Power grid equipment fault analysis method and device, computer equipment and storage medium

Abstract

The invention discloses a power grid equipment fault analysis method and device, computer equipment and a storage medium, and relates to the technical field of power grid equipment fault diagnos.The power grid equipment fault analysis method and device integrate various complex and diverse data sources which are widely distributed in a power grid, construct a highly structured and logically clear data architecture based on a snowflake model, and analyze the fault of the power grid equipment. And data related to fault analysis can be quickly positioned from mass data in a short time. Meanwhile, data storage is carried out in the Hadoop distributed file system, the information dimension of fault analysis is greatly enriched by utilizing the powerful distributed storage, data redundancy and fault-tolerant mechanism of the Hadoop distributed file system, and a stable data basis is provided for power grid equipment fault analysis. And multiple complex factors of power grid equipment faults are comprehensively considered, a comprehensive and high-adaptability model system is constructed, various complex fault types can be accurately identified, fault diagnosis errors caused by single or inaccurate models are effectively avoided, and the fault prediction and diagnosis capability is improved especially under the complex conditions of severe weather and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid equipment fault diagnosis, and in particular to a power grid equipment fault analysis method and device, computer equipment and a storage medium. BACKGROUND

[0002] In modern power systems, there are a large number of power grid equipment with complex structures, and their operating states are affected by multiple factors. With the development of information technology, the power system has generated a large amount of multi-source heterogeneous data, including equipment operating data, monitoring data, weather data, etc.

[0003] In related technologies, the traditional power grid equipment fault analysis method usually directly constructs a fault diagnosis model using equipment operating data for fault diagnosis. However, the applicant realizes that the model constructed by the traditional method has a single structure and is difficult to effectively process these large-scale and complex data, resulting in insufficient accuracy and timeliness of fault analysis; at the same time, the traditional method considers only a single factor, which makes it difficult to accurately identify various complex fault types under adverse weather conditions, and is prone to fault diagnosis errors. SUMMARY

[0004] Therefore, the present application provides a power grid equipment fault analysis method and device, computer equipment and a storage medium, which mainly aims to solve the problem that the model constructed by the traditional method has a single structure and is difficult to effectively process these large-scale and complex data, resulting in insufficient accuracy and timeliness of fault analysis; at the same time, the traditional method considers only a single factor, which makes it difficult to accurately identify various complex fault types under adverse weather conditions, and is prone to fault diagnosis errors.

[0005] According to a first aspect of the present application, a power grid equipment fault analysis method is provided, which comprises:

[0006] Obtaining power grid equipment related historical data and weather historical data, preprocessing the power grid equipment related historical data and the weather historical data to obtain target feature data;

[0007] Constructing a snowflake model using the target feature data, storing the snowflake model in a Hadoop cluster according to a preset format, and establishing a global index based on the stored snowflake model;

[0008] Obtaining first training samples and second training samples according to the global index, constructing a fault prediction model using the first training samples based on a recurrent neural network algorithm, and constructing a fault diagnosis model using the second training samples based on a support vector machine algorithm and a decision tree machine learning algorithm;

[0009] The real-time data of the power grid equipment and the real-time meteorological data are acquired, the real-time data of the power grid equipment is input into the fault diagnosis model, the real-time meteorological data is input into the fault prediction model for fault diagnosis and early warning, and a power grid equipment fault analysis result is obtained.

[0010] According to the second aspect of the present application, a power grid equipment fault analysis device is provided, which comprises:

[0011] The processing module is configured to acquire power grid equipment related historical data and meteorological historical data, and preprocess the power grid equipment related historical data and the meteorological historical data to obtain target feature data.

[0012] The storage module is configured to construct a snowflake model using the target feature data, store the snowflake model in a Hadoop cluster according to a preset format, and establish a global index based on the stored snowflake model.

[0013] The construction module is configured to acquire first training samples and second training samples according to the global index, construct a fault prediction model using the first training samples based on a recurrent neural network algorithm, and construct a fault diagnosis model using the second training samples based on a support vector machine algorithm and a decision tree machine learning algorithm.

[0014] The analysis module is configured to acquire real-time data of the power grid equipment and real-time meteorological data, input the real-time data of the power grid equipment into the fault diagnosis model and input the real-time meteorological data into the fault prediction model for fault diagnosis and early warning, and obtain a power grid equipment fault analysis result.

[0015] According to the third aspect of the present application, a computer device is provided, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method of any one of the first aspect when executing the computer program.

[0016] According to the fourth aspect of the present application, a storage medium is provided, which stores a computer program, and the computer program implements the steps of the method of any one of the first aspect when executed by a processor.

[0017] Through the above technical solutions, the technical solutions provided by the embodiments of the present application have at least the following advantages:

[0018] The application provides a power grid equipment fault analysis method and device, computer equipment and a storage medium. The application obtains power grid equipment related historical data and meteorological historical data, pre-processes the power grid equipment related historical data and meteorological historical data to obtain target feature data, constructs a snowflake model by using the target feature data, stores the snowflake model in a Hadoop cluster according to a preset format, establishes a global index based on the stored snowflake model, obtains first training samples and second training samples according to the global index, constructs a fault prediction model by using the first training samples based on a recurrent neural network algorithm, constructs a fault diagnosis model by using the second training samples based on a support vector machine algorithm and a decision tree machine learning algorithm, obtains power grid equipment real-time data and meteorological real-time data, respectively inputs the power grid equipment real-time data into the fault diagnosis model and inputs the meteorological real-time data into the fault prediction model for fault diagnosis and early warning, and obtains a power grid equipment fault analysis result. The application integrates various widely distributed and complex data sources in the power grid, constructs a highly structured and logically clear data architecture based on the snowflake model, greatly improves the data query and retrieval efficiency, and can quickly locate the key data related to fault analysis from massive data in a short time. Meanwhile, data storage is performed in the Hadoop distributed file system, the powerful distributed storage, data redundancy and fault tolerance mechanism is fully utilized, the information dimension of fault analysis is greatly enriched, the high reliability and safety of data are ensured, and a stable data foundation is provided for power grid equipment fault analysis. Moreover, the application comprehensively considers various complex factors of power grid equipment faults, constructs a comprehensive and adaptable model system, can accurately identify various complex fault types, effectively avoids fault diagnosis failure caused by single or inaccurate model, especially in complex conditions such as severe weather, improves the fault prediction and diagnosis capability, and ensures the safe and stable operation of the power grid.

[0019] The above description is only a summary of the technical scheme of the application. In order to more clearly understand the technical means of the application, the application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following specific embodiments of the application are described. BRIEF DESCRIPTION OF DRAWINGS

[0020] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not meant to limit the scope of the application. Moreover, the same reference numerals in the attached drawings indicate the same or similar components. In the drawings:

[0021] Figure 1 A method flow diagram of power grid equipment fault analysis provided by an embodiment of the application is shown;

[0022] Figure 2Another method flow diagram for analyzing power grid equipment failure provided by the embodiments of the application is shown.

[0023] Figure 3 A structural diagram of the method for analyzing power grid equipment failure provided by the embodiments of the application is shown.

[0024] Figure 4 An apparatus structure diagram of the computer device provided by the embodiments of the application is shown. DETAILED DESCRIPTION

[0025] In the description of the present application, it should be understood that the terms "first", "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0026] Exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided so that the present application can be more thoroughly understood and the scope of the present application can be accurately conveyed to those skilled in the art.

[0027] The embodiments of the present application provide a method for analyzing power grid equipment failure, as shown in the method includes: Figure 1

[0028] 101, obtaining power grid equipment related historical data and meteorological historical data, preprocessing the power grid equipment related historical data and meteorological historical data to obtain target feature data.

[0029] ​To solve the problem of low fault diagnosis accuracy caused by insufficient data processing capacity and single model construction of traditional methods, the application provides a power grid equipment fault analysis method. The related historical data of power grid equipment and meteorological historical data are preprocessed, the data is processed and stored based on the Hadoop framework and the snowflake model, then a fault prediction model under adverse weather based on the recurrent neural network algorithm is established, and a fault diagnosis model based on the support vector machine algorithm and the decision tree machine learning algorithm is constructed. The fault prediction model and the fault diagnosis model are used for fault diagnosis and early warning. Based on the snowflake model, a highly structured and logically clear data architecture is constructed, which can quickly locate the key data related to fault analysis from massive data in a short time. At the same time, data storage is carried out in the Hadoop distributed file system, which fully utilizes the data redundancy and fault tolerance mechanism to ensure the high reliability and security of the data, and provides a stable data foundation for power grid equipment fault analysis. Moreover, various complex factors of power grid equipment fault are considered comprehensively, and a comprehensive and adaptable model system is constructed, which effectively overcomes many deficiencies of existing technologies in data processing, model construction and diagnosis efficiency, especially in complex conditions such as adverse weather, improves the ability of fault prediction and diagnosis, and ensures the safe and stable operation of the power grid. The execution subject of the application can be a fault analysis system, which relies on the computing power of the server to provide services for users. The server can be a standalone server, or it can provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and big data and artificial intelligence platforms, etc. The basic cloud computing server is used to realize accurate diagnosis of power grid equipment faults.

[0030] In the embodiments of the application, the related historical data of power grid equipment is data collected from multiple data sources, including structured data and unstructured data. The structured data is, for example, device account information, real-time running parameters, etc., and the unstructured data is, for example, device inspection images, sound data, etc. The meteorological historical data is adverse weather related information, which is unstructured data, such as meteorological data, meteorological image videos, etc. By obtaining the related historical data of power grid equipment and meteorological historical data, the data types can be enriched, and a more comprehensive data foundation can be provided for fault analysis.

