Power supply fault area address analysis method based on electric power customer service work order

Through the fault address analysis method based on power customer service big data, Chinese text analysis and neural network model are used to identify the power grid fault area, solving the problem of slow grid fault identification and response, and improving the fault processing efficiency and accuracy.

CN119005169APending Publication Date: 2024-11-22国家电网有限公司客户服务中心
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411239088.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

When dealing with regional grid faults, State Grid Corporation relies on manual statistics and information reporting, which makes it difficult to identify and risk assessment and slow response, especially in the event of sudden failures, operating personnel workload, which affects on-site control.

Method used

Using the power supply fault area address analysis method based on power customer service big data, the fault address information is obtained through 95598 acceptance work orders, and Chinese text analysis, One-Hot encoding and neural network model training is carried out, combined with the K-means clustering algorithm and cosine similarity and distance discrimination method to identify the address of the fault area.

Benefits of technology

It realizes rapid positioning of fault areas, reduces the workload of operation personnel, improves the response speed and efficiency of fault handling, and provides a data basis for quickly formulating fault handling plans.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005169A_ABST
    Figure CN119005169A_ABST
Patent Text Reader

Abstract

The invention relates to a power supply fault area address analysis method based on a power customer service work order. The analysis method comprises the following steps: obtaining a fault repair electricity utilization address based on a 95598 acceptance work order and a fault processing work order; in the same time period, segmenting the address character string into different levels of address information; counting unique address vocabulary information in all word segmentation results, and constructing an address vocabulary; mapping a word segmentation result to an index in a vocabulary, and generating a corresponding One-Hot vector; using the vectors after One-Hot encoding as features to be input into a neural network model, ensuring that the model is fully trained, and generating word vectors; and any one of a K-means clustering algorithm, a cosine similarity method and a distance discrimination method is selected to determine the generated word vector. According to the invention, a data basis is provided for rapidly positioning a centralized fault area and rapidly formulating a fault area disposal scheme.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a power data processing technology, and in particular to a power supply fault area address analysis method based on a power customer service work order. Background Art

[0002] At present, the State Grid Corporation of China mainly relies on information reporting from provincial companies and manual statistical analysis to perceive regional grid faults and substation-level problems. It is relatively passive in identifying fault problems, difficult to assess risk of problems, and slow to respond to problem handling. Taking the identification of fault areas as an example, the current identification method relies on operation and management personnel at all levels to manually count the regional distribution of repair work orders and infer the current fault area. During peak traffic periods, especially during the peak summer period, when sudden faults occur in many places, the workload of operators to count fault areas is large, which may affect on-site management and control work. Therefore, with the help of big data technology, it is urgent to analyze the address of power supply fault areas and systems. Summary of the invention

[0003] The present invention proposes a method for analyzing the address of power supply fault areas based on power customer service big data. The specific technical solution is as follows:

[0004] Method 1 for analyzing the address of power supply fault area based on power customer service work order includes the following process:

[0005] S1: Based on the 95598 acceptance work order and fault handling work order, obtain the power address for fault repair; the power address information includes province, city, district / county, street / township, community / village, residential area / natural village, and house number;

[0006] S2: In the same time period, Chinese text is parsed for the addresses that collectively reflect the faults, and the address strings are divided into address information of different levels;

[0007] S3: Count the unique address vocabulary information in all word segmentation results and build an address vocabulary table;

[0008] S4: Map the word segmentation result to the index in the vocabulary and generate the corresponding One-Hot vector; mark the index position as 1 and the rest as 0;

[0009] S5: Input the one-hot encoded vector as a feature into the neural network model, set the learning rate and iteration number hyperparameters to ensure that the model is fully trained; generate word vectors;

[0010] S6: Select any one of the three methods: K-means clustering algorithm, cosine similarity, and distance discrimination method, and judge the generated word vector to form a cluster center.

[0011] Method 2 for analyzing the address of power supply fault area based on power customer service work order includes the following process:

[0012] S1: Based on the 95598 acceptance work order and fault handling work order, obtain the power address for fault repair; the power address information includes province, city, district / county, street / township, community / village, residential area / natural village, and house number;

[0013] S2: In the same time period, Chinese text is parsed for the addresses that collectively reflect the faults, and the address strings are divided into address information of different levels;

[0014] S3: Count the unique address vocabulary information in all word segmentation results and build an address vocabulary table;

[0015] S4: Use the segmented address data to train the Word2Vec model; select the Skip-gram model, set the window size, dimension size, number of trainings, and learning rate parameters; the model learns the distributed representation of vocabulary through training, and these word vectors can capture the semantic information of address vocabulary and the relationship between different address levels in the context;

[0016] S5: Generate word vectors. After the model training is completed, the word segmentation results are input into the trained Word2Vec model to generate word vectors for each address slice.

