Frequency stop monitoring and early warning first-aid repair method based on real-time measurement data

Abstract

The invention belongs to the technical field of power distribution network power failure early warning and first-aid repair methods, and particularly relates to a frequency stop monitoring and early warning and first-aid repair method based on real-time measurement data. Comprising the following steps: establishing a multi-source data fusion processing mechanism, performing standardized fusion processing on multi-source data, and performing fusion mutual correction by using the multi-source data; performing power failure range analysis based on the multi-source real-time measurement data; power failure planning operation and existing operation are matched and planned; studying and judging the reset data based on the real-time signal; pushing a first-aid repair work order based on the research and judgment result and the real-time data; repairing and filing are completed; according to the power failure early warning first-aid repair method based on the real-time measurement data, an effective distribution network low-voltage fault comprehensive research and judgment scheme of multi-source data construction fusion mutual correction is established, the comprehensive research and judgment accuracy of the power failure event reaches 80% or above through local field verification display, the practical level is high, and good popularization and development prospects are achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of distribution network frequency outage monitoring and early warning repair methods, and in particular relates to a frequency outage monitoring and early warning repair method based on real-time measurement data. Background Art

[0002] According to the key task requirements formulated by State Grid Corporation of China, all power supply and distribution companies should fully promote "grid-based" active repair, give full play to the role of power supply service command platform, optimize and mobilize repair service resources, and restore power to customers quickly. At present, the HPLC area terminals in major power supply areas in China have achieved a large coverage, but due to the large power supply area, large user base, and large number of low-voltage meters, the effective coverage of the active reporting capability of frequent power outages in the HPLC area is limited. In addition, due to factors such as complex communication environment, there are a large number of omissions and false alarms (signal jitter, marketing replacement, etc.) in the frequent power outages of low-voltage users. Although the power supply and service systems of power supply companies in various provinces have built relevant functions for frequent power outage analysis and judgment, and realized the overall process of active repair dispatch, the current daily push work order volume is small, and there is a large gap with the number of frequent power outages of low-voltage residents from actual repair sources, and there is a low accuracy rate, which makes it difficult to promote and apply related modules. At present, the actual source of low-voltage repair work is still based on hotlines such as 95598. Therefore, there is an urgent need for the informatization of the perception capability of frequent power outages. Summary of the invention

[0003] The purpose of the present invention is to provide a method for monitoring and warning emergency repair based on real-time measurement data, which can improve the comprehensiveness and efficiency of low-voltage emergency repair work and provide more accurate and timely decision support for emergency repair work, in order to meet the needs of the power supply and distribution system for the collection, analysis and emergency repair of frequent power outages.

[0004] To achieve the above purpose, the present invention adopts the following technical solution.

[0005] A method for monitoring and early warning emergency repair of frequent outages based on real-time measurement data comprises the following steps:

[0006] Step 1: Establish a multi-source data fusion processing mechanism, perform standardized fusion processing on multi-source data, and use multi-source data for fusion and mutual verification; including:

[0007] Obtain real-time data on frequent power outages in the distribution network, standardize the data, extract and analyze the characteristics of the distribution network data based on the time series characteristics of the distribution network data, and verify multi-source data;

[0008] Step 2: Analyze the scope of frequent power outages based on multi-source real-time measurement data, including:

[0009] Receive distribution network fault data from the distribution automation system or the power consumption information collection system based on the data center; call the corresponding fault location algorithm for location according to different fault data; analyze the scope of frequent power outages after locating the distribution network fault;

[0010] Step 3: Plan for frequent power outages and match existing operations, including:

[0011] The known planned maintenance work of the PMS system is used to form a planned information pool, and the frequent power outage data is matched with the planned information pool data to determine whether it is a known plan. If it is a known plan, no work order is issued; if it is not a known frequent power outage, go to step 4;

[0012] Step 4: Analyze the reset data based on the real-time signal, including:

[0013] Apply power restoration monitoring to filter active emergency repair data and obtain short-term power outage signals. Within 5 minutes, monitor these signals to determine whether they are short-term power restoration data. If they are short-term power restoration data, no active work order will be pushed; if they are not short-term frequent power outage data, proceed to step 5.