[0031] Secondly, the fault analysis system adopts different data processing methods for structured data and unstructured data. For structured data, large-scale data cleaning operations can be quickly performed, and abnormal values and missing values can be accurately identified and processed through self-defined distributed algorithms, improving data quality and providing more valuable information for fault analysis. For unstructured data, with the help of professional image processing and audio processing tools, combined with deep learning algorithms, highly representative features can be extracted from device inspection images, sound data, and weather image videos, making unstructured data also available for fault analysis. By preprocessing the relevant historical data of power grid equipment and weather historical data, target feature data is obtained, providing higher quality data that meets the needs of subsequent fault analysis, which helps to improve the accuracy and reliability of fault analysis.

[0032] 102. Construct a snowflake model using the target feature data, store the snowflake model in the Hadoop cluster according to the preset format, and establish a global index based on the stored snowflake model.

[0033] In the embodiments of the present application, the fault analysis system designs the architecture of the snowflake model according to the requirements of power grid equipment fault diagnosis, divides the target feature data into multiple dimension tables and fact tables, for example, constructs the model with equipment, fault type, geographic location, time as dimensions and equipment operation data as facts. The snowflake model is a standardized data model that can reduce data redundancy and improve data consistency and accuracy. In the power grid fault analysis scenario, accurate and standardized data is crucial for accurately analyzing fault causes and predicting equipment status. For example, the equipment account information is stored in the snowflake model according to the standardized structure, which facilitates data maintenance and query, and reduces analysis errors caused by inconsistent data.

[0034] Secondly, the fault analysis system stores the snowflake model in the Hadoop cluster according to the preset format, because the Hadoop cluster has strong storage capacity and can handle large-scale target feature data. The amount of relevant historical data of power grid equipment and weather historical data is usually large, and the Hadoop cluster can meet the storage needs of these data, ensuring the integrity and availability of the data.

[0035] The specific process of building a Hadoop cluster environment is as follows: configuring multiple servers as Hadoop nodes, including at least one NameNode master node and multiple DataNode slave nodes. Set the related parameters of the Hadoop Distributed File System (HDFS) on the NameNode master node, configure the storage path and network parameters on the DataNode slave node, ensure that the DataNode slave node can communicate normally with the NameNode master node, and can efficiently store and read data blocks. Since Hadoop cluster adopts distributed storage and computing mode, it can process data in parallel on multiple computers, which is very important for processing massive power grid data, and can improve data processing efficiency. For example, when analyzing device fault data, a large amount of historical data can be quickly processed, and the fault analysis time can be shortened.

[0036] Finally, the fault analysis system can speed up data retrieval by establishing a global index. In fault analysis, it can quickly locate related data and improve analysis efficiency. For example, when searching for fault data of a specific device in a specific time period, the global index can quickly locate the relevant data records, reducing data query time. Moreover, the global index makes data organization and management more orderly, facilitating data updating, deletion and other operations. In the case of continuous updating of power grid data, the global index can ensure data consistency and manageability, providing convenience for subsequent data analysis and fault diagnosis. In addition, when performing complex fault analysis, it may be necessary to query data associated with multiple tables. The global index can support these complex queries to ensure the accuracy and timeliness of the query results, which helps to analyze the fault causes and device operating status in depth.

[0037] 103. Obtain the first training sample and the second training sample according to the global index, construct a fault prediction model based on the recurrent neural network algorithm using the first training sample, and construct a fault diagnosis model based on the support vector machine algorithm and the decision tree machine learning algorithm using the second training sample.

[0038] In the embodiments of the present application, the first training sample includes historical severe weather data, equipment operation data under severe weather, and equipment failure data under severe weather, and the second training sample includes equipment normal operation data and equipment historical failure data. Different types of sample data are obtained through the global index, which can accurately obtain data related to fault prediction and diagnosis, ensure that the selected sample data is closely related to the actual fault condition, and improve the effectiveness and pertinence of the sample data. Moreover, the use of the global index helps to comprehensively obtain the required data and ensure the integrity of the sample data. Whether it is equipment normal operation data, equipment historical failure data, or related data under severe weather, it can be accurately obtained through the global index, avoiding the influence of data missing on model training, and making the constructed model more consistent with the actual situation.

[0039] The fault analysis system constructs a fault prediction model based on a recurrent neural network algorithm (RNN). RNN can well process time series data and effectively analyze and learn the characteristics of equipment operation data changing over time, such as fluctuations in equipment operation parameters over time and trends in severe weather. Through RNN, time series information in the data can be captured, thereby more accurately predicting the occurrence of equipment failure and improving the accuracy and reliability of fault prediction.

[0040] The fault analysis system constructs a fault diagnosis model based on a support vector machine algorithm (SVM) and a decision tree machine learning algorithm. The two algorithms have different characteristics and advantages. SVM performs well in processing small samples and high-dimensional data, and can find the optimal classification hyperplane to accurately classify equipment normal operation data and equipment historical failure data. The decision tree algorithm has the characteristics of being intuitive and easy to understand, and can make hierarchical decisions based on the characteristics of the data to quickly and accurately diagnose the causes of equipment failure. The combination of the two can fully utilize their respective advantages to improve the accuracy and efficiency of fault diagnosis and enhance the model generalization ability. When faced with new equipment operation data, the model can accurately judge the running state of the equipment and timely discover equipment failure, providing strong support for the maintenance and management of the equipment.

[0041] Therefore, by constructing a fault prediction model and a fault diagnosis model respectively, the equipment failure can be analyzed from different angles. The fault prediction model can discover possible equipment failures in advance to provide a basis for preventive maintenance of the equipment, and the fault diagnosis model can accurately determine the causes of equipment failure to provide guidance for fault repair. The combination of the two can achieve comprehensive analysis of equipment failure and improve the reliability and stability of power grid equipment. In the fault analysis system, accurate fault prediction and diagnosis can shorten the fault handling time, reduce the equipment downtime, and reduce the loss caused by the fault.

[0042] 104. Obtain power grid equipment real-time data and meteorological real-time data, input the power grid equipment real-time data into the fault diagnosis model and input the meteorological real-time data into the fault prediction model for fault diagnosis and early warning, and obtain power grid equipment fault analysis results.

[0043] In the embodiments of the present application, the power grid equipment real-time data and meteorological real-time data are obtained and input into the corresponding models for analysis in a timely manner, so that potential faults and abnormal conditions of the power grid equipment can be quickly found. For example, when the real-time operation parameters of the equipment appear abnormal fluctuations or the weather conditions may affect the equipment, the model can quickly respond to provide timely information for fault diagnosis and early warning, so that the operation and maintenance personnel can take timely measures to reduce the possibility of faults or reduce the loss caused by faults.

[0044] Secondly, the fault analysis system can continuously obtain real-time data, fuse and analyze multi-source data, thereby realizing real-time monitoring of the power grid equipment, discovering problems in the operation process of the equipment in a timely manner, and improving the reliability and stability of the equipment. Different types of data can complement each other to provide more comprehensive equipment operation information. For example, meteorological data can reflect the influence of the external environment on the equipment, and equipment real-time data can reflect the running state of the equipment itself. By comprehensively analyzing these data, the health status of the equipment can be more accurately evaluated, and potential fault risks can be found.

[0045] Compared with the prior art, the power grid equipment fault analysis method provided in the embodiments of the present application comprises the following steps: obtaining power grid equipment related historical data and meteorological historical data, preprocessing the power grid equipment related historical data and the meteorological historical data to obtain target feature data, constructing a snowflake model by using the target feature data, storing the snowflake model in a Hadoop cluster according to a preset format, establishing a global index based on the stored snowflake model, obtaining first training samples and second training samples according to the global index, constructing a fault prediction model by using the first training samples based on a recurrent neural network algorithm, constructing a fault diagnosis model by using the second training samples based on a support vector machine algorithm and a decision tree machine learning algorithm, obtaining power grid equipment real-time data and meteorological real-time data, inputting the power grid equipment real-time data into the fault diagnosis model and inputting the meteorological real-time data into the fault prediction model for fault diagnosis and early warning, and obtaining power grid equipment fault analysis results. The present application integrates various widely distributed and complex data sources in the power grid, constructs a highly structured and logically clear data architecture based on the snowflake model, greatly improves the data query and retrieval efficiency, and can quickly locate the key data related to fault analysis from massive data in a short time. At the same time, data storage is performed in the Hadoop distributed file system, which fully utilizes the powerful distributed storage, data redundancy and fault tolerance mechanism, greatly enriches the information dimension of fault analysis, ensures the high reliability and security of data, and provides a stable data foundation for power grid equipment fault analysis. Moreover, the present application comprehensively considers various complex factors of power grid equipment faults, constructs a comprehensive and adaptable model system, can accurately identify various complex fault types, effectively avoids fault diagnosis errors caused by single or inaccurate models, especially in complex conditions such as severe weather, improves the ability of fault prediction and diagnosis, and ensures the safe and stable operation of the power grid.

[0046] Further, as a refinement and expansion of the above embodiment, in order to completely describe the specific implementation process of the present embodiment, the present embodiment provides another power grid equipment fault analysis method, as shown in Figure 2 The method comprises the following steps:

[0047] 201、In the power grid equipment related historical data, structured data is extracted, and a minimum-maximum normalization method is used to calculate the structured data to obtain a plurality of normalized data.

