CN-121981227-A - Method and device for generating harmonic knowledge graph of power distribution network

CN121981227ACN 121981227 ACN121981227 ACN 121981227ACN-121981227-A

Abstract

The application discloses a method and a device for generating a harmonic knowledge graph of a power distribution network, belonging to the field of operation and maintenance of power systems, wherein the method comprises the steps of carrying out knowledge extraction on topology structure data of the power distribution network and power grid ledger data to generate a plurality of text triples; performing discrete Fourier transform on waveform data of operation equipment in a power distribution network to obtain a complex frequency spectrum in the operation state of the power distribution network, calculating to obtain power spectrum density according to the complex frequency spectrum, determining harmonic thresholds at all frequencies according to the power spectrum density, generating a time-frequency spectrogram, identifying energy characteristic peaks exceeding the corresponding harmonic thresholds in the time-frequency spectrogram as harmonic components, determining operation equipment information, frequency and phase information corresponding to the harmonic components to perform knowledge extraction, generating a plurality of harmonic triplets, and generating a power distribution network harmonic knowledge spectrogram according to the harmonic triplets and the text triplets. By implementing the method and the device, the problem of low efficiency of the related query of the harmonic information and the running equipment information in the prior art can be solved.

Inventors

LI HAOBIN
CHEN ZHIWEI
ZHAO RUIFENG
LAN TIAN
WANG CHEN
ZHONG WEI

Assignees

广东电网有限责任公司电力调度控制中心

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (10)