[0014] Step 5: Push emergency repair work orders based on the analysis results and real-time data, including:

[0015] Apply real-time information verification to filter active repair data; perform data verification on power failure signals, and analyze whether the frequent power outage data is real based on the verified judgment results. If it is not real frequent power outage data, do not push active work orders; if it is real frequent power outage data, push active work orders and go to step 6;

[0016] Step 6: Complete emergency repair and archiving work, including:

[0017] Generate proactive repair work orders on the distribution and emergency platform and automatically push them to the "Accepted" module of the repair process management. The service agent will manually assign a repair team to dispatch the order, and the repair team will handle it through the order acceptance, return and archiving processes.

[0018] A further improvement or specific implementation method of the above-mentioned frequency outage monitoring and early warning repair method based on real-time measurement data, wherein the step one further comprises the following steps: The step one specifically comprises:

[0019] A1. According to the preset distribution network data standard specifications, the data specifications, dimensions, data types, data magnitudes and other attributes of the corresponding multi-source data are transformed and integrated to obtain basic data, and pre-processed using the cloud platform fusion service component;

[0020] A2. For the preprocessed data, select a suitable feature extraction model for feature extraction fusion according to the data structure characteristics of the main data;

[0021] A3. Use the reconstruction model to reconstruct the abnormal data and analyze the difference between it and the original data to verify whether the data is normal.

[0022] A further improvement or specific implementation method of the aforementioned frequency outage monitoring and early warning repair method based on real-time measurement data, the feature extraction fusion in step A2 includes: obtaining data structure feature attributes based on a dynamic graph attention network model combined with the structural relationship of the data; establishing a time series feature machine learning network based on a fully connected network to obtain data time series feature attributes; creating a fusion feature based on feature splicing and position encoding to obtain a fusion feature attribute.

[0023] A further improvement or specific implementation of the above-mentioned frequency outage monitoring and early warning repair method based on real-time measurement data, the feature extraction and fusion in step A2 specifically includes:

[0024] A21. Based on the dynamic graph attention network model and the structural relationship of the data, the data structure feature attributes are obtained. Specifically, the original data is vectorized, and a network structure model containing all the data is established based on the data dependency relationship, where each data variable constitutes each node in the spatial model. Based on the dynamic graph attention network model, an embedding layer is created to learn the data structure features.

[0025] A22. Based on the fully connected network, a time series feature machine learning network is established to obtain the time series feature attributes of the data. Specifically, based on the time series relationship of the data, a suitable time window is determined according to the power supply and distribution cycle of the distribution network. For the data {x 1 ,x 2 ......x w}, through the fully connected network Q, the data x 1 Convert to the factor vector Q x =W Q x 1 +q, and at the same time, other data x i , i∈2~w is converted into a solution vector P through a fully connected network P i =W P x j +p, optimize the network through machine learning, and use the activation function to get the data x 1 With data x i The attention weights are: The data x under the time scale is obtained by weighted aggregation i For data x 1 A fusion of influences Get data {x1 ,x 2 ......x w} Consider the temporal feature expression {x 1 ',x' 2 ......x' w};

[0026] A23. Create fused features based on feature concatenation and position encoding to obtain fused feature attributes. Specifically: establish a position encoding network, express the structural feature parameters of the data in vector form through position embedding to form a structural feature vector, express the time series features in the form of a time series feature vector, and concatenate and aggregate the structural feature vector and the time series feature vector to obtain the full attribute feature.

[0027] As a further improvement or specific implementation of the above-mentioned frequency outage monitoring and early warning repair method based on real-time measurement data, the step A2 also includes: using a plurality of layers of perceptron network models to perform dimensionality reduction processing on the embedded initial structural features.