[0048] In the embodiments of the present application, the fault analysis system extracts structured data from the power grid equipment related historical data. For each original data in the structured data, the maximum value and the minimum value of the original data are determined, and the minimum-maximum normalization method is used to calculate the original data, the maximum value and the minimum value of the original data to obtain normalized data, and the calculation formula is as follows formula 1:

[0049] Formula 1:

[0050] wherein a new denotes the normalized data, which ranges from 0 to 1, facilitating subsequent data analysis and model training, a denotes the original data, a max denotes the maximum value of the original data, a min denotes the minimum value of the original data. For example, the original data is one of the samples of the equipment account information, and the maximum value of the original data is the maximum value of the data characteristics of the sample. The maximum and minimum values of different types of data are different. Since the original units and magnitudes of different structured data can differ greatly, such as the voltage value of the equipment being several hundred volts and the current value being several amperes, the min-max normalization method can uniformly map the data to the [0, 1] interval, eliminate the dimensional differences, avoid the model training being biased towards the characteristics with large magnitudes due to different data magnitudes, and ensure that each characteristic has a reasonable weight in data analysis and model training.

[0051] Then, the fault analysis system normalizes each original data in the structured data to obtain a plurality of normalized data. The normalized data is more uniformly distributed, which can more efficiently update the parameters, make the model converge faster, and reduce the model training time and computational resource consumption.

[0052] 202、In the power grid equipment related historical data, unstructured data is extracted, and a convolutional neural network is used to extract features from the unstructured data and meteorological historical data to obtain a plurality of feature data.

[0053] In the embodiments of the present application, the fault analysis system extracts unstructured data from the power grid equipment related historical data, and uses a convolutional neural network to extract features from the unstructured data and meteorological historical data to obtain a plurality of feature data. For example, features can be extracted from equipment inspection images using a convolutional neural network (CNN). The CNN processes images through convolutional layers, pooling layers, and fully connected layers to output feature vectors of the images. For an m x n pixel image x, after a convolution operation with a convolution kernel ω of size p x q, a feature map C is obtained, and the calculation formula is as follows.

[0054] Formula 2:

[0055] wherein i and j denote the position coordinates of the feature map C, a and b are index variables for traversing the convolution kernel ω, ω(a, b) denotes the weight value of the convolution kernel at position (a, b), x(i + a, j + b) denotes the pixel value of the input image x at the corresponding position, τ is a bias term for adjusting the output value of the feature map. Then, a downsampling operation is performed through the pooling layer, and finally a feature vector is output through the fully connected layer.

[0056] Through the extracted rich and accurate feature data, the fault analysis system can more accurately judge the equipment running state and fault condition. For example, in analyzing the equipment inspection image features, potential equipment fault hidden dangers such as line aging and joint loosening can be more sensitively perceived, providing a reliable basis for fault diagnosis. For feature extraction of meteorological historical data, it is helpful to find the association rules between meteorological conditions and equipment faults. For example, by analyzing the meteorological image video features, combined with equipment running data, the possible faults of the equipment under adverse weather can be predicted in advance, so that preventive measures can be taken in advance to reduce the loss of faults. Compared with traditional manual feature extraction or simple algorithms, CNN can quickly process a large amount of unstructured data, extract valuable features in a short time, speed up the fault analysis process, and enable power grid operation and maintenance personnel to obtain analysis results in time and make decisions.

[0057] 203、Determine the mean data and the standard deviation using the plurality of normalized data and the plurality of feature data.

[0058] In the embodiments of the present application, the fault analysis system takes the plurality of normalized data and the plurality of feature data as target structure data, performs mean calculation on the target structure data to obtain mean data, and the calculation formula is formula 3 as follows:

[0059] Formula 3:

[0060] Wherein, μ represents the mean data, a1...an represents the plurality of normalized data, ai represents the i-th data sample value in the plurality of normalized data, and n represents the number of data samples in the plurality of normalized data. i represents the target structure data, a i represents the i-th data sample value in the target structure data, and n represents the number of data samples in the target structure data. By calculating the mean data, the central tendency of the target structure data can be clearly understood, providing a benchmark reference for subsequent analysis. Then, the fault analysis system calculates the standard deviation using the mean data and the target structure data, as formula 4 as follows:

[0061] Formula 4:

[0062] Wherein, σ represents the standard deviation, which is used to measure the dispersion degree of data, μ represents the mean data, x i represents the i-th data sample value in the target structure data, and n represents the number of data samples in the target structure data. The calculation of the standard deviation can accurately measure the dispersion degree of the data. In the power grid scenario, the equipment running state will fluctuate, and the data collected at different times will be different. The standard deviation can quantify the degree of difference and understand the data dispersion.

[0063] 204、According to the standard deviation, the mean data is corrected to obtain the target feature data.

[0064] In the embodiments of the present application, the fault analysis system acquires the maximum data distribution characteristic constant, determines the deletion range by using the standard deviation, the mean data and the maximum data distribution characteristic constant, and the following formula 5:

[0065] Formula 5: [μ-k'σ, μ+k'σ],

[0066] wherein σ represents the standard deviation, μ represents the mean data, and k' represents the maximum data distribution characteristic constant. Then, the data outside the deletion range in the target structure data is taken as the data to be deleted, and the data to be deleted is deleted in the target structure data to obtain the first corrected data. By setting the deletion range, the abnormal data deviating too far from the mean value can be identified and removed from the target structure data. In the power grid equipment data, these abnormal values can be caused by sensor failure, transmission error, etc. Removing them can avoid misleading the subsequent fault analysis and improve the reliability and authenticity of the data.

[0067] Subsequently, the fault analysis system acquires the target data distribution characteristic constant, determines the correction range by using the standard deviation, the mean data and the target data distribution characteristic constant, and the following formula 6:

[0068] Formula 6: [μ-kσ, μ+kσ],

[0069] wherein σ represents the standard deviation, μ represents the mean data, and k represents the target data distribution characteristic constant. Then, the data outside the correction range in the first corrected data is taken as the data to be corrected, and the data to be corrected in the first corrected data is corrected to obtain the target feature data. By using the correction range, the data outside a certain reasonable range but not directly deleted is corrected. When there is a small range of fluctuation anomalies in the power grid operation data, this method can make the data return to a reasonable interval, ensure the accuracy and consistency of the data, and provide a better data basis for fault analysis.

[0070] The above data correction process is essentially a deep cleaning of the collected data, which can effectively remove noise and error data, and at the same time fill or process missing data, so that the data is more complete and clean, reduces data errors, improves data quality, and makes the subsequent fault analysis model based on these data more stable and accurate.

[0071] In terms of data processing and feature extraction, the application discards the traditional way limited to a small number of data sources and simple data processing, and innovatively integrates various types of widely distributed and complex data sources in the power grid. Using the powerful distributed storage and computing capabilities of the Hadoop framework, it can efficiently process massive amounts of structured and unstructured data. For structured data such as device operating parameters and weather data, not only can large-scale data cleaning operations be performed quickly, but also abnormal values and missing values can be accurately identified and processed through self-defined distributed algorithms. In addition, advanced data analysis techniques can be used to deeply mine the potential associations between data. For unstructured data such as device inspection images and sound recordings, with the help of professional image processing and audio processing tools, combined with deep learning algorithms, highly representative features can be extracted. For example, using convolutional neural networks to extract features such as device component wear and deformation from device images, and using audio spectrum analysis techniques to identify abnormal noise frequency characteristics from device operating sounds, greatly enriching the information dimension of fault diagnosis and significantly improving the availability of data and the effectiveness of features, providing a comprehensive and accurate basis for subsequent fault diagnosis, effectively avoiding misdiagnosis or missed diagnosis problems caused by insufficient data processing or inaccurate feature extraction.

[0072] 205、adopt target feature data to construct a snowflake model.

[0073] In the embodiments of the application, the fault analysis system adopts target feature data to construct a snowflake model, wherein the snowflake model includes a fact table and a plurality of dimension tables, the plurality of dimension tables include a device dimension table, a time dimension table, a fault type dimension table, and a geographic location dimension table, and the fact table includes foreign keys for associating the device dimension table, the time dimension table, the fault type dimension table, and the geographic location dimension table. This structure follows a strict paradigm design, reduces data redundancy, and makes data storage more standardized. For example, the device dimension table centrally manages device-related attributes, avoiding repeated storage of the same device information in multiple places, improving data consistency and accuracy. Secondly, since the data is stored in different dimension tables according to the theme, when updating or maintaining the data, such as modifying device model information or adding new fault type descriptions, only the corresponding dimension table needs to be operated, without affecting other dimension tables and the fact table, reducing the difficulty and error probability of data maintenance, and facilitating long-term management of power grid data.

[0074] Taking the equipment dimension table as an example, its primary key is the equipment number, which includes attributes such as equipment name, model, manufacturer, and installation location. The time dimension table's primary key is the timestamp, containing time-level information such as year, month, day, hour, and minute. The fault type dimension table's primary key is the fault type code, containing detailed descriptions of various faults. The fact table focuses on equipment operation data, including electrical parameters and operating status of the equipment at different points in time, and is linked to related dimension tables such as the equipment and time dimension tables through foreign keys. The hierarchical structure of the snowflake model enables efficient data organization and association, facilitating subsequent data analysis and mining.

[0075] 206. Store the snowflake model in the Hadoop cluster according to the preset format.