[0017] S6: Select any one of the three methods: K-means clustering algorithm, cosine similarity, and distance discrimination method to judge the generated word vector; identify the address cluster center.

[0018] Preferably, in the method 1 and the method 2, the specific steps of the K-means clustering algorithm are as follows:

[0019] S6.1: Determine the optimal number of clusters k using either the elbow method or the silhouette coefficient method; the elbow method plots the sum of squared errors (SSE) for each k value and selects the "elbow" position as the optimal k value; the silhouette coefficient method calculates the silhouette coefficients for different k values ​​and selects the k value corresponding to the maximum silhouette coefficient;

[0020] S6.2: Initialize the centroid; select the k-means++ centroid initialization method to ensure the rationality of the initial position of the centroid;

[0021] S6.3: Set the maximum number of iterations to control the convergence speed of the algorithm; set the tolerance to control the convergence accuracy of the algorithm;

[0022] S6.4: Use the K-means algorithm to cluster word vectors; the algorithm iteratively adjusts the centroid position until the preset number of iterations is reached or the tolerance requirement is met; each address word vector is assigned to the nearest centroid to form several clusters.

[0023] Preferably, in the method 1 and the method 2, the specific steps of the cosine similarity algorithm are as follows:

[0024] S6.11: Normalize the vectors; Normalize each word vector by calculating the modulus of the vector and dividing each component of the vector by the modulus to obtain a unit vector;

[0025] S6.12: Calculate dot product; Calculate the dot product between two normalized word vectors, i.e. cosine similarity; the formula for dot product is:

[0026]

[0027] Represent two standardized word vectors respectively;

[0028] S6.13: Similarity quantification; cosine similarity values ​​range from -1 to 1, where 1 indicates complete similarity, 0 indicates no similarity, and -1 indicates complete opposite;

[0029] S6.14: Form cluster centers based on cosine similarity distance.

[0030] Preferably, in the method 1 and the method 2, the specific steps of the distance determination method are as follows:

[0031] The Euclidean distance formula is used to calculate the straight-line distance between two word vectors. The formula is as follows:

[0032]

[0033] Among them, the A vector is (x 11 ,x 12 ,…,x 1n ), the B vector is (x 21 ,x 22 ,…,x 2n );

[0034] Cluster centers are formed based on Euclidean distance.

[0035] Preferably, the following process is included between S3 and S4:

[0036] S3.1: Data preprocessing: load address data and clean it, remove extra spaces and special characters, and standardize address information;

[0037] S3.2: Word segmentation: Use Chinese text word segmentation tools to segment address data and divide address strings into different levels of address information;

[0038] S3.3: Verification of word segmentation results: Verify and correct the word segmentation results to ensure that each address word can accurately reflect the address level information.

[0039] The present invention extracts the commonality of 95598 service work order data reflected in the concentrated fault area, and combines it with the concentrated fault area identification model to provide a data basis for quickly locating the concentrated fault area and quickly formulating a fault area disposal plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the analysis method in Example 1 of the present invention.

[0041] Figure 2 Schematic diagram of the analysis method in Example 2 of the present invention. DETAILED DESCRIPTION

[0042] Example 1

[0043] The power supply fault area address analysis method based on the power customer service work order includes the following processes:

[0044] S1: Based on the 95598 acceptance work order and fault handling work order, obtain the power address for fault reporting; the power address information includes province, city, district / county, street / township, community /

[0045] Village, community / natural village, house number;

[0046] It usually includes the following main information: basic customer information including customer name, contact number and customer number; fault reporting information including: reporting time, fault type, such as power outage, line failure, etc., and fault description.

[0047] The power supply address information includes: province, city, district / county, street / township, community / village, neighborhood / natural village and house number; this is the 7-level address database information of the power industry; this information is used to locate the specific location of the fault, so that the staff can respond and handle it quickly.