[0028] A further improvement or specific implementation of the above-mentioned frequency outage monitoring and early warning repair method based on real-time measurement data, wherein step A3 specifically includes:

[0029] A31. Create a variational autoencoder network model (VAE), train the variational autoencoder network model using all attribute features corresponding to historical data, and extract the real data features of normal data;

[0030] A32. The trained autoencoder network model (VAE) will be used to perform deviation verification on all attribute features of real-time data and filter out abnormal data.

[0031] A further improvement or specific implementation method of the above-mentioned frequency outage monitoring and early warning repair method based on real-time measurement data, the input of the autoencoder network model is a vector set consisting of all attribute features obtained by splicing and aggregation, and the input is fitted by the encoder of the autoencoder network model so that it has a normal distribution mean and variance, and then the fitting data of the above-mentioned normal distribution is sampled by selecting appropriate sampling parameters through machine learning to obtain encoded data, and the encoded data is decoded by the decoder of the autoencoder network model to obtain a reconstructed vector set, which is the real data feature expression.

[0032] A further improvement or specific implementation method of the above-mentioned frequency outage monitoring and early warning repair method based on real-time measurement data, wherein the step 2 specifically includes:

[0033] B1. Extract and fuse the data features of the collected real-time distribution fault data and screen the abnormal data based on step 1, determine the abnormal data, repair or eliminate the abnormal data through field investigation or data processing, and obtain effective distribution network fault data;

[0034] B2. Extract and classify effective distribution network fault data, call corresponding fault location algorithms according to different fault data, and determine the faulty equipment and fault type;

[0035] B3. Determine the faulty equipment that is subject to frequent power outages according to the frequent power outage specifications of the distribution network, and search and determine the scope of frequent power outages by combining the faulty equipment and fault type information with the distribution network topology model.

[0036] A further improvement or specific implementation of the above-mentioned frequency outage monitoring and early warning repair method based on real-time measurement data, wherein step B3 specifically includes:

[0037] B31. Based on the determined faulty equipment and fault type, and in combination with the frequent power outage specifications of the distribution network, determine the faulty equipment information that is to be frequently powered off;

[0038] B32. Determine the scope of frequent power outages by searching based on the information of the equipment that is expected to be frequently powered outages in combination with the distribution network topology model, specifically including: performing a single-line search in the distribution network topology model with the power equipment that is expected to be frequently powered outages as the search origin to obtain the power supply node that supplies power to the power equipment and all distribution transformers in the search path; performing a multi-line search in parallel in the distribution network topology model with the power supply equipment that is expected to be frequently powered outages as the search origin to obtain all encountered transformers.

[0039] A further improvement or specific implementation method of the above-mentioned frequency outage monitoring and early warning repair method based on real-time measurement data, the single-line search includes: for the bus on the search path, comparing whether the bus voltage is consistent with the voltage level of the attached substation, if they are consistent, continuing the search until any distribution transformer is found, if they are inconsistent, stopping the search and treating the bus front-stage equipment as a power supply node; for the circuit breaker on the search path, retrieving whether it is a closing device determined based on the frequent power outage specification of the distribution network, if so, continuing the search until any distribution transformer is found, if not, stopping the search;

[0040] The multi-line search includes: for the busbars on the search path, comparing the busbar voltage with the voltage level of the attached substation to see if they are consistent, if they are consistent, continuing the search until any distribution transformer is found, if they are inconsistent, stopping the search.

[0041] Its beneficial effects are:

[0042] This application has established an effective multi-source data construction, fusion and mutual verification distribution network low-voltage fault comprehensive analysis and judgment scheme based on the frequency outage monitoring and early warning repair method of real-time measurement data. The local field verification shows that the comprehensive analysis and judgment accuracy of frequent power outage events is more than 80%, which has a high level of practicality and good promotion and development prospects.