[0076] In this embodiment, the fault analysis system determines the historical data volume, annual growth rate, and expected service life based on the target characteristic data, and calculates the estimated value of the fact table using the historical data volume, annual growth rate, and expected service life. The calculation formula is as follows: Formula 7:

[0077] Formula 7: S = V(1 + r) t ,

[0078] Here, S represents the estimated value of the fact table, V represents the historical data volume, r represents the annual growth rate, and t represents the expected service life. Next, the fault analysis system allocates storage space in the Hadoop cluster according to the estimated value of the fact table. This allows for advance planning of storage resources based on the historical data volume, growth trend, and expected service life of the power grid, avoiding waste or insufficiency of storage space. For example, more space can be allocated to the fact table of power grid equipment with rapidly growing data, ensuring the sustainability of data storage. Subsequently, the fault analysis system obtains the data storage format and partitioning strategy corresponding to the fact table from a preset format and stores the fact table in the Hadoop cluster according to this format and strategy. Similarly, it obtains the data storage format and partitioning strategy corresponding to each dimension table from a preset format and stores each dimension table in the Hadoop cluster according to its corresponding format and partitioning strategy. Hadoop can perform efficient computation and analysis on the data stored in the cluster. When building fault analysis models and mining data correlations, the powerful computing capabilities of the cluster can be utilized to quickly process large-scale data, shorten fault analysis time, and improve analysis efficiency.

[0079] For example, a corresponding database and table structure are created on a Hadoop cluster using a tool such as Hive or Spark SQL. When creating the table, the data storage format and partitioning strategy, such as partitioning by time dimension, are specified to improve storage efficiency and save storage space. For the device running data fact table, daily or hourly partitioning storage can be performed, so that when data in a specific time period is queried, the corresponding partition can be quickly located, reducing unnecessary data scanning.

[0080] In data modeling and organization, the present application constructs a highly structured and logically clear data architecture based on the snowflake model. Through careful design of multiple dimension tables and fact tables, device information, time information, geographic information, and various running data and fault data are organically integrated. Taking the device dimension table as an example, key information such as device model, manufacturer, installation time, and maintenance history is recorded in detail, and is closely associated with the fact table through the primary key; the time dimension table is accurate to the second or even millisecond level timestamp, facilitating tracking and analysis of the dynamic changes of the device running state. This data architecture greatly improves data query and retrieval efficiency, enabling quick positioning of key data related to fault diagnosis from massive data in a short time. At the same time, data is stored in the Hadoop distributed file system, fully utilizing its data redundancy and fault tolerance mechanism to ensure high reliability and security of the data, providing a stable data foundation for power grid device fault diagnosis and avoiding situations where data loss or damage affects the diagnosis result.

[0081] 207、Based on the stored snowflake model, a global index is established.

[0082] In the embodiments of the present application, the fault analysis system creates a global data index in Hive or other tools that support indexing to improve data query and retrieval efficiency. Optionally, for fields such as device number and timestamp that are often used as query conditions, B-tree index or bitmap index is created, which can greatly shorten the data query time. For example, a composite index based on device number and timestamp is created on the device running data fact table, which can quickly locate the relevant data records when querying the data of a certain device in a specific time period, avoiding full table scanning and greatly improving the query performance, enabling the operation and maintenance personnel to obtain the required fault analysis data in a timely manner.

[0083] Optionally, an inverted index can also be created for keyword fields in text data (such as device description information), which can effectively improve the accuracy of text data retrieval. Specifically, each keyword is mapped to a list of documents containing that keyword, i.e., data records. The term frequency–inverse document frequency (TF-IDF) algorithm is used to calculate the weight of the keyword. Let tf(k,d) be the term frequency of keyword k in a document, representing the number of times keyword k appears in document d. Let df(k) be the number of documents containing keyword k in document set D, used to count the number of documents containing the keyword. The total number of documents is N. Then, the TF-IDF weight calculation formula for keyword k in a document is as follows: Formula 8:

[0084] Formula 8:

[0085] Then, the index is sorted by weight to improve query efficiency, helping maintenance personnel quickly obtain detailed equipment information and assisting in fault analysis and diagnosis. It should be noted that unique indexes can be created for different data characteristics. Whether it's numerical equipment operating parameters or textual fault descriptions, appropriate indexes can improve query efficiency and enhance the fault analysis system's ability to process and adapt to various types of data. This application's embodiments do not impose specific limitations on these aspects.

[0086] 208. Obtain the first and second training samples based on the global index.

[0087] In this embodiment, the fault analysis system obtains historical severe weather data, equipment operation data under severe weather conditions, and equipment failure data under severe weather conditions based on a global index, and uses these data as the first training sample. Next, the system obtains normal equipment operation data and historical equipment failure data based on the global index, and uses these data as the second training sample. The global index provides an efficient search path for data querying, allowing the fault analysis system to quickly locate historical severe weather data, equipment operation data, and failure data without blindly searching through massive amounts of data. This significantly shortens the time for obtaining training samples, improves sample collection efficiency, and allows the fault analysis process to proceed more quickly. This high-quality, targeted training sample enables the model to learn more effective features and patterns. The first training sample is used to train the fault prediction model, enabling it to better capture the correlation between severe weather and equipment failure; the second training sample is used to train the fault diagnosis model, allowing it to accurately identify the differences between normal and faulty equipment operation states, thereby improving the accuracy and reliability of both models in the fault analysis task.

[0088] 209. Based on the recurrent neural network algorithm, the first training sample is used to construct a fault prediction model.

[0089] In the embodiments of the present application, a suitable artificial intelligence algorithm is selected, such as a recurrent neural network algorithm, historical severe weather data, equipment operation data in severe weather, and corresponding fault occurrence are taken as training samples, input into the artificial intelligence algorithm for training, the algorithm parameters are adjusted, the model performance is optimized, and the model can accurately predict the fault risk of the equipment in severe weather.

[0090] Specifically, the fault analysis system constructs an initial prediction model based on a recurrent neural network algorithm, the initial prediction model includes an input layer, a hidden layer and an output layer, the input sequence of the input layer is as follows formula 9:

[0091] Formula 9: X=(x1, x2, …, xT) T ),

[0092] Wherein, X is the input sequence, and T is the sequence length. The state update formula of the hidden layer is as follows formula 10:

[0093] Formula 10: h t =tanh(W ih x t +W hh h t-1 +b h ),

[0094] Wherein, W ih represents the weight matrix from the input layer to the hidden layer, with a dimension of h x x, h represents the number of neurons in the hidden layer, x represents the dimension of the input sequence, and each element W ih (i,j) in W ih represents the connection weight from the jth neuron of the input layer to the ith neuron of the hidden layer, W hh represents the weight matrix from the hidden layer to the hidden layer, with a dimension of h x h, each element W hh (i,j) in W hh represents the connection weight from the jth neuron of the hidden layer to the ith neuron of the hidden layer, b h represents the bias vector of the hidden layer, with a dimension of h x 1, and b h (i) represents the bias value of the ith neuron of the hidden layer. The prediction calculation formula of the output layer is as follows formula 11:

[0095] Formula 11: y t =W hy h t +b y ,

[0096] Wherein, W hyThis represents the weight matrix from the hidden layer to the output layer, with dimensions y×h, where y represents the dimension of the output data of the output layer, and W... hy (i,y) represents the connection weight from the j-th neuron in the hidden layer to the i-th neuron in the output layer, b y This represents the bias vector of the output layer, with dimensions y×1, b y (i) represents the bias value of the i-th neuron in the output layer. Next, the fault analysis system trains the initial prediction model using the first training sample and calculates the loss function, as shown in Formula 12 below:

[0097] Formula 12:

[0098] Where L represents the loss function, N represents the number of training samples in the first training sample, and y n Indicates the true label, This represents the predicted value output by the initial prediction model, which intuitively reflects the accuracy of the model's predictions and provides a clear direction for model optimization. Subsequently, the fault analysis system calculates the gradients of the loss function and the weight matrices from multiple input layers to hidden layers using the backpropagation algorithm, obtaining the updated weight matrices from multiple input layers to hidden layers. The calculation formula is shown in Formula 13 below:

[0099] Formula 13:

[0100] Among them, W ih ' represents the updated weight matrix from the input layer to the hidden layer, and α represents the learning rate, which controls the step size for each parameter update. The loss function is represented by W. ih The partial derivatives of the model allow the model to clearly know which parameters need to be adjusted and in what direction, providing a precise basis for subsequent parameter updates and improving the relevance and effectiveness of model training.

[0101] Finally, the fault analysis system, based on the gradient descent algorithm, updates the model parameters using the updated weight matrices from multiple input layers to hidden layers, thus obtaining the fault prediction model. By continuously updating the model parameters according to the gradient descent algorithm, the fault prediction model can gradually adjust the weight matrix and optimize its structure to better fit the features and patterns in the training data. As training progresses, the model's accuracy in predicting equipment faults continuously improves, enabling more precise early warnings of equipment failures and providing reliable support for power grid operation and maintenance.

[0102] 210. Based on the support vector machine algorithm and the decision tree machine learning algorithm, a fault diagnosis model is constructed using the second training sample.

[0103] In this embodiment, a fault diagnosis model is constructed based on normal equipment operation data and historical fault data using support vector machine algorithm and decision tree machine learning algorithm. Key features in the equipment operation data, such as current and voltage fluctuation features and equipment temperature change features, are extracted and used as input to the model. The fault type is used as output for model training, thereby improving the model's ability to identify different fault types.

[0104] Specifically, firstly, a fault type classification is established using a support vector machine. The fault analysis system extracts multiple feature vectors from the second training samples, along with the fault type label corresponding to each feature vector, as model training samples. The model training samples are (x′) i ,y′ i ), where i = 1, 2, ..., m, x′ i Represents a feature vector with dimension n, y′ i Indicates the fault type label, y′ i ∈{-1,1}.

