An intelligent power data retrieval system
The intelligent power data retrieval system utilizes twin neural networks and hash functions to process multi-source power data, combined with deep neural network encryption, to solve the problems of uniformity and security in multi-source power data retrieval, achieving efficient and accurate data retrieval and security assurance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- MAINTENANCE & TEST CENTRE CSG EHV POWER TRANSMISSION CO
- Filing Date
- 2026-02-02
- Publication Date
- 2026-06-12
AI Technical Summary
Existing power data retrieval technologies struggle to perform unified retrieval of multi-source power data, suffer from semantic biases, are unable to compare real-time or historical data, and lack security, making them prone to data leakage.
It employs a data acquisition module, a data processing module, a retrieval and interaction module, an intelligent matching module, and a retrieval security module. It achieves unified processing and secure retrieval of multi-source power data through twin neural networks and hash functions, and uses deep neural networks for encryption to generate variable keys to ensure data security.
It enables efficient and accurate retrieval of multi-source power data, improves retrieval accuracy and response speed, ensures data security, and meets the comparison needs of multi-source power data.
Smart Images
Figure CN122196037A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power data processing technology, and in particular to an intelligent power data retrieval system. Background Technology
[0002] With the digital transformation of the power industry, the scale of power data generated by the power system is constantly expanding, and the data types are becoming increasingly complex, including structured power generation data and transmission line parameter data, as well as unstructured equipment operation and maintenance reports, fault diagnosis records, and other multi-source power data. This power data is an important basis for power system optimization, fault early warning, and planning and design. How to quickly and accurately retrieve the required information from massive amounts of power data has become a major problem facing the power industry.
[0003] Existing power data retrieval technologies mostly employ traditional keyword matching methods, which have the following drawbacks: First, they struggle to perform unified retrieval of multi-source power data; second, traditional keyword matching is prone to discrepancies, such as identical keywords with different semantics or the same semantics but different keywords, failing to accurately understand the semantics of the user's search request and leading to missed or false detections; third, they cannot simultaneously retrieve real-time or historical multi-source power data and output related historical or real-time multi-source power data, failing to meet the comparison needs of multi-source power data; and fourth, they lack sufficient security, as power data involves the safety of power grid operation, and existing retrieval systems lack robust security protection mechanisms, making them susceptible to data leakage. Summary of the Invention
[0004] Therefore, the purpose of this invention is to provide an intelligent power data retrieval system to solve or at least partially solve the aforementioned problems existing in the prior art.
[0005] To achieve the above objectives, the present invention provides an intelligent power data retrieval system, the system comprising: Data acquisition module: used to collect real-time and historical multi-source power data and store the multi-source power data in the power database; Data processing module: used to preprocess the collected multi-source power data and extract power data features; Search interaction module: Used to receive user search request information and parse the search request information; Intelligent matching module: used to perform similarity matching between the parsed search request and the characteristics of power data; Search output module: Used to return the matching results of the intelligent matching module to the user in a preset format; Security Module: Used to ensure the security of multi-source power data during collection, transmission, storage and retrieval.
[0006] Furthermore, the data acquisition module collects multi-source power data through a sensor network and an internal interface of the power system. The multi-source power data includes power generation data, substation data, distribution data, power consumption data, and power equipment operation and maintenance data.
[0007] Furthermore, the data processing module is specifically used to perform the following steps: S11. Perform preprocessing operations on the collected multi-source power data, including data cleaning, data standardization, and format conversion; S12. The preprocessed multi-source power data is transformed into a multi-dimensional word vector representation, forming a real-time power data matrix and a historical power data matrix respectively. S13. The real-time power data matrix and the historical power data matrix are encoded using the dual-branch parameter sharing structure in the Siamese neural network. Then, an interactive attention mechanism is introduced to calculate the feature correlation weights between the real-time power data matrix and the historical power data matrix, and an interactive attention matrix is generated. S14. The interactive attention matrix is combined with the real-time power data matrix and the historical power data matrix to form the real-time weighted representation matrix and the historical weighted representation matrix, respectively. S15. Concatenate the row vectors of the real-time power data matrix and the real-time weighted representation matrix, and concatenate the row vectors of the historical power data matrix and the historical weighted representation matrix to form a new real-time power data matrix and a historical power data matrix. Then, extract power features from the real-time power data matrix and the historical power data matrix through the encoder network to obtain real-time power features and historical power features, and calculate the power feature similarity between the real-time power features and the historical power features.