[0043] The solution of this application improves the comprehensiveness and efficiency of low-voltage emergency repair work, enhances the work order control, fault management, and proactive emergency repair capabilities for single meter box events, helps to more comprehensively grasp the low-voltage emergency repair work order data, and provides more accurate and timely decision support for emergency repair work; the solution of this application fully considers the current planned work content, effectively eliminates invalid work orders generated by planned work such as low-voltage meter replacement, thereby ensuring the accuracy of proactive work order dispatch and improving overall operational efficiency;

[0044] The technical solution based on this application provides an effective way to realize the sharing of distribution network data and make full use of the construction achievements of existing data centers and real-time measurement centers. At the same time, it realizes the collection and integration of frequent power outage signals on the low-voltage side, and can efficiently and conveniently realize data intercommunication with existing data management systems.

[0045] This application relies on real-time measurement data, and completes real-time and standardized access to multi-source measurement information by extracting and applying data features, avoiding problems such as chaotic access channels and uneven quality. Through multi-source data fusion and mutual verification, the validity of signal data is enhanced, providing a data base support for fault diagnosis, and at the same time providing technical support for further analysis of equipment abnormalities for measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of a frequency outage monitoring and early warning repair method based on real-time measurement data. DETAILED DESCRIPTION

[0047] The present invention is described in detail below in conjunction with specific embodiments.

[0048] The present application discloses a method for monitoring and early warning repair of power outages based on real-time measurement data, which is mainly used to meet the needs of distribution network companies to realize the transformation of the emergency repair mode of low-voltage faults in distribution networks from "manual passive" to "intelligent active", and makes full use of fault information fusion and mutual verification as well as cleaning and improvement of fault information to realize early warning analysis of frequent power outage fault (frequent power outage fault) data, and at the same time optimizes the dispatching of active emergency repair work orders and the emergency repair process, effectively improving the timeliness of discovering and diagnosing low-voltage frequent outages, and providing technical support for active emergency repair work.

[0049] The steps include:

[0050] Step 1: Establish a multi-source data fusion processing mechanism, perform standardized fusion processing on multi-source data, and use multi-source data for fusion and mutual verification; including: obtaining real-time data of frequent power outages in the distribution network, standardizing the data, extracting and analyzing the characteristics of the distribution network data based on the time series characteristics of the distribution network data, and verifying the multi-source data; specifically including steps A1 to A3:

[0051] A1. According to the preset distribution network data standard specifications, the data specifications, dimensions, data types, data magnitudes and other attributes of the corresponding multi-source data are transformed and integrated to obtain basic data, and pre-processed using the cloud platform fusion service component;

[0052] The aforementioned transformation integration mainly includes data form, data magnitude, graphic data extraction, chart data conversion and other contents, which mainly converts the original measurement data with different expressions into a unified expression form, which can be completed by manual processing, program conversion and other methods. In order to achieve the best data processing efficiency, this application mainly relies on the existing cloud data center service modules within each power distribution company to complete;

[0053] A2. For the preprocessed data, select a suitable feature extraction model for feature extraction fusion according to the data structure characteristics of the main data; specifically including A21~A23:

[0054] A21. Based on the dynamic graph attention network model and the structural relationship of the data, the data structure feature attributes are obtained. Specifically, the original data is vectorized, and a network structure model containing all the data is established based on the data dependency relationship, where each data variable constitutes each node in the spatial model. Based on the dynamic graph attention network model, an embedding layer is created to learn the data structure features.

[0055] The dynamic graph attention network model (GAT) is used for the attention mechanism to realize the local structural information between each data and extract the structural features between the data. Here, the structural features refer to the dependency and influence relationship of the data in the generalized space. In the actual implementation process, multiple graph neural network layers can be further created through the machine learning platform to further complete the compression and extraction of data structure features, so as to effectively obtain global data features;