[0105] Based on the support vector machine algorithm, the objective function of the quadratic programming problem is determined as shown in Formula 14 below:

[0106] Formula 14:

[0107] Where ω represents the hyperplane normal vector, V represents the penalty parameter used to balance margin maximization and classification error, and ξ i Let ξ represent the slack variable. i >0 allows a certain number of samples to be misclassified within the interval. This represents the sum of slack variables for the second training sample. The penalty parameter setting can flexibly balance the maximization of the interval and the classification error. When there are noisy or abnormal samples in the power grid fault data, the penalty parameter value can be adjusted to allow a certain degree of sample misclassification, avoiding the model from pursuing perfect classification and causing overfitting, so that the model can still maintain good performance under complex data conditions.

[0108] Based on the support vector machine algorithm, the constraints of the quadratic programming problem are determined as shown in Formula 15 below:

[0109] Formula 15: y′ i (ω T x′ i +b)≥1-ξ i ,

[0110] Where, y′ i Indicates the fault type label, x′ i Let ξ denote the eigenvector, ω denote the hyperplane normal vector, and ξ denote the eigenvector. i Let b represent the slack variable and b be the intercept.

[0111] The model training samples are used to solve the quadratic programming problem to obtain the optimal hyperplane normal vector and optimal intercept. The optimal hyperplane normal vector and optimal intercept are then used to determine the classification hyperplane, as shown in Formula 16 below:

[0112] Formula 16: (ω′) T x+b′=0,

[0113] Where ω′ represents the optimal hyperplane normal vector with a dimension of n×1, each element ω′(j) in ω′ represents the weight of the j-th dimension in the feature space, and b′ represents the optimal intercept.

[0114] Support Vector Machines (SVMs) excel at processing high-dimensional data. Power grid equipment operation data contains multiple dimensions of information, including current and voltage fluctuations, and equipment temperature variations. SVMs can find the optimal classification hyperplane in the high-dimensional feature space, accurately classifying different fault types and avoiding the curse of dimensionality. This effectively utilizes various features from the equipment operation data for fault diagnosis. Furthermore, by maximizing the classification margin, SVMs can produce models with good generalization ability. In power grid fault diagnosis scenarios, even when encountering new fault data that is not entirely identical to the training data, the model can still make accurate judgments due to its good generalization ability, reducing the risk of overfitting and improving the reliability of identifying different fault types.

[0115] Secondly, a decision tree is constructed to extract fault features and achieve more accurate classification. In the decision tree construction, indicators such as information gain or the Gini index are used to select features for node splitting. Taking information gain as an example, the entropy value of the second training sample is calculated using the following formula (Formula 17):

[0116] Formula 17:

[0117] Where Ent(D) represents the entropy value, D represents the second training sample, K represents the set of fault types, and p k This represents the proportion of samples in the second training sample that belong to the k-th type of fault.

[0118] Multiple feature vectors are extracted from the model training samples. For each feature vector, the information gain of the feature vector with respect to the second training sample is calculated using the following formula 18:

[0119] Formula 18:

[0120] Where Gain(D,A) represents information gain, Ent(D) represents the entropy value of the second training sample, D represents the second training sample, A represents the feature vector, V represents the number of values ​​in the feature vector, and D... v |D represents a subset of samples whose feature vectors take the value v. v| / |D| represents a subset of samples D v The proportion in the second training sample, Ent(D) v ) represents the sample subset D v The entropy value. By calculating information gain, the most effective features for distinguishing fault types can be selected from numerous equipment operation feature vectors. When the power grid equipment operation data contains a variety of complex features, information gain can quantify the contribution of each feature to fault classification, identify key features such as current and voltage fluctuation features that can significantly reduce the uncertainty of the sample set, avoid the use of redundant or irrelevant features, and improve the efficiency of model training and the accuracy of fault diagnosis.

[0121] Each feature vector is processed separately to obtain the information gain corresponding to each feature vector. The information gain with the largest value among multiple information gains is selected as the target information gain, and the feature vector corresponding to the target information gain among multiple feature vectors is used as the target splitting feature.

[0122] Based on the decision tree machine learning algorithm, a target splitting feature is recursively applied to the second training sample to construct the target decision tree. Information gain is then used to select features to build the decision tree, aiming to minimize the model's classification error on the training set. In power grid fault diagnosis scenarios, this means that the decision tree can more accurately classify different fault types, correctly categorize equipment fault data into the corresponding fault types, improve fault diagnosis accuracy, and provide reliable fault diagnosis results for maintenance personnel.

[0123] Decision trees present the decision-making process in a tree-like structure, and the generated rules are easy to understand and interpret. For power grid operation and maintenance personnel, they can intuitively see how different characteristics of equipment operating data are used to determine the fault type step by step, facilitating their understanding of fault diagnosis logic and making it easier to verify and adjust the model results. Moreover, decision trees can handle both numerical and categorical data, effectively processing various feature types in power grid equipment operating data, reducing redundant information interference, and improving the efficiency and accuracy of fault diagnosis.

[0124] Based on the above process, the fault analysis system constructs an initial fault diagnosis model using a classification hyperplane and a target decision tree, specifically by building the fault diagnosis model on a Hadoop cluster using support vector machine and decision tree algorithms. Then, based on cross-validation, a second set of training samples is used to train and optimize the initial fault diagnosis model, resulting in the final fault diagnosis model.

[0125] Support Vector Machines excel at classifying high-dimensional and complex data, while decision trees stand out in feature selection and rule interpretation. Combining the two leverages their respective strengths to enhance the overall performance of fault diagnosis models. This approach accurately identifies fault types and clearly explains the diagnostic process and its rationale, providing a more powerful tool for power grid equipment fault diagnosis. Furthermore, analyzing and learning from different perspectives using various algorithms can uncover more comprehensive fault characteristics and patterns, reducing the limitations of a single algorithm. This leads to more accurate identification of various fault types, improves fault diagnosis accuracy, and provides strong support for the reliable operation of power grid equipment.

[0126] In terms of fault diagnosis model construction, this application comprehensively considers various complex factors of power grid equipment faults, constructing a comprehensive and highly adaptable model system. For different types of equipment faults, various advanced machine learning and artificial intelligence algorithms are employed for targeted modeling. For example, for faults with obvious time-series characteristics, such as progressive aging faults, modeling can effectively capture the changing trends of equipment operating status over time. For faults based on the physical structure and electrical characteristics of equipment, support vector machines combined with professional domain knowledge are used for modeling. By carefully selecting appropriate kernel functions and parameters, fault categories are accurately classified. Simultaneously, during model training, a large amount of historical fault data and normal operation data are fully utilized for training and validation, continuously optimizing the model's parameters and structure to ensure high accuracy and generalization ability. This allows the model to accurately identify various complex fault types and effectively avoid fault diagnosis errors caused by a single or inaccurate model.

[0127] 211. Obtain real-time data of power grid equipment and real-time meteorological data. Input the real-time data of power grid equipment into the fault diagnosis model and the real-time meteorological data into the fault prediction model to perform fault diagnosis and early warning, and obtain the fault analysis results of power grid equipment.

[0128] In this embodiment, real-time operational data of power grid equipment is collected and input into a fault diagnosis model for analysis. When the fault diagnosis model determines that equipment may fail, it issues an early warning. Simultaneously, combined with severe weather forecast information, real-time meteorological data is input into a fault prediction model. If the fault prediction model predicts an increased risk of equipment failure under severe weather conditions, corresponding preventative measures are taken in advance, such as adjusting equipment operating parameters and arranging inspections. Specifically, the fault analysis system obtains a first preset format corresponding to the fault diagnosis model and adjusts the real-time data of the power grid equipment according to the first preset format, such as performing feature extraction and normalization. Then, the adjusted real-time data of the power grid equipment is input into the fault diagnosis model to obtain the equipment failure probability value output by the fault diagnosis model. A second preset format corresponding to the fault prediction model is obtained, and the real-time meteorological data is adjusted according to the second preset format. The adjusted real-time meteorological data is then input into the fault prediction model to obtain the equipment failure risk assessment value output by the fault prediction model. It should be noted that the preprocessing process for the real-time data of the power grid equipment and the real-time meteorological data is similar to the preprocessing process for the aforementioned historical data of the power grid equipment and historical meteorological data.

[0129] The system obtains a fault probability threshold and a fault risk threshold. When the equipment fault probability value is greater than or equal to the fault probability threshold, a fault diagnosis result is generated indicating that the power grid equipment has failed. For example, let p be the equipment fault probability output by the fault diagnosis model. When p ≥ p threshold When a potential equipment malfunction is detected, an early warning is issued, where p threshold This is a preset fault probability threshold, such as 0.7. When the equipment fault risk assessment value is greater than or equal to the fault risk threshold, a fault prediction result is generated indicating that the power grid equipment has a fault risk under the current weather conditions. For example, suppose the equipment fault risk assessment value output by the fault prediction model is r, when r ≥ r threshold When this occurs, corresponding preventative measures are triggered, where r threshold The preset fault risk threshold is determined based on the equipment's importance and risk tolerance. Finally, the fault diagnosis and prediction results are used as the fault analysis results for the power grid equipment. The fault analysis system integrates real-time data from power grid equipment and real-time meteorological data, comprehensively considering both the equipment's own condition and the impact of the external environment. Meteorological conditions often affect equipment operation; for example, severe weather may lead to equipment failure. Comprehensive analysis of both types of data can more accurately assess the probability of equipment failure, uncover potential fault correlations, and improve the comprehensiveness of fault analysis.