[0048] S2: In the same time period, Chinese text parsing is performed on the addresses that reflect the faults in a centralized manner, and the address strings are divided into address information of different levels; the 7-level address database of the power industry is a detailed address classification system, which includes seven levels from province to house number, and the specific structure is as follows: province, the highest level of administrative regions in the country, such as "Guangdong Province"; city, a city under a province, such as "Guangzhou City"; district / county, a district or county under a city, such as "Tianhe District" or "Conghua County"; street / township, a street or town under a district or county, such as "Chebei Street" or "Shengang Town"; community / village, a community or village under a street or town, such as "Dongpu Community" or "Shangchong Village"; community / natural village, a community or natural village under a community or village, such as "Zhonghai Kangcheng Community" or "Donghu Village"; house number, a specific building or address number, such as "No. 201, Longteng Building" or "Room 502, Unit 3, Building 1".

[0049] S3: Count the unique address vocabulary information in all word segmentation results and build an address vocabulary table;

[0050] S3.1: Data preprocessing: load address data and clean it, remove extra spaces and special characters, and standardize address information;

[0051] S3.2: Word segmentation: Use Chinese text word segmentation tools, such as Jieba, to segment address data and divide address strings into different levels of address information;

[0052] S3.3: Verification of word segmentation results: Verify and correct the word segmentation results to ensure that each address word can accurately reflect the address level information;

[0053] S4: Map the word segmentation result to the index in the vocabulary and generate the corresponding One-Hot vector; mark the index position as 1 and the rest as 0;

[0054] S5: Input the one-hot encoded vector as a feature into the neural network model, set the learning rate and iteration number hyperparameters to ensure that the model is fully trained; generate word vectors;

[0055] S6: Select any one of the three methods, namely, K-means clustering algorithm, cosine similarity, and distance discrimination method, to judge the generated word vector; identify the address cluster center; and extract the cluster addresses above level 5 as the concentrated fault area addresses.

[0056] The specific steps of the K-means clustering algorithm are as follows:

[0057] S6.1: Determine the optimal number of clusters k using either the elbow method or the silhouette coefficient method; the elbow method plots the sum of squared errors (SSE) for each k value and selects the "elbow" position as the optimal k value; the silhouette coefficient method calculates the silhouette coefficients for different k values ​​and selects the k value corresponding to the maximum silhouette coefficient;

[0058] S6.2: Initialize the centroid: Select the k-means++ centroid initialization method to ensure the rationality of the initial position of the centroid;

[0059] S6.3: Set the maximum number of iterations, such as 300, to control the convergence speed of the algorithm; set the tolerance, such as 0.0001, to control the convergence accuracy of the algorithm;

[0060] S6.4: Use the K-means algorithm to cluster the word vectors; the algorithm iteratively adjusts the centroid position until the preset number of iterations is reached or the tolerance requirement is met; each address word vector is assigned to the nearest centroid to form several clusters;

[0061] S6.5: Verify the clustering results and evaluate the clustering effect; check the effectiveness and rationality of the clustering by visualizing the clustering results or calculating clustering evaluation indicators, such as the silhouette coefficient.

[0062] The specific steps of the cosine similarity algorithm are as follows:

[0063] S6.11: Normalize the vectors; Normalize each word vector by calculating the modulus of the vector and dividing each component of the vector by the modulus to obtain a unit vector;

[0064] S6.12: Calculate dot product; Calculate the dot product between two normalized word vectors, i.e. cosine similarity; the formula for dot product is:

[0065]

[0066] Represent two standardized word vectors respectively;

[0067] S6.13: Similarity quantification; cosine similarity values ​​range from -1 to 1, where 1 indicates complete similarity, 0 indicates no similarity, and -1 indicates complete opposite;

[0068] Parameter setting and optimization: Select appropriate vector dimensions and training times to optimize the word vector quality and cosine similarity calculation effect; to improve calculation efficiency, batch processing can be used to calculate cosine similarity; through the above steps, cosine similarity can be used to effectively calculate the similarity between address fragment word vectors, helping to evaluate the relationship between different address fragments in the address segmentation task.

[0069] The specific steps of the distance determination method are as follows:

[0070] The Euclidean distance formula is used to calculate the straight-line distance between two word vectors. The formula is as follows:

[0071]

[0072] Among them, the A vector is (x 11 ,x 12 ,…,x 1n ), the B vector is (x 21 ,x 22 ,…,x 2n );

[0073] Cluster centers are formed based on Euclidean distance.