1. The method for generating the harmonic knowledge graph of the power distribution network is characterized by comprising the following steps of: acquiring topological structure data and power grid account data of a power distribution network and waveform data of running equipment in the power distribution network; knowledge extraction is carried out on the topological structure data and the grid standing book data, and a plurality of text triples are generated; performing discrete Fourier transform on the waveform data to obtain a complex frequency spectrum in the running state of the power distribution network, and calculating to obtain power spectrum density according to the complex frequency spectrum; According to the power spectrum density, determining a harmonic threshold under each frequency, and generating a time-frequency spectrogram; Identifying energy characteristic wave peaks exceeding corresponding harmonic thresholds in the time-frequency spectrogram as harmonic components, and determining operation equipment information, frequency and phase information corresponding to each harmonic component; knowledge extraction is carried out on the operation equipment information, the frequency information and the phase information corresponding to each harmonic component, and a plurality of harmonic triples are generated; and generating a harmonic knowledge graph of the power distribution network according to the harmonic triplets and the text triplets.
2. The method for generating a harmonic knowledge graph of a power distribution network according to claim 1, further comprising, before acquiring waveform data of an operating device in the power distribution network: Acquiring original waveform data of operation equipment in a power distribution network; Framing the original waveform data to obtain spare waveform data after framing; Windowing is carried out on the standby waveform data subjected to framing processing, and the standby waveform data subjected to windowing is obtained; and taking the standby waveform data subjected to the windowing treatment as the waveform data of the running equipment in the power distribution network.
3. The method for generating a harmonic knowledge graph of a power distribution network according to claim 1, wherein the step of performing knowledge extraction on the topology data and the grid ledger data to generate a plurality of text triples comprises: Carrying out knowledge extraction on the topological structure data and the grid ledger data through a deep learning model for named entity recognition to generate a plurality of text triples; The knowledge extraction is performed on the topology structure data and the grid ledger data through a deep learning model for named entity recognition, and a plurality of text triples are generated, including: according to the BERT layer in the deep learning model, word segmentation is carried out on the topological structure data and the grid ledger data to obtain a plurality of word segmentation results, and a bidirectional coding mechanism is adopted to generate a high-dimensional dense vector sequence containing context semantic information according to each word segmentation result; Extracting sequence features of the high-dimensional dense vector sequence according to BiLSTM layers in the deep learning model to generate a feature vector sequence fused with bidirectional semantics; and marking entity relations of the feature vector sequences according to a CRF layer in the deep learning model, and generating a plurality of text triples.
4. The method for generating a harmonic knowledge graph of a power distribution network according to claim 1, wherein the knowledge extraction is performed on the operation equipment information, the frequency and the phase information corresponding to each harmonic component to generate a plurality of harmonic triples, including: For each harmonic component, forming a harmonic triplet according to the operation equipment information, frequency and phase information corresponding to the harmonic component; And obtaining harmonic triples corresponding to the harmonic components.
5. The method for generating a harmonic knowledge graph of a power distribution network according to claim 1, wherein the generating the harmonic knowledge graph of the power distribution network according to the harmonic triplet and the text triplet comprises: Calculating a first word vector representation of each of the harmonic triples, and calculating a second word vector representation of each of the text triples; For each harmonic triplet, calculating cosine similarity between a first word vector representation and each second word vector representation of the current harmonic triplet; taking the text triplet corresponding to the second word vector representation with cosine similarity exceeding the preset threshold value as a target text triplet; connecting the current harmonic triplet with each target text triplet to generate a plurality of entity association nodes containing the current harmonic triplet; and integrating all generated entity association nodes to form the harmonic knowledge graph of the power distribution network.
6. The utility model provides a power distribution network harmonic knowledge graph generation device which characterized in that includes: The data acquisition module is used for acquiring topological structure data and power grid account data of the power distribution network and waveform data of running equipment in the power distribution network; the text triplet generation module is used for carrying out knowledge extraction on the topological structure data and the grid ledger data to generate a plurality of text triples; The power spectrum density calculation module is used for performing discrete Fourier transform on the waveform data to obtain a complex frequency spectrum in the running state of the power distribution network, and calculating the power spectrum density according to the complex frequency spectrum; the spectrogram generation module is used for determining harmonic wave threshold values under each frequency according to the power spectral density and generating a time-frequency spectrogram; The harmonic information generation module is used for identifying energy characteristic wave peaks exceeding a corresponding harmonic threshold in the time-frequency spectrogram as harmonic components and determining operation equipment information, frequency and phase information corresponding to each harmonic component; the harmonic triplet generation module is used for carrying out knowledge extraction on the operation equipment information, the frequency and the phase information corresponding to each harmonic component to generate a plurality of harmonic triples; And the knowledge graph generation module is used for generating a harmonic knowledge graph of the power distribution network according to the harmonic triplet and the text triplet.
7. The harmonic knowledge graph generation device of the power distribution network, as set forth in claim 6, further comprising a waveform data processing module; The waveform data processing module is used for acquiring original waveform data of operation equipment in the power distribution network; Framing the original waveform data to obtain spare waveform data after framing; Windowing is carried out on the standby waveform data subjected to framing processing, and the standby waveform data subjected to windowing is obtained; and taking the standby waveform data subjected to the windowing treatment as the waveform data of the running equipment in the power distribution network.
8. The power distribution network harmonic knowledge graph generation device according to claim 6, wherein the text triplet generation module is specifically configured to: Carrying out knowledge extraction on the topological structure data and the grid ledger data through a deep learning model for named entity recognition to generate a plurality of text triples; The knowledge extraction is performed on the topology structure data and the grid ledger data through a deep learning model for named entity recognition, and a plurality of text triples are generated, including: according to the BERT layer in the deep learning model, word segmentation is carried out on the topological structure data and the grid ledger data to obtain a plurality of word segmentation results, and a bidirectional coding mechanism is adopted to generate a high-dimensional dense vector sequence containing context semantic information according to each word segmentation result; Extracting sequence features of the high-dimensional dense vector sequence according to BiLSTM layers in the deep learning model to generate a feature vector sequence fused with bidirectional semantics; and marking entity relations of the feature vector sequences according to a CRF layer in the deep learning model, and generating a plurality of text triples.
9. The power distribution network harmonic knowledge graph generation device according to claim 6, wherein the harmonic triplet generation module is specifically configured to: For each harmonic component, forming a harmonic triplet according to the operation equipment information, frequency and phase information corresponding to the harmonic component; And obtaining harmonic triples corresponding to the harmonic components.
10. The harmonic knowledge graph generation device of claim 6, wherein the knowledge graph generation module is specifically configured to: Calculating a first word vector representation of each of the harmonic triples, and calculating a second word vector representation of each of the text triples; For each harmonic triplet, calculating cosine similarity between a first word vector representation and each second word vector representation of the current harmonic triplet; taking the text triplet corresponding to the second word vector representation with cosine similarity exceeding the preset threshold value as a target text triplet; connecting the current harmonic triplet with each target text triplet to generate a plurality of entity association nodes containing the current harmonic triplet; and integrating all generated entity association nodes to form the harmonic knowledge graph of the power distribution network.