[0130] Optionally, this application employs efficient distributed computing and iterative optimization strategies. In a Hadoop cluster environment, complex computational tasks are decomposed into multiple sub-tasks for parallel processing, significantly shortening model training and diagnosis time. During fault diagnosis, a rule-based fast screening algorithm is first used to preliminarily judge the real-time collected data, eliminating some obviously normal situations. Then, suspected fault data is input into the trained model for in-depth analysis. By continuously iterating and updating the model's prediction results and evaluation indicators, and based on set strict thresholds and confidence levels, the system accurately determines whether equipment has malfunctioned and the type and severity of the fault. This iterative solution and diagnosis method effectively addresses the dynamic changes and uncertainties in the operating status of power grid equipment, providing rapid and accurate fault diagnosis results, timely decision support for maintenance personnel, significantly improving the efficiency and reliability of power grid equipment fault diagnosis, reducing power outage time and economic losses. It has extremely high application value and broad prospects for promotion in the power industry, powerfully driving the intelligent and efficient development of power grid equipment maintenance and management.

[0131] In summary, the fault analysis system achieves accurate diagnosis of power grid equipment faults through effective processing and analysis of multi-source heterogeneous data from the power system. Especially under complex conditions such as severe weather, it improves the ability to predict and diagnose faults, ensures the safe and stable operation of the power grid, effectively overcomes many shortcomings of existing technologies in data processing, model building, and diagnostic efficiency, and greatly enhances the accuracy and timeliness of power grid equipment fault diagnosis.

[0132] This application provides a method for power grid equipment fault analysis. Compared with existing technologies, this invention acquires historical data related to power grid equipment and historical meteorological data, preprocesses this data to obtain target feature data, constructs a snowflake model using the target feature data, stores the snowflake model in a preset format to a Hadoop cluster, and establishes a global index based on the stored snowflake model. First and second training samples are obtained based on the global index. A fault prediction model is constructed using the first training sample based on a recurrent neural network algorithm, and a fault diagnosis model is constructed using the second training sample based on a support vector machine algorithm and a decision tree machine learning algorithm. Real-time data from power grid equipment and real-time meteorological data are acquired, and the real-time data from power grid equipment is input into the fault diagnosis model, while the real-time meteorological data is input into the fault prediction model for fault diagnosis and early warning, respectively, to obtain the power grid equipment fault analysis results. This application integrates various widely distributed and complex data sources in the power grid, and constructs a highly structured and logically clear data architecture based on the snowflake model, significantly improving data query and retrieval efficiency, and enabling rapid location of key data related to fault analysis from massive amounts of data in a short time. Meanwhile, data is stored in the Hadoop Distributed File System, fully leveraging its powerful distributed storage, data redundancy, and fault tolerance mechanisms to greatly enrich the information dimensions of fault analysis, ensuring high data reliability and security, and providing a stable data foundation for power grid equipment fault analysis. Furthermore, this application comprehensively considers various complex factors of power grid equipment faults, constructing a comprehensive and highly adaptable model system capable of accurately identifying various complex fault types. This effectively avoids fault diagnosis errors caused by single or inaccurate models, especially under complex conditions such as severe weather, improving fault prediction and diagnosis capabilities and ensuring the safe and stable operation of the power grid.

[0133] Furthermore, as Figure 1 To specifically implement the method, this application provides a power grid equipment fault analysis device, such as... Figure 3 As shown, the device includes: a processing module 301, a storage module 302, a construction module 303, and an analysis module 304.

[0134] Processing module 301 is used to acquire historical data related to power grid equipment and historical meteorological data, and to preprocess the historical data related to power grid equipment and historical meteorological data to obtain target feature data;

[0135] Storage module 302 is used to construct a snowflake model using the target feature data, store the snowflake model in a preset format to a Hadoop cluster, and establish a global index based on the stored snowflake model;

[0136] The construction module 303 is used to obtain a first training sample and a second training sample according to the global index, construct a fault prediction model using the first training sample based on the recurrent neural network algorithm, and construct a fault diagnosis model using the second training sample based on the support vector machine algorithm and the decision tree machine learning algorithm.

[0137] The analysis module 304 is used to acquire real-time data of power grid equipment and real-time meteorological data. The real-time data of power grid equipment is input into the fault diagnosis model and the real-time meteorological data is input into the fault prediction model for fault diagnosis and early warning, so as to obtain the fault analysis results of power grid equipment.

[0138] In a specific application scenario, the processing module 301 is used to extract structured data from the historical data related to the power grid equipment; for each original data in the structured data, the maximum and minimum values ​​of the original data are determined, and a minimum-maximum normalization method is used to calculate the original data, the maximum and minimum values ​​of the original data, to obtain normalized data.

[0139]

[0140] Among them, a new Let a represent the normalized data, and let a represent the original data. max a represents the maximum value of the original data. min The minimum value of the original data is represented by [reference needed]. Multiple normalized data points are obtained by normalizing each original data point in the structured data. Unstructured data is extracted from the historical data related to power grid equipment, and features are extracted from the unstructured data and the historical meteorological data using a convolutional neural network to obtain multiple feature data points. The multiple normalized data points and the multiple feature data points are used as target structured data, and the mean of the target structured data is calculated to obtain the mean data.

[0141]

[0142] Where μ represents the mean data, a1...a i Represents the target structure data, a i Let represent the value of the i-th data sample in the target structure data, and n represent the number of data samples in the target structure data; the standard deviation is calculated using the mean data and the target structure data.

[0143]

[0144] Where σ represents the standard deviation, μ represents the mean data, and x iThe value of the i-th data sample in the target structure data is represented by , and n represents the number of data samples in the target structure data. The target structure data is corrected based on the standard deviation and the mean data to obtain the target feature data.

[0145] In specific application scenarios, the processing module 301 is used to obtain the maximum data distribution characteristic constant, and to determine the deletion range using the standard deviation, the mean data, and the maximum data distribution characteristic constant.

[0146] [μ-k′σ,μ+k′σ],

[0147] Where σ represents the standard deviation, μ represents the mean data, and k′ represents the maximum data distribution characteristic constant; data outside the deletion range in the target structure data are taken as data to be deleted, and the data to be deleted is deleted from the target structure data to obtain the first corrected data; the target data distribution characteristic constant is obtained, and the correction range is determined using the standard deviation, the mean data, and the target data distribution characteristic constant.

[0148] [μ-kσ,μ+kσ],

[0149] Where σ represents the standard deviation, μ represents the mean data, and k represents the target data distribution characteristic constant; the data outside the correction range in the first correction data are taken as the data to be corrected, and the data to be corrected in the first correction data are corrected to obtain the target feature data.

[0150] In a specific application scenario, the storage module 302 is used to construct a snowflake model using the target feature data. The snowflake model includes a fact table and multiple dimension tables, including a device dimension table, a time dimension table, a fault type dimension table, and a geographic location dimension table. The fact table includes foreign keys for linking the device dimension table, the time dimension table, the fault type dimension table, and the geographic location dimension table. Based on the target feature data, the historical data volume, annual growth rate, and expected service life are determined. The estimated value of the fact table is calculated using the historical data volume, the annual growth rate, and the expected service life.

[0151] S = V(1 + r) t ,

[0152] Wherein, S represents the estimated value of the fact table, V represents the amount of historical data, r represents the annual growth rate, and t represents the expected usage period; the storage space of the Hadoop cluster is allocated according to the estimated value of the fact table; the data storage format and partitioning strategy corresponding to the fact table are obtained from the preset format, and the fact table is stored in the Hadoop cluster according to the data storage format and partitioning strategy corresponding to the fact table; the data storage format and partitioning strategy corresponding to each dimension table are obtained from the preset format, and each dimension table is stored in the Hadoop cluster according to the data storage format and partitioning strategy corresponding to each dimension table.

[0153] In a specific application scenario, the construction module 303 is used to obtain historical severe weather data, equipment operation data under severe weather, and equipment failure data under severe weather according to the global index, and use the historical severe weather data, the equipment operation data under severe weather, and the equipment failure data under severe weather as the first training sample; obtain normal equipment operation data and historical equipment failure data according to the global index, and use the normal equipment operation data and the historical equipment failure data as the second training sample; construct an initial prediction model based on the recurrent neural network algorithm, the initial prediction model including an input layer, a hidden layer, and an output layer, and the input sequence of the input layer is:

[0154] X = (x1, x2, ..., x) T ),

[0155] Where X represents the input sequence, and T represents the sequence length; the state update formula of the hidden layer is:

[0156] h t =tanh(W ih x t +W hh h t-1 +b h ),

[0157] Among them, W ih W represents the weight matrix from the input layer to the hidden layer, with dimensions h×x, where h represents the number of neurons in the hidden layer, x represents the dimension of the input sequence, and W ih Each element W in ih (i,j) represents the connection weight from the j-th neuron in the input layer to the i-th neuron in the hidden layer, W hh W represents the weight matrix from the hidden layer to the hidden layer, with dimensions h×h. hh Each element W in hh (i,j) represents the connection weight from the j-th neuron in the hidden layer to the i-th neuron in the hidden layer, bh This represents the bias vector of the hidden layer, with dimensions h×1, b h (i) represents the bias value of the i-th neuron in the hidden layer; the prediction calculation formula for the output layer is:

[0158] y t =W hy h t +b y ,

[0159] Among them, W hy This represents the weight matrix from the hidden layer to the output layer, with dimensions y×h, where y represents the dimension of the output data of the output layer, and W... hy (i, y) represents the connection weight from the j-th neuron in the hidden layer to the i-th neuron in the output layer, b y This represents the bias vector of the output layer, with dimensions y×1, b y (i) represents the bias value of the i-th neuron in the output layer; the initial prediction model is trained using the first training sample, and the loss function is calculated.

