initial values ​​involves using the normalized half-cycle area difference, discontinuity duration, and phase difference sequence of each phase in the parameters to be confirmed as the initial standard parameter values ​​for the new fault type, thus creating an initial entry for the new standard waveform parameter template. Binding it to the corresponding knowledge unit and storing it in the knowledge base involves establishing a relationship between the new standard waveform parameter template and the handling knowledge unit provided by the scheduling expert for this fault type, and then writing it into storage according to the unified field structure of the knowledge base, so that the new fault type can participate in the deviation calculation in step S321 during subsequent matching. Steps C1 and C2 together constitute a complete coverage of the state judgment regarding the existence of a corresponding standard waveform parameter template in the knowledge base.

[0150] It should be noted that step C2 enables the knowledge base to have the ability to self-expand to new fault scenarios. When a new fault first appears, it triggers confirmation by the scheduling expert and an initial entry can be established in the knowledge base. In step S32, the waveform recordings of similar faults will participate in the matching with the new standard waveform parameter template. After gradually accumulating the set of historical waveform feature parameters, the parameters will be updated through step C1, forming a complete knowledge accumulation path from the first identification to parameter stabilization.

[0151] Preferably, after step S3242 is completed, the source information of the fault waveform, the target fault type, the total sorting of the comprehensive deviation set, and the confirmation results of the scheduling expert can be written into the fault handling log of the scheduling terminal, so that each fault judgment process has a complete operation record, which can be used for subsequent procedure updates and training material compilation.

[0152] Preferably, after step S32 is completed, the difference between the minimum and the second smallest total value of the comprehensive deviation set in this match can be calculated. If the difference is lower than the minimum discrimination threshold determined by those skilled in the art based on the parameter discrimination ability, the second smallest fault type and the corresponding handling knowledge unit are additionally marked in the standardized answer in step S4 to remind the dispatcher to pay attention to the uncertainty of the matching result, which is connected with the triggering logic of the preset prompt interval in step S42.

[0153] Preferably, after step S3 is completed, the target fault type, the total amount of the comprehensive deviation set, and the parameter values ​​on which the matching is based can be cached in the intermediate result storage area of ​​this fault waveform analysis task, so that step S4 can directly call it when generating the standardized answer, avoiding repeated execution of matching calculations and reducing the consumption of computing resources when multiple fault waveform analysis tasks are processed concurrently at the scheduling end.

[0154] Step S4: Retrieve the corresponding knowledge unit from the knowledge base according to the target fault type, organize it into a standardized answer and output it to the scheduling terminal.

[0155] It should be noted that dispatchers need to confirm the handling plan within a short period of time after receiving a fault alarm. The knowledge units stored in the knowledge base contain fields such as execution steps, preconditions, applicable scenarios, risk warnings, and procedure sources, representing a structured accumulation of historical fault handling experience. Traditionally, dispatchers need to manually consult procedure documents, which can easily delay handling opportunities when the fault type is complex or multiple faults occur concurrently due to excessive document retrieval time. For example, if a shift worker receives a 220kV line trip alarm at night, they need to simultaneously check protection action records, confirm the circuit breaker location, and determine if reclosing conditions are met. If relying on manual procedure consultation, the total time spent executing each step sequentially may exceed the optimal time window for on-site handling. Step S4 uses the target fault type determined in step S3 as an index, directly retrieves the corresponding knowledge unit from the knowledge base, and organizes it into a standardized answer containing handling steps, risk warnings, and procedure sources. This allows dispatchers to obtain an executable handling plan without consulting documents, shortening the response time from fault identification to handling decision.

[0156] See Figure 4 , Figure 4 The overall storage and service architecture of the power grid dispatch knowledge base is demonstrated. Figure 4 The application service interface layer provides external interfaces for knowledge resource invocation, knowledge resource retrieval and recommendation, knowledge resource management, and knowledge graph management, enabling unified access and management of the knowledge base. The Neo4j-based knowledge base layer, grounded in the Neo4j core API and Cypher query language, ensures data consistency and query efficiency through mechanisms such as transaction management, lock management, transaction logs, page caching, and record files. Knowledge graph data is ultimately persisted to graph data disk storage. Multiple types of knowledge resources in the underlying storage device are structured and then written to the graph database. Step S4... Figure 4 The knowledge unit retrieval and answer organization are completed with the support of the knowledge resource retrieval interface shown. In some embodiments, step S4 includes steps S41 to S43: Step S41: Retrieve the corresponding knowledge unit from the knowledge base according to the target fault type, and add the target fault type, the total amount of the comprehensive deviation set, and the historical fault case number on which the matching is based to the standardized answer.