[0074] Example 2

[0075] The power supply fault area address analysis method based on the power customer service work order includes the following processes:

[0076] S1: Based on the 95598 acceptance work order and fault handling work order, obtain the power address for fault reporting; the power address information includes province, city, district / county, street / township, community /

[0077] Village, community / natural village, house number;

[0078] S2: In the same time period, Chinese text is parsed for the addresses that collectively reflect the faults, and the address strings are divided into address information of different levels;

[0079] S3: Count the unique address vocabulary information in all word segmentation results and build an address vocabulary table;

[0080] S3.1: Data preprocessing: load address data and clean it, remove extra spaces and special characters, and standardize address information;

[0081] S3.2: Word segmentation: Use Chinese text word segmentation tools, such as Jieba, to segment address data and divide address strings into different levels of address information;

[0082] S3.3: Verification of word segmentation results: Verify and correct the word segmentation results to ensure that each address word can accurately reflect the address level information;

[0083] S4: Use the segmented address data to train the Word2Vec model; select the Skip-gram model, set the window size, dimension size, number of trainings, and learning rate parameters; the model learns the distributed representation of vocabulary through training, and these word vectors can capture the semantic information of address vocabulary and the relationship between different address levels in the context; in this embodiment, the window size can be selected as 5, the word vector dimension can be selected as 100, and the number of trainings can be selected as 5;

[0084] S5: Generate word vectors. After the model training is completed, the word segmentation results are input into the trained Word2Vec model to generate word vectors for each address slice. These high-dimensional vectors represent the position of the words in the semantic space.

[0085] S6: Select any one of the three methods, namely, K-means clustering algorithm, cosine similarity, and distance discrimination method, to judge the generated word vector; identify the address cluster center, and extract cluster addresses above level 5 as concentrated fault area addresses.

[0086] The K-means clustering method, cosine similarity algorithm, and distance discrimination method are exactly the same as those in Example 1.

Claims

1. A method for analyzing the address of a power supply fault area based on a power customer service work order, characterized in that: The process includes the following: S1: Based on the 95598 acceptance work order and fault handling work order, obtain the power address for fault repair; the power address information includes province, city, district / county, street / township, community / village, residential area / natural village, and house number; S2: In the same time period, Chinese text is parsed for the addresses that collectively reflect the faults, and the address strings are divided into address information of different levels; S3: Count the unique address vocabulary information in all word segmentation results and build an address vocabulary table; S3.1: Data preprocessing: load address data and clean it, remove extra spaces and special characters, and standardize address information; S3.2: Word segmentation: Use Chinese text word segmentation tools to segment address data and divide address strings into different levels of address information; S3.3: Verification of word segmentation results: Verify and correct the word segmentation results to ensure that each address word can accurately reflect the address level information; S4: Map the word segmentation result to the index in the vocabulary and generate the corresponding One-Hot vector; mark the index position as 1 and the rest as 0; S5: Input the one-hot encoded vector as a feature into the neural network model, set the learning rate and iteration number hyperparameters to ensure that the model is fully trained; generate word vectors; S6: Select any one of the three methods: K-means clustering algorithm, cosine similarity, and distance discrimination method, judge the generated word vector, and identify the address cluster center.

2. According to claim 1, the method for analyzing the power supply fault area address based on the power customer service work order is characterized in that: The specific steps of the K-means clustering algorithm are as follows: S6.1: Determine the optimal number of clusters k using either the elbow method or the silhouette coefficient method; the elbow method plots the sum of squared errors (SSE) for each k value and selects the "elbow" position as the optimal k value; the silhouette coefficient method calculates the silhouette coefficients for different k values ​​and selects the k value corresponding to the maximum silhouette coefficient; S6.2: Initialize the centroid: Select the k-means++ centroid initialization method to ensure the rationality of the initial position of the centroid; S6.3: Set the maximum number of iterations to control the convergence speed of the algorithm; set the tolerance to control the convergence accuracy of the algorithm; S6.4: Use the K-means algorithm to cluster word vectors; the algorithm iteratively adjusts the centroid position until the preset number of iterations is reached or the tolerance requirement is met; each address word vector is assigned to the nearest centroid to form several cluster centers.

3. According to claim 1, the method for analyzing the power supply fault area address based on the power customer service work order is characterized in that: The specific steps of the cosine similarity algorithm are as follows: S6.11: Normalize the vectors; Normalize each word vector by calculating the modulus of the vector and dividing each component of the vector by the modulus to obtain a unit vector; S6.12: Calculate dot product; Calculate the dot product between two normalized word vectors, i.e. cosine similarity; the formula for dot product is: Represent two standardized word vectors respectively; S6.13: Similarity quantification; cosine similarity values ​​range from -1 to 1, where 1 indicates complete similarity, 0 indicates no similarity, and -1 indicates complete opposite; S6.14: Form cluster centers based on cosine similarity distance.