[0160]

[0161] Where L represents the loss function, N represents the number of training samples for the first training sample, and y n Indicates the true label, This represents the predicted value output by the initial prediction model; the gradient of the loss function and the weight matrices from the input layers to the hidden layers are calculated using the backpropagation algorithm to obtain the updated weight matrices from the input layers to the hidden layers.

[0162]

[0163] Among them, W ih ' represents the updated weight matrix from the input layer to the hidden layer, and α represents the learning rate. The loss function is expressed as W ih The partial derivatives; based on the gradient descent algorithm, the updated weight matrices from the input layers to the hidden layers are used to update the model parameters, thereby obtaining the fault prediction model.

[0164] In a specific application scenario, the construction module 303 is used to extract multiple feature vectors and a fault type label corresponding to each feature vector from the second training sample, as model training samples, wherein the model training samples are (x′ i ,y′ i ), where i = 1, 2, ..., m, x′ i Represents a feature vector with dimension n, y′i Indicates the fault type label, y′ i ∈{-1,1}; Based on the support vector machine algorithm, determine the objective function of the quadratic programming problem.

[0165]

[0166] Where ω represents the hyperplane normal vector, C represents the penalty parameter, and ξ i Let ξ represent the slack variable. i >0, Let represent the sum of slack variables of the second training sample; based on the support vector machine algorithm, determine the constraints of the quadratic programming problem.

[0167] y′ i (ω T x′ i +b)≥1-ξ i ,

[0168] Where, y′ i Indicates the fault type label, x′ i Let ξ denote the eigenvector, ω denote the hyperplane normal vector, and ξ denote the eigenvector. i Let represent the slack variable, and b be the intercept. The model is used to train samples to solve the quadratic programming problem, yielding the optimal hyperplane normal vector and optimal intercept. The optimal hyperplane normal vector and optimal intercept are then used to determine the classification hyperplane.

[0169] (ω′) T x+b′=0,

[0170] Where ω′ represents the optimal hyperplane normal vector with dimension n×1, each element ω′(j) in ω′ represents the weight of the j-th dimension in the feature space, and b′ represents the optimal intercept; calculate the entropy value of the second training sample.

[0171]

[0172] Where Ent(D) represents the entropy value, D represents the second training sample, K represents the set of fault types, and p k This represents the proportion of samples belonging to the k-th type of fault in the second training sample; the multiple feature vectors are extracted from the model training sample; for each feature vector, the information gain of the feature vector with respect to the second training sample is calculated.

[0173]

[0174] Where Gain(D,A) represents information gain, Ent(D) represents the entropy value of the second training sample, D represents the second training sample, A represents the feature vector, V represents the number of values ​​of the feature vector, and D... v Denotes the subset of samples whose feature vector takes the value v, |D v | / |D| represents a subset of samples D v The proportion in the second training sample, Ent(D) v ) represents the sample subset D v The entropy value is calculated; each feature vector is processed to obtain the information gain corresponding to each feature vector; the information gain with the largest value among multiple information gains is selected as the target information gain, and the feature vector corresponding to the target information gain among the multiple feature vectors is used as the target splitting feature; based on the decision tree machine learning algorithm, the target splitting feature is used to recursively operate on the second training sample to construct the target decision tree; the classification hyperplane and the target decision tree are used to construct the initial fault diagnosis model; based on the cross-validation method, the second training sample is used to train and optimize the initial fault diagnosis model to obtain the fault diagnosis model.

[0175] In a specific application scenario, the analysis module 304 is used to obtain a first preset format corresponding to the fault diagnosis model, adjust the real-time data of the power grid equipment according to the first preset format, input the adjusted real-time data of the power grid equipment into the fault diagnosis model, and obtain the equipment fault probability value output by the fault diagnosis model; obtain a second preset format corresponding to the fault prediction model, adjust the real-time meteorological data according to the second preset format, input the adjusted real-time meteorological data into the fault prediction model, and obtain the equipment fault risk assessment value output by the fault prediction model; obtain a fault probability threshold and a fault risk threshold; when the equipment fault probability value is greater than or equal to the fault probability threshold, generate a fault diagnosis result indicating that the power grid equipment has a fault; when the equipment fault risk assessment value is greater than or equal to the fault risk threshold, generate a fault prediction result indicating that the power grid equipment has a fault risk under the current weather conditions; and use the fault diagnosis result and the fault prediction result as the fault analysis result of the power grid equipment.

[0176] This application provides an apparatus that, compared with the prior art, acquires historical data related to power grid equipment and historical meteorological data, preprocesses this data to obtain target feature data, constructs a snowflake model using the target feature data, stores the snowflake model in a preset format to a Hadoop cluster, and establishes a global index based on the stored snowflake model. First and second training samples are obtained based on the global index. A fault prediction model is constructed using the first training sample based on a recurrent neural network algorithm, and a fault diagnosis model is constructed using the second training sample based on a support vector machine algorithm and a decision tree machine learning algorithm. Real-time data from power grid equipment and real-time meteorological data are acquired, and the real-time data from power grid equipment is input into the fault diagnosis model, while the real-time meteorological data is input into the fault prediction model for fault diagnosis and early warning, respectively, to obtain the power grid equipment fault analysis results. This application integrates various widely distributed and complex data sources in the power grid, and constructs a highly structured and logically clear data architecture based on the snowflake model, which significantly improves data query and retrieval efficiency, enabling rapid location of key data related to fault analysis from massive amounts of data in a short time. Meanwhile, data is stored in the Hadoop Distributed File System, fully leveraging its powerful distributed storage, data redundancy, and fault tolerance mechanisms to greatly enrich the information dimensions of fault analysis, ensuring high data reliability and security, and providing a stable data foundation for power grid equipment fault analysis. Furthermore, this application comprehensively considers various complex factors of power grid equipment faults, constructing a comprehensive and highly adaptable model system capable of accurately identifying various complex fault types. This effectively avoids fault diagnosis errors caused by single or inaccurate models, especially under complex conditions such as severe weather, improving fault prediction and diagnosis capabilities and ensuring the safe and stable operation of the power grid.

[0177] It should be noted that other corresponding descriptions of the functional units involved in the power grid equipment fault analysis device provided in this application embodiment can be found by referring to... Figure 1 and Figure 2 The corresponding description in [the document] will not be repeated here.

[0178] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0179] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0180] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

[0181] In an exemplary embodiment, see Figure 4 The invention also provides a computer device including a bus, a processor, a memory, and a communication interface. It may also include an input / output interface and a display device, wherein the various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the power grid equipment fault analysis method described in the above embodiments.

[0182] A storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the power grid equipment fault analysis method.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented in hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0184] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application.

[0185] Those skilled in the art will understand that the modules in the apparatus of the implementation scenario can be distributed within the apparatus of the implementation scenario as described, or they can be located in one or more apparatuses different from this implementation scenario, with corresponding changes. The modules of the above-described implementation scenario can be combined into one module, or they can be further divided into multiple sub-modules.

[0186] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of the implementation scenario.

[0187] The above disclosures are only a few specific implementation scenarios of this application. However, this application is not limited to these. Any variations that can be conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for fault analysis of power grid equipment, characterized in that, include: Acquire historical data related to power grid equipment and historical meteorological data, and preprocess the historical data related to power grid equipment and historical meteorological data to obtain target feature data; A snowflake model is constructed using the target feature data, the snowflake model is stored in a Hadoop cluster according to a preset format, and a global index is built based on the stored snowflake model; The first training sample and the second training sample are obtained according to the global index. Based on the recurrent neural network algorithm, the first training sample is used to construct a fault prediction model. Based on the support vector machine algorithm and the decision tree machine learning algorithm, the second training sample is used to construct a fault diagnosis model. Real-time data of power grid equipment and real-time meteorological data are acquired. The real-time data of power grid equipment is input into the fault diagnosis model and the real-time meteorological data is input into the fault prediction model for fault diagnosis and early warning, respectively, to obtain the fault analysis results of power grid equipment.

2. The method according to claim 1, characterized in that, The preprocessing of historical data related to the power grid equipment and historical meteorological data to obtain target feature data includes: Extract structured data from the historical data related to the power grid equipment; For each original data point in the structured data, determine the maximum and minimum values ​​of the original data. Then, use a min-max normalization method to calculate the normalized data based on the original data, its maximum and minimum values. Among them, a new Let a represent the normalized data, and let a represent the original data. max a represents the maximum value of the original data. min This represents the minimum value of the original data; By normalizing each original data point in the structured data, multiple normalized data points are obtained. Unstructured data is extracted from the historical data related to the power grid equipment. A convolutional neural network is used to extract features from the unstructured data and the historical meteorological data to obtain multiple feature data. Using the multiple normalized data and the multiple feature data as target structure data, the mean of the target structure data is calculated to obtain the mean data. Where μ represents the mean data, a1...a i Represents the target structure data, a i This represents the value of the i-th data sample in the target structure data, and n represents the number of data samples in the target structure data; The standard deviation is calculated using the mean data and the target structure data. Where σ represents the standard deviation, μ represents the mean data, and x i This represents the value of the i-th data sample in the target structure data, and n represents the number of data samples in the target structure data; The target structure data is corrected based on the standard deviation and the mean data to obtain the target feature data.