[0056] In a stable and mature power supply and distribution area, the power consumption data of the power supply and distribution end and the power consumption end have periodic characteristics. In a complete power supply and distribution cycle, the power supply and power consumption measurement data are relatively stable, and in the absence of major emergencies, the fault information on both sides of the power supply and power consumption also has certain periodic characteristics. For example, in extreme weather cycles such as hot summers and cold winters, specific faults occur far more often than in other weather conditions. Based on this, there are time series characteristics between real-time measurement data. By extracting and analyzing time series characteristics, the correlation between measurement data can be further reflected, and then the data characteristics that can characterize the validity of a large amount of measurement data can be better extracted. Because the real-time measurement data is huge in magnitude, this application uses a fully connected network to complete it. The specific steps include:

[0057] A22. Based on the fully connected network, a time series feature machine learning network is established to obtain the time series feature attributes of the data. Combined with the time series relationship of the data, the appropriate time window is determined according to the power supply and distribution cycle of the distribution network. For the w measurement data {x 1 ,x 2 ......x w}, through the fully connected network Q, the data x 1 Convert to the factor vector Q x =W Q x 1 +q, and at the same time, other data x i , i∈2~w is converted into a solution vector P through a fully connected network P i =W P x j +p; the selection of the time window is mainly determined by the power supply and distribution cycle of the current distribution network. In principle, a longer time window can better extract the continuous time series characteristics between data, but in actual selection, the current system data and computing power capacity should be considered for optimization to ensure the operation efficiency of the system;

[0058] Then, the network model is optimized through machine learning, and the activation function is used to obtain the data x 1 With data x i The attention weights are: The data x under the time scale is obtained by weighted aggregation i For data x 1 A fusion of influences Get data {x 1 ,x 2 ......x w} Consider the temporal feature expression {x 1 ',x' 2 ......x' w}; where Wx+y is the formal expression of the fully connected network;

[0059] Through the above two steps, we can obtain the unified temporal and spatial structural characteristics of multi-source data. By screening and extracting effective feature attributes, we can realize the integration and compression of multi-source data, extract effective data, and eliminate invalid data, thereby providing a basis for realizing multi-end and multi-source data intercommunication. On this basis, through further data fusion processing, we can further compress the data volume and complete the mutual verification between multi-source data. Specifically:

[0060] A23. Create fused features based on feature concatenation and position encoding to obtain fused feature attributes. Specifically: establish a position encoding network, express the structural feature parameters of the data in vector form through position embedding to form a structural feature vector, use several layers of perceptron network models to reduce the dimension of the embedded initial structural features, express the time series features in the form of time series feature vectors, and concatenate and aggregate the structural feature vectors with the time series feature vectors to obtain full attribute features.

[0061] The fusion feature is mainly used to unify the spatial structure attributes and time attributes of real-time measurement data. We use the position encoding network model to embed the multi-dimensional measurement data into a vector-shaped time series feature. In actual implementation, considering the computational pressure caused by multi-dimensional features, this application further applies the multi-layer perception network model alignment to perform dimensionality reduction processing, so as to compress the data dimension and reduce the total amount of feature data.

[0062] By extracting fusion features, the multi-source heterogeneous measurement data are extracted and merged to have unified feature attributes, thereby realizing mutual verification analysis between multi-source data and further improving the validity of the data, which is specifically step A3.

[0063] A3. Use the reconstruction model to reconstruct the abnormal data and analyze the difference between it and the original data to verify whether the data is normal, including A31~A32:

[0064] A31. Create a variational autoencoder network model (VAE), train the variational autoencoder network model using all attribute features corresponding to historical data, and extract the real data features of normal data;

[0065] The reconstruction model uses historical data to learn its characteristic distribution, where the reconstruction model is mainly selected based on the data structure characteristics and the machine learning model. This application preferably adopts the principle of long short-term memory network model and completes data reconstruction through variational autoencoder network model;

[0066] Among them, the input of the autoencoder network model is a vector set composed of all attribute features obtained by splicing and aggregation. The input is fitted by the encoder of the autoencoder network model to make it have the mean and variance of the normal distribution. Then, the appropriate sampling parameters are selected through machine learning to sample the fitted data of the above-mentioned normal distribution to obtain the encoded data. The encoded data is decoded by the decoder of the autoencoder network model to obtain the reconstructed vector set, which is the feature expression of the real data.