[0157] The knowledge unit is a structured entry in the knowledge base bound to the target fault type. It includes five fields: execution steps, preconditions, applicable scenarios, risk warnings, and procedure source. Retrieval is performed by searching the knowledge base for the corresponding entry using the target fault type identifier as the key and reading all field content. The standardized answer is the output structure of the five fields of the knowledge unit, organized according to the scheduling end's display standards. Scheduling personnel can directly obtain executable handling steps and risk warnings through the standardized answer. The historical fault case number is a unique identifier for the historical fault waveform record associated with this knowledge unit in the knowledge base. It can be a historical case number entered when the knowledge base was built, or a recent case number written after confirmation by scheduling experts in step C1. It is appended to the standardized answer for scheduling personnel to trace and match. The total comprehensive deviation set is appended to the standardized answer, allowing scheduling personnel to intuitively assess the reliability of the matching result.

[0158] It is easy to understand that by adding the target fault type, the total comprehensive deviation set, and the historical fault case number, the standardized answer is upgraded from a simple output of handling steps to a verifiable conclusion with complete traceability information. Before executing the handling steps, dispatchers can check the on-site situation of the reference case by the historical fault case number, and judge the similarity between the current match and the historical case by combining the total comprehensive deviation set value, so as to avoid blindly executing the handling steps when the reference basis is unclear.

[0159] It should be noted that step S41 establishes a complete mapping from the target fault type to the standardized answer. The five fields of the knowledge unit cover all the key information required by the dispatcher to perform the disposal operation, eliminating the need to consult additional documents outside the knowledge base. This concentrates the information source for disposal decisions within a single standardized answer, consistent with the design goal of the knowledge base to ensure traceability and verifiability of the answers. For the organization of resource metadata in the knowledge base, please refer to [link to relevant documentation]. Figure 3 , Figure 3This document showcases a metadata feature model for various types of knowledge resources, structured in a tree format and divided into five attribute categories. The general core attributes include title, knowledge point, knowledge point identifier, training resource identifier, description, keywords, status, total usage count, and initial rating, used for basic identification and retrieval of knowledge resources. Resource type attributes include resource type, interaction type, typical learning duration, difficulty, description, professional direction, and level, used to characterize the resource's formal features and applicable scenarios, supporting filtering and recommendation by type. Lifecycle attributes include uploading unit, production unit, copyright ownership, release time, and speaker, recording the complete source information of the resource from production to inclusion in the database, ensuring traceability and copyright compliance. Resource format attributes include format, file size, and storage location, used to support physical location and multi-format compatible access. Resource evaluation attributes include comprehensive evaluation results, content evaluation results, formal evaluation results, and data statistical evaluation results, reflecting the quality assessment of the resource after user feedback, providing a basis for dynamic optimization of the knowledge base. The above five types of attributes together constitute a complete metadata description framework for knowledge resources, enabling dispatchers to further access training videos or case documents corresponding to the target fault type after obtaining standardized answers, thereby extending from fault handling to experience learning.

[0160] Step S42: When the total amount of the comprehensive deviation set is within the preset prompt range, obtain the second smallest fault type of the comprehensive deviation set, integrate the handling steps of the knowledge unit corresponding to the second smallest fault type and the handling steps corresponding to the target fault type into the standardized answer, mark the total amount of the comprehensive deviation set and the difference description corresponding to each fault type in the standardized answer, and send a manual confirmation prompt to the dispatcher.

[0161] The preset prompt interval refers to the numerical range where the minimum value of the total comprehensive deviation set is less than or equal to the preset deviation threshold, but the difference between it and the second smallest value is lower than the minimum distinction threshold determined by a person skilled in the art based on the parameter distinction capability. This reflects a situation where the matching degree between the target fault type and the second smallest fault type is relatively close, and the automatic judgment has a certain degree of uncertainty. The second smallest fault type refers to the fault type ranked second in the total comprehensive deviation set. It can be a fault type with similar waveform characteristics to the target fault type. For example, the normalized half-cycle area difference values ​​of single-phase grounding fault and phase-to-phase short-circuit fault are relatively close under certain operating conditions. When the waveform quality is average, the total comprehensive deviation set values ​​of the two may not differ much. The difference description is a textual description of the key differences between the target fault type and the second smallest fault type in the handling steps. It is automatically generated by comparing the handling step fields of the two types of knowledge units in the knowledge base. For example, if there are differences between the two fault types in key operations such as whether reclosing is allowed or whether the faulty phase needs to be isolated, the differences will be marked one by one in the standardized answer.