3. The method according to claim 2, characterized in that, The step of correcting the target structure data based on the standard deviation and the mean data to obtain the target feature data includes: Obtain the maximum data distribution characteristic constant, and use the standard deviation, the mean data, and the maximum data distribution characteristic constant to determine the deletion range. [μ-k′σ,μ+k′σ], Where σ represents the standard deviation, μ represents the mean data, and k′ represents the maximum data distribution characteristic constant; Data outside the deletion range in the target structure data is taken as data to be deleted, and the data to be deleted is deleted from the target structure data to obtain the first corrected data; Obtain the target data distribution characteristic constant, and determine the correction range using the standard deviation, the mean data, and the target data distribution characteristic constant. [μ-kσ,μ+kσ], Where σ represents the standard deviation, μ represents the mean data, and k represents the distribution characteristic constant of the target data; The data outside the correction range in the first correction data are taken as the data to be corrected, and the data to be corrected in the first correction data are corrected to obtain the target feature data.

4. The method according to claim 1, characterized in that, The step of constructing a snowflake model using the target feature data and storing the snowflake model in a preset format to a Hadoop cluster includes: A snowflake model is constructed using the target feature data. The snowflake model includes a fact table and multiple dimension tables, including a device dimension table, a time dimension table, a fault type dimension table, and a geographic location dimension table. The fact table includes foreign keys for associating the device dimension table, the time dimension table, the fault type dimension table, and the geographic location dimension table. Based on the target characteristic data, determine the historical data volume, annual growth rate, and expected service life. Use the historical data volume, annual growth rate, and expected service life to calculate the estimated value of the fact table: S = V(1 + r). t , Where S represents the estimated value of the fact table, V represents the amount of historical data, r represents the annual growth rate, and t represents the expected number of years of use; The storage space of the Hadoop cluster is allocated according to the estimated values ​​in the fact table; Obtain the data storage format and partitioning strategy corresponding to the fact table from the preset format, and store the fact table in the Hadoop cluster according to the data storage format and partitioning strategy corresponding to the fact table; Obtain the data storage format and partitioning strategy corresponding to each dimension table from the preset format, and store each dimension table to the Hadoop cluster according to the data storage format and partitioning strategy corresponding to each dimension table.

5. The method according to claim 1, characterized in that, The step of obtaining the first training sample and the second training sample according to the global index, and constructing a fault prediction model using the first training sample based on the recurrent neural network algorithm includes: Historical severe weather data, equipment operation data under severe weather, and equipment failure data under severe weather are obtained based on the global index, and the historical severe weather data, equipment operation data under severe weather, and equipment failure data under severe weather are used as the first training sample. Based on the global index, obtain normal operating data and historical fault data of the device, and use the normal operating data and historical fault data of the device as the second training sample; An initial prediction model is constructed based on the recurrent neural network algorithm. The initial prediction model includes an input layer, a hidden layer, and an output layer. The input sequence of the input layer is as follows: X=(x1,x2,…,x T ), Where X represents the input sequence and T represents the sequence length; The state update formula for the hidden layer is: h t =tanh(W ih x t +W hh h t-1 +b h ), Among them, W ih W represents the weight matrix from the input layer to the hidden layer, with dimensions h×x, where h represents the number of neurons in the hidden layer, x represents the dimension of the input sequence, and W ih Each element W in ih (i,j) represents the connection weight from the j-th neuron in the input layer to the i-th neuron in the hidden layer, W hh W represents the weight matrix from the hidden layer to the hidden layer, with dimensions h×h. hh Each element W in hh (i,j) represents the connection weight from the j-th neuron in the hidden layer to the i-th neuron in the hidden layer, b h This represents the bias vector of the hidden layer, with dimensions h×1, b h (i) represents the bias value of the i-th neuron in the hidden layer; The prediction calculation formula for the output layer is: y t =W hy h t +b y , Among them, W hy This represents the weight matrix from the hidden layer to the output layer, with dimensions y×h, where y represents the dimension of the output data of the output layer, and W... hy (i, y) represents the connection weight from the j-th neuron in the hidden layer to the i-th neuron in the output layer, b y This represents the bias vector of the output layer, with dimensions y×1, b y (i) represents the bias value of the i-th neuron in the output layer; The initial prediction model is trained using the first training sample, and the loss function is calculated. Where L represents the loss function, N represents the number of training samples for the first training sample, and y n Indicates the true label, This represents the predicted value output by the initial prediction model; The gradients of the loss function and the weight matrices from the input layers to the hidden layers are calculated using the backpropagation algorithm to obtain the updated weight matrices from the input layers to the hidden layers. Among them, W ih ' represents the updated weight matrix from the input layer to the hidden layer, and α represents the learning rate. The loss function is expressed as w ih The partial derivatives; Based on the gradient descent algorithm, the updated weight matrices from the input layers to the hidden layers are used to update the model parameters, thereby obtaining the fault prediction model.

6. The method according to claim 1, characterized in that, The fault diagnosis model, constructed using the second training samples based on the support vector machine algorithm and decision tree machine learning algorithm, includes: Multiple feature vectors and a fault type label corresponding to each feature vector are extracted from the second training sample and used as model training samples, wherein the model training samples are (x′). i ,y′ i ), where i = 1, 2, ..., m, x′ i Represents a feature vector with dimension n, y′ i Indicates the fault type label, y′ i ∈{-1,1}; Based on the aforementioned support vector machine algorithm, the objective function for the quadratic programming problem is determined. Where ω represents the hyperplane normal vector, C represents the penalty parameter, and ξ i Let ξ represent the slack variable. i >0, This represents the sum of the slack variables of the second training sample; Based on the support vector machine algorithm, the constraints of the quadratic programming problem are determined. y′ i (oh T x′ i +b)≥1-ξ i , Where, y′ i Indicates the fault type label, x′ i Let ξ denote the eigenvector, ω denote the hyperplane normal vector, and ξ denote the eigenvector. i Let b represent the slack variable and b be the intercept. The model is used to train samples to solve the quadratic programming problem, obtaining the optimal hyperplane normal vector and optimal intercept. The optimal hyperplane normal vector and optimal intercept are then used to determine the classification hyperplane. (ω′) T x+b′=0, Where ω′ represents the optimal hyperplane normal vector with a dimension of n×1, each element ω′(j) in ω′ represents the weight of the j-th dimension in the feature space, and b′ represents the optimal intercept. Calculate the entropy value of the second training sample. Where Ent(D) represents the entropy value, D represents the second training sample, K represents the set of fault types, and p k This represents the proportion of samples in the second training sample that belong to the k-th type of fault; Extract the multiple feature vectors from the model training samples; For each feature vector, calculate the information gain of the feature vector with respect to the second training sample. Where Gain(D,A) represents information gain, Ent(D) represents the entropy value of the second training sample, D represents the second training sample, A represents the feature vector, V represents the number of values ​​of the feature vector, and D... v Denotes the subset of samples whose feature vector takes the value v, |D v | / |D| represents a subset of samples D v The proportion in the second training sample, Ent(D) v ) represents the sample subset D v The entropy value; Each feature vector is processed separately to obtain the information gain corresponding to each feature vector; Among multiple information gains, the information gain with the largest value is selected as the target information gain, and the feature vector corresponding to the target information gain among the multiple feature vectors is used as the target splitting feature. Based on the decision tree machine learning algorithm, the target splitting feature is used to recursively operate on the second training sample to construct the target decision tree; An initial fault diagnosis model is constructed using the classification hyperplane and the target decision tree. Based on the cross-validation method, the initial fault diagnosis model is trained and optimized using the second training sample to obtain the fault diagnosis model.

7. The method according to claim 1, characterized in that, The process involves inputting real-time data from the power grid equipment into the fault diagnosis model and real-time meteorological data into the fault prediction model for fault diagnosis and early warning, respectively, to obtain power grid equipment fault analysis results, including: Obtain the first preset format corresponding to the fault diagnosis model, adjust the real-time data of the power grid equipment according to the first preset format, input the adjusted real-time data of the power grid equipment into the fault diagnosis model, and obtain the equipment fault probability value output by the fault diagnosis model. Obtain the second preset format corresponding to the fault prediction model, adjust the real-time meteorological data according to the second preset format, input the adjusted real-time meteorological data into the fault prediction model, and obtain the equipment fault risk assessment value output by the fault prediction model. Obtain the failure probability threshold and the failure risk threshold; When the failure probability value of the equipment is greater than or equal to the failure probability threshold, a fault diagnosis result for the power grid equipment failure is generated. When the equipment failure risk assessment value is greater than or equal to the failure risk threshold, a failure prediction result is generated for the power grid equipment under the current weather conditions. The fault diagnosis results and the fault prediction results are used as the fault analysis results of the power grid equipment.

8. A power grid equipment fault analysis device, characterized in that, include: The processing module is used to acquire historical data related to power grid equipment and historical meteorological data, and to preprocess the historical data related to power grid equipment and historical meteorological data to obtain target feature data; The storage module is used to construct a snowflake model using the target feature data, store the snowflake model in a preset format to the Hadoop cluster, and establish a global index based on the stored snowflake model. The construction module is used to obtain a first training sample and a second training sample according to the global index, construct a fault prediction model using the first training sample based on the recurrent neural network algorithm, and construct a fault diagnosis model using the second training sample based on the support vector machine algorithm and the decision tree machine learning algorithm. The analysis module is used to acquire real-time data of power grid equipment and real-time meteorological data. The real-time data of power grid equipment is input into the fault diagnosis model and the real-time meteorological data is input into the fault prediction model for fault diagnosis and early warning, so as to obtain the fault analysis results of power grid equipment.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images