[0067] A32. The trained autoencoder network model (VAE) will be used to verify the deviation of all attribute features of real-time data and filter out abnormal data;

[0068] Step 2: Analyze the scope of frequent power outages based on multi-source real-time measurement data, and receive distribution network fault data from the distribution automation system or the power consumption information collection system based on the data center; call the corresponding fault location algorithm for location according to different fault data; and analyze the scope of frequent power outages after the distribution network fault is located; specifically, steps B1 to B3:

[0069] B1. Extract and fuse the data features of the collected real-time distribution fault data and screen the abnormal data based on step 1, determine the abnormal data, repair or eliminate the abnormal data through field investigation or data processing, and obtain effective distribution network fault data;

[0070] B2. Extract and classify effective distribution network fault data, call corresponding fault location algorithms according to different fault data, and determine the faulty equipment and fault type;

[0071] B3. Determine the faulty equipment that is subject to frequent power outages according to the frequent power outage specifications of the distribution network, and search and determine the scope of frequent power outages by combining the faulty equipment and fault type information with the distribution network topology model.

[0072] Step B3 specifically includes B31-B32:

[0073] B31. Based on the determined faulty equipment and fault type, and in combination with the frequent power outage specifications of the distribution network, determine the faulty equipment information that is to be frequently powered off;

[0074] B32. Determine the scope of frequent power outages by searching based on the information of the equipment that is expected to be frequently powered outages in combination with the distribution network topology model, specifically including: performing a single-line search in the distribution network topology model with the power equipment that is expected to be frequently powered outages as the search origin to obtain the power supply node that supplies power to the power equipment and all distribution transformers in the search path; performing a multi-line search in parallel in the distribution network topology model with the power supply equipment that is expected to be frequently powered outages as the search origin to obtain all encountered transformers.

[0075] The single-line search at least includes: for the bus on the search path, comparing whether the bus voltage is consistent with the voltage level of the attached substation, if they are consistent, continuing the search until any distribution transformer is found, if they are not consistent, stopping the search and treating the bus front-stage equipment as a power supply node; for the circuit breaker on the search path, retrieving whether it is a closing device determined in the frequent power outage specification of the distribution network, if so, continuing the search until any distribution transformer is found, if not, stopping the search;

[0076] Among them, the multi-line search at least includes: for the bus on the search path, comparing the bus voltage with the voltage level of the attached substation to see if they are consistent, if they are consistent, continuing the search until any distribution transformer is found, if they are inconsistent, stopping the search.

[0077] Step 3: Plan for frequent power outages and match existing operations, including:

[0078] The known planned maintenance work of the PMS system is used to form a planned information pool, and the frequent power outage data is matched with the planned information pool data to determine whether it is a known plan. If it is a known plan, no work order is issued; if it is not a known frequent power outage, go to step 4;

[0079] Step 4: Analyze the reset data based on the real-time signal, including:

[0080] Apply power restoration monitoring to filter active emergency repair data and obtain short-term power outage signals. Within 5 minutes, monitor these signals to determine whether they are short-term power restoration data. If they are short-term power restoration data, no active work order will be pushed; if they are not short-term frequent power outage data, proceed to step 5.

[0081] Step 5: Push emergency repair work orders based on the analysis results and real-time data, including:

[0082] Apply real-time information verification to filter active repair data; perform data verification on power failure signals, and analyze whether the frequent power outage data is real based on the verified judgment results. If it is not real frequent power outage data, do not push active work orders; if it is real frequent power outage data, push active work orders and go to step 6;

[0083] Step 6: Complete emergency repair and archiving work, including:

[0084] Generate proactive repair work orders on the distribution and emergency platform and automatically push them to the "Accepted" module of the repair process management. The service agent will manually assign a repair team to dispatch the order, and the repair team will handle it through the order acceptance, return and archiving processes.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for monitoring and early warning of frequent outages based on real-time measurement data. It is characterized in that The steps include: Step 1: Establish a multi-source data fusion processing mechanism, perform standardized fusion processing on multi-source data, and use multi-source data for fusion and mutual verification; including: Obtain real-time data on power outages in the distribution network, standardize the data, extract and analyze the characteristics of the distribution network data based on the time series characteristics of the distribution network data, and verify multi-source data; Step 2: Analyze the scope of power outage based on multi-source real-time measurement data, including: Receive distribution network fault data from the distribution automation system or the power consumption information collection system based on the data center; call the corresponding fault location algorithm for location according to different fault data; analyze the power outage scope after locating the distribution network fault; Step 3: Power outage planning and matching planning with existing operations, including: Use the known planned maintenance work of the PMS system to form a planned information pool, match the power outage data with the planned information pool data, and determine whether it is a known plan. If it is a known plan, no work order will be issued; if it is not a known power outage, go to step 4; Step 4: Analyze the reset data based on the real-time signal, including: Apply power restoration monitoring to filter active repair data and obtain short-term power outage signals. Within 5 minutes, monitor these signals to determine whether they are short-term power restoration data. If they are short-term power restoration data, no active work order will be pushed; if they are not short-term power outage data, proceed to step 5. Step 5: Push emergency repair work orders based on the analysis results and real-time data, including: Apply real-time information verification to filter active repair data; perform data verification on power outage signals, and analyze whether it is real power outage data based on the verified analysis results. If it is not real power outage data, do not push active work orders; if it is real power outage data, push active work orders and go to step 6; Step 6: Complete emergency repair and archiving work, including: Generate an active emergency repair work order on the emergency repair platform and automatically push it to the "Accepted" module of the emergency repair process management. The service agent will manually assign the emergency repair team to dispatch the order, and the emergency repair team will handle it through the order acceptance, return and archiving processes.

2. According to the method of claim 1, a method for monitoring and early warning of frequency outages based on real-time measurement data, It is characterized in that The step 1 further comprises the following steps: The step 1 specifically comprises: A1. According to the preset distribution network data standard specifications, the data specifications, dimensions, data types, data magnitudes and other attributes of the corresponding multi-source data are transformed and integrated to obtain basic data, and pre-processed using the cloud platform fusion service component; A2. For the preprocessed data, select a suitable feature extraction model for feature extraction fusion according to the data structure characteristics of the main data; A3. Use the reconstruction model to reconstruct the abnormal data and analyze the difference between it and the original data to verify whether the data is normal.

3. According to the method of claim 2 for monitoring and early warning of frequency outages based on real-time measurement data, It is characterized in that The feature extraction fusion in step A2 includes: obtaining data structure feature attributes based on a dynamic graph attention network model combined with the structural relationship of the data; establishing a time series feature machine learning network based on a fully connected network to obtain data time series feature attributes; and creating a fusion feature based on feature splicing and position encoding to obtain a fusion feature attribute.

4. According to the method of claim 2 for monitoring and early warning of frequency outages based on real-time measurement data, It is characterized in that The feature extraction and fusion in step A2 specifically includes: A21. Based on the dynamic graph attention network model and the structural relationship of the data, the data structure feature attributes are obtained. Specifically, the original data is vectorized, and a network structure model containing all the data is established based on the data dependency relationship, where each data variable constitutes each node in the spatial model. Based on the dynamic graph attention network model, an embedding layer is created to learn the data structure features. A22. Based on the fully connected network, a time series feature machine learning network is established to obtain the time series feature attributes of the data. Specifically, based on the time series relationship of the data, a suitable time window is determined according to the power supply and distribution cycle of the distribution network. For the data {x 1 ,x 2 ......x w }, through the fully connected network Q, the data x 1 Convert to the factor vector Q x =W Q x 1 +q, and at the same time, other data x i , i∈2~w is converted into a solution vector P through a fully connected network P i =W P x j +p, optimize the network through machine learning, and use the activation function to get the data x 1 With data x i The attention weights are: The data x under the time scale is obtained by weighted aggregation i For data x 1 A fusion of influences Get data {x 1 ,x 2 ......x w } Consider the temporal feature expression {x 1 ',x' 2 ......x' w }; A23. Create fused features based on feature concatenation and position encoding to obtain fused feature attributes. Specifically: establish a position encoding network, express the structural feature parameters of the data in vector form through position embedding to form a structural feature vector, express the time series features in the form of a time series feature vector, and concatenate and aggregate the structural feature vector and the time series feature vector to obtain the full attribute feature.