[0162] Understandably, when there is uncertainty in the matching results, if only a single handling step corresponding to the target fault type is output to the dispatcher, the dispatcher will have no way of knowing that the current judgment is ambiguous. This could lead to the execution of the incorrect handling procedure when the lesser fault type is actually the true fault type. Integrating the handling steps for both fault types into a standardized answer with explanations of the differences allows dispatchers to compare the key differences between the two solutions before execution, making a final judgment based on the actual situation on site, thus reducing the risk of incorrect handling direction due to matching uncertainty.

[0163] It should be noted that step S42 introduces a manual confirmation prompt when the confidence level is insufficient, rather than forcibly outputting a single conclusion. This reflects the design principle of this application to ensure the safety of handling in the context of matching uncertainty through human-machine collaboration. It forms a tiered guarantee with the logic in step S3242 of sending the parameters to be confirmed to the scheduling expert when the preset deviation threshold is exceeded, covering different degrees of uncertainty from complete matching failure to ambiguity in the matching result.

[0164] Step S43: If the total amount of the comprehensive deviation set is not within the preset prompt range, output the standardized answer to the scheduling terminal.

[0165] The statement that the total comprehensive deviation set is not within the preset warning range means that the difference between the minimum and second minimum values ​​of the total comprehensive deviation set exceeds the minimum distinction threshold, indicating that there is a sufficient matching gap between the target fault type and other candidate fault types, and the automatic judgment result is reliable. In this case, the standardized answer obtained in step S41 is directly output to the scheduling terminal without triggering a manual confirmation prompt. The scheduling personnel then directly execute the handling steps in the knowledge unit corresponding to the target fault type. Steps S42 and S43 together constitute a complete coverage of the state judgment regarding whether the total comprehensive deviation set is within the preset warning range.

[0166] It should be noted that step S43 ensures that no additional manual confirmation burden is generated when the matching result is reliable, enabling dispatchers to directly obtain a definitive conclusion and quickly execute the handling operation in most common fault scenarios. The dual-scheme integration output of step S42 is only triggered in the case of matching ambiguity, so as to minimize the additional attention of dispatchers while ensuring the safety of the handling.

[0167] Preferably, after step S4 is completed, the entire process of this fault waveform analysis can be written to the fault handling log at the dispatching end. The record includes the source information of the fault waveform, the target fault type determined in step S3, the total amount of the comprehensive deviation set, the output timestamp of the standardized answer, and the final confirmation selection by the dispatcher when step S42 is triggered. The fault handling log can serve as a reference for subsequent procedure updates, training material compilation, and knowledge base maintenance, ensuring that each fault handling process has a complete operation record and meeting the requirements of power grid dispatching operations for traceability of handling decisions.

[0168] See Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. The electronic device 40 includes: a processor 41, a memory 42, and a computer program; wherein, The memory 42 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0169] The processor 41 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0170] Alternatively, the memory 42 can be either standalone or integrated with the processor 41.

[0171] When the memory 42 is a device independent of the processor 41, the device may further include: Bus 43 is used to connect the memory 42 and the processor 41.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for constructing a professional knowledge base and intelligently querying information for power grid dispatching, characterized in that: include: The fault waveforms uploaded from the dispatch terminal are processed into coordinates to identify the waveform profiles of each phase, including: Coordinate processing was performed with the lower left corner of the fault waveform as the origin to determine the time coordinates and amplitude coordinates of each pixel. For each phase channel, the number of pixels corresponding to each amplitude coordinate is counted. The amplitude coordinates where the number of pixels in each phase channel exceeds a preset density threshold are determined as the zero-value baseline for each phase, including: The number of pixels corresponding to each amplitude coordinate in each phase channel is counted to obtain the pixel distribution of each phase amplitude. When it is determined that there are two amplitude coordinates in the amplitude pixel distribution of each phase where the number of pixels both exceed the preset dense threshold, the middle amplitude coordinate of the two amplitude coordinates is taken as the zero value baseline of each phase. When it is determined that there is only one amplitude coordinate in the amplitude pixel distribution of each phase where the number of pixels exceeds a preset dense threshold, the amplitude coordinate is used as the zero value baseline of each phase; The sequence of pixels that continuously deviate from the zero baseline in each phase channel is identified as the waveform profile of each phase, resulting in the waveform profile of phase A, phase B, and phase C. The waveform profile of each phase is divided into positive and negative half cycles using the zero-value baseline corresponding to the waveform profile of each phase, and the positive and negative half cycle division results are obtained. Based on the positive and negative half-cycle segmentation results, the matching parameters of each phase are extracted, and the matching parameters are matched with the standard waveform parameter templates corresponding to each fault type in the knowledge base to obtain the target fault type. Based on the target fault type, the corresponding knowledge unit is retrieved from the knowledge base, organized into a standardized answer, and output to the scheduling end, including: Based on the target fault type, retrieve the corresponding knowledge unit from the knowledge base, and add the target fault type, the total amount of the comprehensive deviation set, and the historical fault case number on which the matching is based to the standardized answer; When the total amount of the comprehensive deviation set is determined to be within the preset prompt range, the fault type with the second smallest total amount of the comprehensive deviation set is obtained. The handling steps of the knowledge unit corresponding to the second smallest fault type and the handling steps corresponding to the target fault type are integrated into the standardized answer. The total amount of the comprehensive deviation set and the difference description corresponding to each fault type are marked in the standardized answer, and a manual confirmation prompt is sent to the dispatcher. When the total set of comprehensive deviations is not within the preset prompt range, the standardized answer will be output to the scheduling terminal.