[0008] Furthermore, the parsing of the retrieval request information specifically includes the following steps: S21. Perform preprocessing operations on the user-input search request, including cleaning, deduplication, and word segmentation. S22. The preprocessed retrieval request information is vectorized and then dimensionality-reduced. S23. The dimensionality-reduced retrieval request information is converted into hash values using a hash function.
[0009] Furthermore, the intelligent matching module is specifically used to perform the following steps: S31. Using a hash function, convert various real-time multi-source power data in the power database into hash values based on real-time power characteristics. This is consistent with the operations in steps S21-S23 above. The hash value is a unique identifier for various real-time multi-source power data. S32. Use the unique identifier as the primary key of the index for the corresponding real-time multi-source power data, so as to index the real-time multi-source power data based on the unique identifier; S33. Perform hash value similarity matching between the hash value of the retrieval request information and the hash value of various real-time multi-source power data, and index the real-time multi-source power data according to the hash value similarity matching results; S34. Based on the similarity of power characteristics, filter out historical multi-source power data related to the indexed real-time multi-source power data, and output the final real-time and historical multi-source power data.
[0010] Furthermore, the retrieval output module supports multiple output formats and batch export of retrieval results, including table format, chart format, document format, and data interface format.
[0011] Furthermore, the retrieval security module is specifically used to perform the following steps: S41. Using a deep neural network combined with time step and power data features, a variable key is generated, as follows:
[0012] in, It is a variable key. For deep neural networks, For power data characteristics, For time steps; S42. Encrypting multi-source power data using a variable key is represented as follows:
[0013] in, For encrypted multi-source power data, For multi-source power data, The hash value of the variable key; S43. Based on the encrypted multi-source power data, a decryption operation is performed on it. The decryption operation is used to restore the multi-source power data by restoring the variable key and performing a reverse encryption operation.
[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention proposes an intelligent power data retrieval system. It achieves comprehensive acquisition of multi-source power data through a data acquisition module, unifies the multi-source power data into structured data and extracts power data features through a data processing module, parses retrieval requests through a retrieval interaction module to improve retrieval accuracy, accurately matches multi-source power data corresponding to the retrieval request through an intelligent matching module, meets the retrieval output needs of different users through a retrieval output module, and ensures the security of multi-source power data acquisition, transmission, storage, and retrieval through a retrieval security module. This invention enables efficient retrieval of power data, improving retrieval accuracy and response speed. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a schematic diagram of the overall structure of an intelligent power data retrieval system provided in an embodiment of the present invention. Detailed Implementation
[0017] The principles and features of the present invention are described below with reference to the accompanying drawings. The listed embodiments are only used to explain the present invention and are not intended to limit the scope of the present invention.
[0018] Reference Figure 1 This embodiment provides an intelligent power data retrieval system, the system comprising: Data acquisition module: Used to collect real-time and historical multi-source power data and store the multi-source power data in the power database, specifically including: The data acquisition module is equipped with a real-time acquisition unit and a historical acquisition unit. The real-time acquisition unit acquires real-time multi-source power data through a sensor network (such as voltage sensors, current sensors, temperature sensors, etc.) and an internal interface of the power system. The historical acquisition unit acquires historical multi-source power data through a third-party data platform and supports batch export by time interval and data category. The multi-source power data includes power generation data, substation data, distribution data, power consumption data, and power equipment operation and maintenance data.
[0019] Data processing module: Used to preprocess the collected multi-source power data and extract power data features. Specifically, it performs the following steps: S11. Perform preprocessing operations on the collected multi-source power data, including data cleaning, data standardization, and format conversion. The data cleaning is used to identify anomalies in the multi-source power data using an anomaly detection algorithm, and remove redundant data, missing data, and abnormal data. The data standardization is used to unify multi-source power data of different formats into a preset format. The format conversion is used to convert unstructured multi-source power data into structured data.
[0020] S12. The preprocessed multi-source power data is transformed into a multi-dimensional word vector representation, forming a real-time power data matrix and a historical power data matrix, respectively. The multi-dimensional word vector representation can map multi-source power data to a vector space of a unified dimension, which can be directly used to construct real-time and historical power data matrices, avoiding matrix redundancy problems caused by multi-source power data encoding.
[0021] S13. The real-time power data matrix and the historical power data matrix are encoded using the two-branch parameter-sharing structure in the Siamese neural network. Then, an interactive attention mechanism is introduced to calculate the feature association weights between the real-time power data matrix and the historical power data matrix. Based on the feature association weights, the corresponding interactive attention matrix is generated, as shown below:
[0022] in, For interactive attention matrix, For real-time power data matrix, This is a matrix of historical power data.