4. The method for analyzing the power supply fault area address based on the power customer service work order according to claim 1 is characterized in that: The specific steps of the distance determination method are as follows: The Euclidean distance formula is used to calculate the straight-line distance between two word vectors. The formula is as follows: Among them, the A vector is (x 11 ,x 12 ,…,x 1n ), the B vector is (x 21 ,x 22 ,…,x 2n ); Cluster centers are formed based on Euclidean distance.

5. The method for analyzing the power supply fault area address based on the power customer service work order according to any one of claims 1 to 4, characterized in that: The following process is included between S3 and S4: S3.1: Data preprocessing: load address data and clean it, remove extra spaces and special characters, and standardize address information; S3.2: Word segmentation: Use Chinese text word segmentation tools to segment address data and divide address strings into different levels of address information; S3.3: Verification of word segmentation results: Verify and correct the word segmentation results to ensure that each address word can accurately reflect the address level information.

6. A method for analyzing the address of a power supply fault area based on a power customer service work order, characterized in that: The process includes the following: S1: Based on the 95598 acceptance work order and fault handling work order, obtain the power address for fault reporting; the power address information includes province, city, district / county, street / township, community / Village, community / natural village, house number; S2: In the same time period, Chinese text is parsed for the addresses that collectively reflect the faults, and the address strings are divided into address information of different levels; S3: Count the unique address vocabulary information in all word segmentation results and build an address vocabulary table; S4: Use the segmented address data to train the Word2Vec model; select the Skip-gram model, set the window size, dimension size, number of trainings, and learning rate parameters; the model learns the distributed representation of vocabulary through training, and these word vectors can capture the semantic information of address vocabulary and the relationship between different address levels in the context; S5: Generate word vectors. After the model training is completed, the word segmentation results are input into the trained Word2Vec model to generate word vectors for each address slice. S6: Select any one of the three methods: K-means clustering algorithm, cosine similarity, and distance discrimination method, judge the generated word vector, and identify the address cluster center.

7. The method for analyzing the power supply fault area address based on the power customer service work order according to claim 6 is characterized in that: The specific steps of the K-means clustering algorithm are as follows: S6.1: Determine the optimal number of clusters k using either the elbow method or the silhouette coefficient method; the elbow method plots the sum of squared errors (SSE) for each k value and selects the "elbow" position as the optimal k value; the silhouette coefficient method calculates the silhouette coefficients for different k values ​​and selects the k value corresponding to the maximum silhouette coefficient; S6.2: Initialize the centroid: Select the k-means++ centroid initialization method to ensure the rationality of the initial position of the centroid; S6.3: Set the maximum number of iterations to control the convergence speed of the algorithm; set the tolerance to control the convergence accuracy of the algorithm; S6.4: Use the K-means algorithm to cluster word vectors; the algorithm iteratively adjusts the centroid position until the preset number of iterations is reached or the tolerance requirement is met; each address word vector is assigned to the nearest centroid to form several clusters.

8. The method for analyzing the power supply fault area address based on the power customer service work order according to claim 6 is characterized in that: The specific steps of the cosine similarity algorithm are as follows: S6.11: Normalize the vectors; Normalize each word vector by calculating the modulus of the vector and dividing each component of the vector by the modulus to obtain a unit vector; S6.12: Calculate dot product; Calculate the dot product between two normalized word vectors, i.e. cosine similarity; the formula for dot product is: Represent two standardized word vectors respectively; S6.13: Similarity quantification; cosine similarity values ​​range from -1 to 1, where 1 indicates complete similarity, 0 indicates no similarity, and -1 indicates complete opposite; S6.14: Form cluster centers based on cosine similarity distance.

9. The method for analyzing the power supply fault area address based on the power customer service work order according to claim 1, characterized in that: The specific steps of the distance determination method are as follows: The Euclidean distance formula is used to calculate the straight-line distance between two word vectors. The formula is as follows: Among them, the A vector is (x 11 ,x 12 ,…,x 1n ), the B vector is (x 21 ,x 22 ,…,x 2n ); Cluster centers are formed based on Euclidean distance.

10. The method for analyzing the power supply fault area address based on the power customer service work order according to any one of claims 6 to 9, characterized in that: The following process is included between S3 and S4: S3.1: Data preprocessing: load address data and clean it, remove extra spaces and special characters, and standardize address information; S3.2: Word segmentation: Use Chinese text word segmentation tools to segment address data and divide address strings into different levels of address information; S3.3: Verification of word segmentation results: Verify and correct the word segmentation results to ensure that each address word can accurately reflect the address level information.