2. The method according to claim 1, characterized in that, The extraction of matching parameters for each phase based on the positive and negative half-cycle segmentation results includes: Obtain all intersection points between the waveform profile of each phase and the zero baseline. Take the waveform profile segment above the zero baseline between two adjacent intersection points as the positive half-cycle profile and the waveform profile segment below the zero baseline as the negative half-cycle profile. When the waveform profile segment between two adjacent intersection points coincides with the zero baseline, the coincident segment is identified as a discontinuity segment. The time coordinate length of all discontinuities in the waveform profile of each phase is counted to obtain the discontinuity duration of each phase. Based on the positive half-cycle area enclosed by the positive half-cycle profile and the zero-value baseline, and the negative half-cycle area enclosed by the negative half-cycle profile and the zero-value baseline, combined with the theoretical area of ​​the whole cycle of each phase waveform profile, the normalized half-cycle area difference of each phase is obtained, where the theoretical area of ​​the whole cycle is determined by the time coordinate span and amplitude coordinate peak value of each phase waveform profile. The time coordinates of the first positive intersection point of the waveform profiles of phases A, B, and C with the zero baseline are subtracted pairwise to obtain the phase difference AB, phase difference BC, and phase difference AC. The phase difference AB, phase difference BC, and phase difference AC are then compared with the standard phase differences pre-stored in the knowledge base to obtain the phase difference sequence. The normalized half-cycle area difference of each phase, the discontinuity duration of each phase, and the phase difference sequence are used as matching parameters.

3. The method according to claim 2, characterized in that, The time coordinate lengths of all discontinuous segments in the waveform profiles of each phase are statistically analyzed to obtain the discontinuity duration of each phase, including: The number of intermediate segments in the waveform profile of each phase is counted to obtain the number of segments between each phase; When the number of discontinuities in any phase is greater than one, the time coordinate lengths of all discontinuities in each phase are summed to obtain the discontinuity duration of each phase. When the number of discontinuities in any phase is equal to one, the time coordinate length of the unique discontinuity in each phase is taken as the discontinuity duration of each phase.

4. The method according to claim 3, characterized in that, The process of counting the number of intermediate segments in the waveform profile of each phase to obtain the number of interphase segments includes: By counting the number of discontinuous segments greater than zero, we obtain the set of discontinuous phases; When the set of discontinuous phases is determined to be empty, it is determined that there are no discontinuous segments in the waveform profile of each phase, and the normalized half-circle area difference and phase difference sequence of each phase are used as the matching basis. When determining that the discontinuous phase set is not empty, the discontinuity duration of each phase in the discontinuous phase set, the normalized half-circuit area difference of each phase, and the phase difference sequence are used together as the matching criteria.

5. The method according to claim 4, characterized in that, The step of matching the matching parameters with the standard waveform parameter templates corresponding to each fault type in the knowledge base to obtain the target fault type includes: Based on the normalized half-cycle area difference of each phase and the corresponding parameters in each standard waveform parameter template, the area deviation set of each fault type is obtained; based on the phase difference sequence and the corresponding parameters in each standard waveform parameter template, the phase deviation set of each fault type is obtained. When it is determined that the set of discontinuous phases is not empty, the discontinuity duration of each phase in the set of discontinuous phases and the corresponding discontinuity duration of each phase in each standard waveform parameter template are used to construct a set of discontinuity deviations. The set of area deviations, the set of phase deviations and the set of discontinuity deviations are merged to obtain the comprehensive deviation set of each fault type. When determining that the set of discontinuous phases is empty, the area deviation set and the phase deviation set are combined as the comprehensive deviation set for each fault type. The fault type with the smallest total set of comprehensive deviations is identified as the target fault type.