[0023] The dual-branch parameter-sharing structure maps real-time and historical power data matrices to the same feature space, facilitating the comparison of feature differences between the two matrices. The interactive attention mechanism enables the Siamese neural network to automatically identify key feature dimensions in the real-time and historical power data matrices and assign them higher attention weights, thus mitigating the interference of noisy features.
[0024] S14. The interactive attention matrix is combined with the real-time power data matrix and the historical power data matrix to form the real-time weighted representation matrix and the historical weighted representation matrix, respectively.
[0025] S15. Concatenate the row vectors of the real-time power data matrix and the real-time weighted representation matrix, and concatenate the row vectors of the historical power data matrix and the historical weighted representation matrix to form new real-time power data matrices and historical power data matrices. Then, use a Transformer encoder network to extract power features from both matrices, obtaining real-time and historical power features, and calculate the power feature similarity between them. The Transformer encoder network can focus on power feature associations in different dimensions, and can establish global associations by calculating the attention weights of any two elements in the real-time and historical power data matrices, without relying on sequence order or local window limitations, thus accurately capturing long-range dependency features in the real-time and historical power data matrices.
[0026] Search interaction module: Used to receive user search request information and parse the search request information, specifically to perform the following steps: S21. Perform preprocessing operations on the user-input search request. The preprocessing operations include cleaning, deduplication, and word segmentation. Cleaning removes invalid information and noisy data from the search request, thus purifying the user-input search request. Deduplication removes duplicate keyword combinations or completely identical search statements from the search request to prevent the search engine from performing repeated calculations on the same content. Word segmentation breaks down the user-input search request into lexical units with independent semantics.
[0027] S22. The preprocessed retrieval request information is vectorized and then dimensionality-reduced. The vectorization represents the lexical units or search terms in the retrieval request as high-dimensional vectors, transforming the retrieval request into computationally achievable numerical features, thus improving retrieval accuracy and efficiency. The dimensionality reduction maps the high-dimensional vector retrieval request to a low-dimensional space, improving retrieval stability and reducing overfitting risk and storage resource consumption of the retrieval system.
[0028] S23. The dimensionality-reduced retrieval request information is converted into hash values using a hash function.
[0029] The intelligent matching module is used to perform similarity matching between the parsed search request and the characteristics of power data, specifically by executing the following steps: S31. Using a hash function, convert various real-time / historical multi-source power data in the power database into hash values based on real-time power characteristics. This is consistent with the operations in steps S21-S23 above. The hash value is a unique identifier for various types of real-time multi-source power data.
[0030] The retrieval request information differs in form from various real-time / historical multi-source power data, making direct matching complex and inaccurate. Therefore, by converting the retrieval request information and various real-time / historical multi-source power data into hash values and mapping them to the same fixed-length space to form standardized feature identifiers, the equivalent semantic comparison between the retrieval request information and various real-time / historical multi-source power data can be achieved.
[0031] S32. Use the unique identifier as the primary key of the index for the corresponding real-time / historical multi-source power data, and use it to index the real-time multi-source power data based on the unique identifier.
[0032] S33. Perform hash value similarity matching between the hash value of the retrieval request information and the hash values of various real-time / historical multi-source power data, and index the real-time / historical multi-source power data according to the hash value similarity matching results.
[0033] S34. Based on the similarity of power characteristics, filter out historical / real-time multi-source power data related to the indexed real-time / historical multi-source power data, and output the final real-time and historical multi-source power data. This facilitates users in comparing the differences between multi-source power data that are related but from different times, and allows users to perform unilateral searches while simultaneously filtering and outputting real-time and historical multi-source power data based on the similarity of power characteristics. This reduces the computational cost of the retrieval system and improves retrieval speed.
[0034] The search output module is used to return the matching results from the intelligent matching module to the user in a preset format, specifically including: The search output module supports multiple output formats, batch export of search results, online preview and printing. The output formats include table format, chart format, document format and data interface format, etc. Users can select the corresponding output format according to their actual needs.
[0035] The retrieval security module also includes a role-based access control mechanism, which can assign different access permissions to different roles, restricting the scope of user operations on power data. It is also used to ensure the security of multi-source power data during collection, transmission, storage, and retrieval, preventing data leakage, tampering, or damage. Specifically, it is used to execute the following steps: S41. Using a deep neural network combined with time step and power data features, a variable key is generated, as follows:
[0036] The deep neural network is used to minimize the probability of variable key collisions. , means as follows:
[0037] in, It is a variable key. For deep neural networks, For power data characteristics, For time steps, For variable keys With variable key Equal probabilities; S42. Encrypting multi-source power data using a variable key is represented as follows:
[0038]
[0039] in, For encrypted multi-source power data, For multi-source power data, The hash value of the variable key. It is a safe hash algorithm; S43. Based on the encrypted multi-source power data, a decryption operation is performed. The decryption operation is used to restore the multi-source power data by restoring the variable key and performing a reverse encryption operation, as shown below:
[0040] Calculate the loss function for the decryption operation and optimize the loss function to ensure consistency between encryption and decryption operations, as shown below:
[0041] in, The loss function for the decryption operation. and These are the functions for decryption and encryption, respectively.