Description

Method and device for generating harmonic knowledge graph of power distribution network Technical Field The invention relates to the field of operation and maintenance of power systems, in particular to a method and a device for generating a harmonic knowledge graph of a power distribution network. Background With the rapid development of the power system to the high voltage, large capacity and intelligent direction, the power grid scale is continuously enlarged, and the multi-type knowledge of the power distribution network is explosively increased, wherein the multi-type knowledge comprises unstructured text description data and time-sequence waveform data. In the operation and maintenance management of a power grid, particularly in a harmonic analysis scene, the efficiency of information query directly influences the timeliness of operation and maintenance decisions, but the current harmonic analysis scene still has the problem of difficult information query, and aiming at the defect of the fusion query capability of text data and waveform data, the device related information of corresponding operation devices cannot be quickly associated through harmonic features, so that the harmonic analysis result is difficult to support subsequent device fault analysis. Disclosure of Invention The invention provides a method and a device for generating a harmonic knowledge graph of a power distribution network, and the method can solve the problem that in the prior art, the related query efficiency of harmonic information and operation equipment information is low. In order to solve the above technical problems, an embodiment of the present invention provides a method for generating a harmonic knowledge graph of a power distribution network, including: acquiring topological structure data and power grid account data of a power distribution network and waveform data of running equipment in the power distribution network; knowledge extraction is carried out on the topological structure data and the grid standing book data, and a plurality of text triples are generated; performing discrete Fourier transform on the waveform data to obtain a complex frequency spectrum in the running state of the power distribution network, and calculating to obtain power spectrum density according to the complex frequency spectrum; According to the power spectrum density, determining a harmonic threshold under each frequency, and generating a time-frequency spectrogram; Identifying energy characteristic wave peaks exceeding corresponding harmonic thresholds in the time-frequency spectrogram as harmonic components, and determining operation equipment information, frequency and phase information corresponding to each harmonic component; knowledge extraction is carried out on the operation equipment information, the frequency information and the phase information corresponding to each harmonic component, and a plurality of harmonic triples are generated; and generating a harmonic knowledge graph of the power distribution network according to the harmonic triplets and the text triplets. Further, before acquiring the waveform data of the running equipment in the power distribution network, the method further comprises: Acquiring original waveform data of operation equipment in a power distribution network; Framing the original waveform data to obtain spare waveform data after framing; Windowing is carried out on the standby waveform data subjected to framing processing, and the standby waveform data subjected to windowing is obtained; and taking the standby waveform data subjected to the windowing treatment as the waveform data of the running equipment in the power distribution network. Further, the knowledge extraction is performed on the topology structure data and the grid ledger data to generate a plurality of text triples, including: Carrying out knowledge extraction on the topological structure data and the grid ledger data through a deep learning model for named entity recognition to generate a plurality of text triples; The knowledge extraction is performed on the topology structure data and the grid ledger data through a deep learning model for named entity recognition, and a plurality of text triples are generated, including: according to the BERT layer in the deep learning model, word segmentation is carried out on the topological structure data and the grid ledger data to obtain a plurality of word segmentation results, and a bidirectional coding mechanism is adopted to generate a high-dimensional dense vector sequence containing context semantic information according to each word segmentation result; Extracting sequence features of the high-dimensional dense vector sequence according to BiLSTM layers in the deep learning model to generate a feature vector sequence fused with bidirectional semantics; and marking entity relations of the feature vector sequences according to a CRF layer in the deep learning model, and generating a p