6. The method according to claim 5, characterized in that, The step of determining the fault type with the smallest total comprehensive deviation set as the target fault type includes: When the minimum value of the total set of comprehensive deviations is less than or equal to the preset deviation threshold, the corresponding fault type is determined as the target fault type. When the minimum value of the total set of comprehensive deviations is greater than the preset deviation threshold, the normalized half-cycle area difference, discontinuity duration and phase difference sequence of each phase are marked as parameters to be confirmed and sent to the scheduling expert terminal. The fault type confirmation result returned by the scheduling expert terminal is received and the fault type confirmation result is used as the target fault type.

7. The method according to claim 6, characterized in that, After receiving the fault type confirmation result returned by the scheduling expert, and using the fault type confirmation result as the target fault type, the method further includes: When it is determined that a standard waveform parameter template corresponding to the fault type confirmation result exists in the knowledge base, the parameter to be confirmed is included in the historical waveform feature parameter set corresponding to the standard waveform parameter template. When the number of parameters in the historical waveform feature parameter set reaches the preset update threshold, the standard waveform parameter template is updated based on the historical waveform feature parameter set. If it is determined that there is no standard waveform parameter template in the knowledge base that corresponds to the fault type confirmation result, a new standard waveform parameter template is constructed with the parameter to be confirmed as the initial value, and then stored in the knowledge base after being bound to the corresponding knowledge unit.

8. The method according to claim 1, characterized in that, The process of performing coordinate processing on the fault waveform uploaded by the dispatch terminal to identify the waveform contours of each phase also includes: When it is determined that the fault waveform diagram contains voltage and current channels, the waveform profiles of each phase corresponding to the voltage and current channels are identified respectively to obtain voltage waveform profile groups and current waveform profile groups. The voltage waveform profile group and the current waveform profile group are divided into positive and negative half-cycles, and the normalized half-cycle area difference, discontinuity duration and phase difference sequence are extracted to obtain the voltage matching parameter group and the current matching parameter group. Obtain the normalized positive half-cycle area ratio of the voltage waveform profile group and the current waveform profile group under the same phase. Based on the normalized positive half-cycle area ratio of the voltage waveform profile group and the normalized positive half-cycle area ratio of the current waveform profile group, obtain the voltage and current positive half-cycle area ratio deviation of each phase. Determine the phases with voltage and current positive half-cycle area ratio deviations greater than a preset deviation threshold as abnormal phases, and obtain the abnormal phase set. The voltage matching parameter group and the current matching parameter group are used to match the standard waveform parameter templates corresponding to each fault type in the knowledge base.

9. The method according to claim 8, characterized in that, After determining the phases with voltage and current positive half-cycle area ratio deviations greater than a preset deviation threshold as abnormal phases and obtaining the abnormal phase set, the method further includes: When it is determined that the abnormal phase set is not empty, the number of abnormal phases and the identifier of each abnormal phase in the abnormal phase set are used as additional parameters, which, together with the voltage matching parameter group and the current matching parameter group, participate in the matching with the standard waveform parameter templates corresponding to each fault type in the knowledge base. After the matching is completed, when it is confirmed that the number of abnormal phases and the identifier of each abnormal phase are consistent with the pre-stored standard waveform parameter template corresponding to the target fault type, the total amount of the comprehensive deviation set obtained by the matching is adjusted to the preset confidence interval to obtain the total amount of the comprehensive deviation set after confidence adjustment. The total set of comprehensive deviations after confidence adjustment is used to determine the target fault type; When the abnormal phase set is determined to be empty, only the voltage matching parameter group and the current matching parameter group are used to match the standard waveform parameter templates corresponding to each fault type in the knowledge base.

10. The method according to claim 2, characterized in that, The step involves comparing the AB phase difference, BC phase difference, and AC phase difference with the pre-stored standard phase differences in the knowledge base to obtain a phase difference sequence, including: When any phase deviation exceeds a preset phase threshold, the corresponding phase is marked as an abnormal phase, and the abnormal phase identifier is included in the phase difference sequence as an additional dimension recorded in the phase difference sequence. When all phase deviations are determined to be within the preset phase threshold, the phase difference sequence is marked as symmetrical and normal, and no phase abnormality identifier is generated.

Images