[0042] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A smart power data retrieval system, characterized in that, The system includes: Data acquisition module: used to collect real-time and historical multi-source power data and store the multi-source power data in the power database; Data processing module: used to preprocess the collected multi-source power data and extract power data features; Search interaction module: Used to receive user search request information and parse the search request information; Intelligent matching module: used to perform similarity matching between the parsed search request and the characteristics of power data; Search output module: Used to return the matching results of the intelligent matching module to the user in a preset format; Security Module: Used to ensure the security of multi-source power data during the collection, transmission, storage and retrieval process.
2. The intelligent power data retrieval system according to claim 1, characterized in that, The data acquisition module collects multi-source power data through a sensor network and an internal interface of the power system. The multi-source power data includes power generation data, substation data, distribution data, power consumption data, and power equipment operation and maintenance data.
3. The intelligent power data retrieval system according to claim 1, characterized in that, The data processing module is specifically used to perform the following steps: S11. Perform preprocessing operations on the collected multi-source power data, including data cleaning, data standardization, and format conversion; S12. The preprocessed multi-source power data is transformed into a multi-dimensional word vector representation, forming a real-time power data matrix and a historical power data matrix respectively. S13. The real-time power data matrix and the historical power data matrix are encoded using the dual-branch parameter sharing structure in the Siamese neural network. Then, an interactive attention mechanism is introduced to calculate the feature correlation weights between the real-time power data matrix and the historical power data matrix, and an interactive attention matrix is generated. S14. The interactive attention matrix is combined with the real-time power data matrix and the historical power data matrix to form the real-time weighted representation matrix and the historical weighted representation matrix, respectively. S15. Concatenate the row vectors of the real-time power data matrix and the real-time weighted representation matrix, and concatenate the row vectors of the historical power data matrix and the historical weighted representation matrix to form a new real-time power data matrix and a historical power data matrix. Then, extract power features from the real-time power data matrix and the historical power data matrix through the encoder network to obtain real-time power features and historical power features, and calculate the power feature similarity between the real-time power features and the historical power features.
4. The intelligent power data retrieval system according to claim 3, characterized in that, The parsing of the retrieval request information specifically includes the following steps: S21. Perform preprocessing operations on the user-input search request, including cleaning, deduplication, and word segmentation. S22. The preprocessed retrieval request information is vectorized and then dimensionality-reduced. S23. The dimensionality-reduced retrieval request information is converted into hash values using a hash function.
5. The intelligent power data retrieval system according to claim 4, characterized in that, The intelligent matching module is specifically used to perform the following steps: S31. Using a hash function, convert various real-time multi-source power data in the power database into hash values based on real-time power characteristics. This is consistent with the operations in steps S21-S23 above. The hash value is a unique identifier for various real-time multi-source power data. S32. Use the unique identifier as the primary key of the index for the corresponding real-time multi-source power data, so as to index the real-time multi-source power data based on the unique identifier; S33. Perform hash value similarity matching between the hash value of the retrieval request information and the hash value of various real-time multi-source power data, and index the real-time multi-source power data according to the hash value similarity matching results; S34. Based on the similarity of power characteristics, filter out historical multi-source power data related to the indexed real-time multi-source power data, and output the final real-time and historical multi-source power data.
6. The intelligent power data retrieval system according to claim 1, characterized in that, The search output module supports multiple output formats and batch export of search results. The output formats include table format, chart format, document format, and data interface format.
7. The intelligent power data retrieval system according to claim 1, characterized in that, The retrieval security module is specifically used to perform the following steps: S41. Using a deep neural network combined with time step and power data features, a variable key is generated, as follows: in, It is a variable key. For deep neural networks, For power data characteristics, For time steps; S42. Encrypting multi-source power data using a variable key is represented as follows: in, For encrypted multi-source power data, For multi-source power data, The hash value of the variable key; S43. Based on the encrypted multi-source power data, a decryption operation is performed on it. The decryption operation is used to restore the multi-source power data by restoring the variable key and performing a reverse encryption operation.