5. According to the method of claim 4 for monitoring and early warning of frequency outages based on real-time measurement data, It is characterized in that The step A2 also includes: using a plurality of layers of perceptron network models to perform dimensionality reduction processing on the initial structural features obtained by embedding.

6. According to the method of claim 1 for monitoring and early warning of frequency outages based on real-time measurement data, It is characterized in that The step A3 specifically includes: A31. Create a variational autoencoder network model (VAE), train the variational autoencoder network model using all attribute features corresponding to historical data, and extract the real data features of normal data; A32. The trained autoencoder network model (VAE) will be used to perform deviation verification on all attribute features of real-time data and filter out abnormal data.

7. A method for monitoring and early warning emergency repair of frequency outages based on real-time measurement data according to claim 6, It is characterized in that The input of the autoencoder network model is a vector set consisting of all attribute features obtained by splicing and aggregation. The input is fitted by the encoder of the autoencoder network model to make it have the mean and variance of the normal distribution. Then, the appropriate sampling parameters are selected through machine learning to sample the fitted data of the normal distribution to obtain the encoded data. The encoded data is decoded by the decoder of the autoencoder network model to obtain the reconstructed vector set, which is the feature expression of the real data.

8. According to the method of claim 1 for monitoring and early warning emergency repair of frequency outages based on real-time measurement data, It is characterized in that The step 2 specifically includes: B1. Extract and fuse the data features of the collected real-time distribution fault data and screen the abnormal data based on step 1, determine the abnormal data, repair or eliminate the abnormal data through field investigation or data processing, and obtain effective distribution network fault data; B2. Extract and classify effective distribution network fault data, call corresponding fault location algorithms according to different fault data, and determine the faulty equipment and fault type; B3. Determine the faulty equipment to be shut down according to the power outage specifications of the distribution network, and search and determine the scope of the power outage by combining the faulty equipment and fault type information with the distribution network topology model.

9. A method for monitoring and early warning emergency repair of frequency outages based on real-time measurement data according to claim 8, It is characterized in that The step B3 specifically includes: B31. Based on the identified faulty equipment and fault type, determine the faulty equipment information to be shut down in combination with the power outage specifications of the distribution network; B32. Determine the scope of the power outage by searching based on the information of the equipment to be powered off and the distribution network topology model, specifically including: performing a single-line search in the distribution network topology model with the power equipment to be powered off as the search origin to obtain the power supply node supplying power to the power equipment and all distribution transformers in the search path; performing a multi-line search in parallel in the distribution network topology model with the power equipment to be powered off as the search origin to obtain all encountered transformers.

10. A method for monitoring and early warning emergency repair of frequency outages based on real-time measurement data according to claim 9, It is characterized in that The single-line search includes: for the bus on the search path, comparing whether the bus voltage is consistent with the voltage level of the attached substation, if they are consistent, continuing the search until any distribution transformer is found, if they are not consistent, stopping the search and treating the bus front-stage device as a power supply node; for the circuit breaker on the search path, retrieving whether it is a closing device determined in the power outage specification of the distribution network, if so, continuing the search until any distribution transformer is found, if not, stopping the search; The multi-line search includes: for the busbars on the search path, comparing the busbar voltage with the voltage level of the attached substation to see if they are consistent, if they are consistent, continuing the search until any distribution transformer is found, if they are inconsistent, stopping the search.

Images