Multi-modal feature integration method for heterogeneous data of power grid safety production review

Abstract

The invention discloses a power grid safety production review heterogeneous data multi-modal feature integration method, relates to the technical field of power grid heterogeneous data processing, and solves the problems that in the prior art, an existing multi-modal data integration method is generally high in calculation complexity, and the efficiency is reduced along with the increase of the data volume. According to the method, the feature correlation degree is analyzed by introducing the Pearson correlation coefficient, the mixed attribute feature matching is completed in combination with the normalization processing, and the accurate integration of the multi-modal features is realized by relying on the adaptive feature selection model based on the channel attention; on the basis, redundant information is eliminated by adopting a maximum matching method, and an integration result is optimized in combination with normalization processing of a trust function, so that the efficiency and accuracy of feature integration are remarkably improved, and efficient data support is provided for auditing of a power grid safety production risk management system.

Description

Technical Field

[0001] This invention relates to the field of heterogeneous data processing technology for power grids, and in particular to a method for integrating multimodal features of heterogeneous data in power grid safety production review. Background Technology

[0002] Data for power grid safety production review originates from multiple information systems and business platforms, encompassing various types such as real-time power grid monitoring data, operation log data, business guidelines, and approval forms. This data exhibits significant multi-source and heterogeneity; each data source and presentation format can be considered an independent modality, collectively forming a complex data system describing power grid safety production review. To fully leverage the value of this data and improve the accuracy and comprehensiveness of power grid safety risk early warning and decision analysis, effective integration of heterogeneous data becomes a crucial step.

[0003] Multimodal data for power grid safety production review can characterize the risk status of power grid safety production from different dimensions. Through information complementarity between modes, the accuracy of perception of power grid safety production risks can be significantly improved. Therefore, multimodal data integration technology has become a research hotspot in the field of power grid safety production management. However, existing technologies still have many limitations in practical applications: Zhong Yufeng et al. proposed a scale-invariant feature transformation algorithm based on edge point features, which achieves high-precision registration by using the distance between bilateral feature points as a confidence constraint. However, this algorithm requires high computing resources and needs to traverse all feature point pairs, resulting in long data integration time and difficulty in meeting the real-time requirements of large-scale data processing; Zhang Yan proposed a multimodal data feature fusion algorithm based on deep learning and DS theory, which selects features by constructing a model structure, calculates the distance between heterogeneous data and features, and combines DS theory to complete the fusion. However, this method requires designing independent RBM structures for different modalities, and the parameter tuning process is complex, which significantly restricts the fusion efficiency; Yang Qiuju proposed a multi-source heterogeneous sensor data fusion model based on fuzzy theory, which reduces computational complexity by using probability allocation functions and distance matrices, and uses binary functions to transform data sources and calculate support values ​​to fuse duplicate data. However, it requires calculating support values ​​for each data source and judging data conflict based on support values ​​and OWA weights, which leads to a sharp increase in time complexity as the amount of data increases; Wu Zonghang et al. proposed a joint SSL-based model... The multimodal fusion recommendation algorithm SFELMMR enhances data representation capabilities by jointly learning features from different modalities to alleviate data sparsity and fuse global and local perspectives. However, this method requires analyzing the potential correlations of each item, and the data integration time increases significantly with the amount of data. In summary, existing multimodal data integration methods generally suffer from high computational complexity and decreased efficiency with increasing data volume, making it difficult to meet the demand for efficient integration of large-scale heterogeneous data in the field of power grid safety production review.

[0004] Therefore, a method for integrating multimodal features of heterogeneous data in power grid safety production review is needed. Summary of the Invention

[0005] To address the common problems of high computational complexity and decreasing efficiency with increasing data volume in existing multimodal data integration methods, this invention provides a method for integrating multimodal features of heterogeneous data from power grid safety production review. This method constructs a multimodal feature matching matrix by analyzing feature correlations and employs an adaptive feature selection model based on channel attention to achieve multimodal feature integration, ultimately yielding efficient and accurate multimodal data feature integration results. The specific technical solution is as follows: A method for integrating multimodal features of heterogeneous data in power grid safety production review includes the following steps: Feature extraction: The heterogeneous data of power grid safety production review to be processed are classified into independent datasets, and then feature extraction is performed and the feature values ​​are normalized. Hybrid attribute feature matching: Calculate the correlation coefficient between multimodal features, and enhance the heterogeneous data of power grid safety production review based on the correlation coefficient to form a feature matching matrix. Then, save the correlation information stored in the feature matching matrix into the spatial basis of the corresponding two modes, and determine whether the feature vectors are assigned to the same feature set based on the central cosine similarity. Multimodal feature integration: After directly concatenating the features, a channel attention mechanism is used to obtain the weight vector. Then, a residual connection is used to add the original features and the weighted features to obtain the integrated features.

[0006] Preferably, the process of mixed attribute feature matching includes the following steps: The correlation magnitude of the multimodal features of safety production review data is calculated using the normalized Pearson correlation coefficient, as shown in the following formula: In the formula, To determine the correlation between the features of multimodal safety production review data, and All of these are multimodal safety production review data samples. For data The average value, For data The average value, For data standard deviation For data Standard deviation; Enhancement processing is applied to the power grid safety production review data to form a feature matching matrix. The calculation formula is shown in the following formula: In the formula, To enhance the initial feature matrix before processing, L ( X ) for The feature matching matrix obtained after performing enhancement processing; Subsequently, the relevance information stored in the feature matching matrix is ​​saved to the spatial basis of the corresponding two modalities; Calculate the center cosine similarity of the power grid safety production review data. When the difference between the calculated center cosine similarity result and 1 is less than the set minimum threshold, they are grouped into the same feature set and stored together; when the difference between the result and 1 is greater than the set maximum threshold, they are stored separately.

[0007] The preferred formula for calculating central cosine similarity is as follows: In the formula, Represents the central cosine similarity. and They represent different spatial bases. This indicates the number of feature vectors.

[0008] Preferably, the weight vector is obtained in the following way: In the formula, The weight vectors generated for the channel attention mechanism are used to represent the importance of different features. For fully connected layer operations, For average pooling operation, For feature cascading operations, As a unique integration feature of modality, These are integrated features shared by all modalities.

[0009] The preferred method for obtaining integrated features is as follows: In the formula, These are the initial integrated features after optimization.

[0010] Preferably, the following steps are also included: Redundancy processing of multi-source heterogeneous data: The maximum matching method is used to remove redundant information in the original power grid safety production data, and a trust function is introduced to quantify the credibility and correlation between different modal data. The normalized trust function is integrated with the redundant multimodal data features to obtain the final integrated result of the multimodal data features.

[0011] Preferably, the process for obtaining the final integration result is as follows: The maximum matching method is used to remove redundant information from the original power grid safety production data, as shown in the following formula: In the formula, This is a set of multimodal data features after redundancy processing. To assess the correlation between features in multimodal data, Here is the criterion function used to select the optimal feature subset. denoted as the standard deviation of this characteristic distribution.

[0012] The trust function for heterogeneous data is calculated using the Pearson correlation coefficient. As shown below: In the formula, Represents the trust function. This represents the set of multimodal features after redundancy processing. Indicates the corresponding target sample. express The average value, express The average value, express standard deviation express Standard deviation; The trust function is normalized, and then integrated with the redundant multimodal data features to obtain the final integrated result of the multimodal data features, as shown in the following formula: In the formula, The final result of integrating multimodal data features. This is the trust function.

[0013] A computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the method for integrating heterogeneous data multimodal features for power grid safety production review as described above.

[0014] A processor for running a program, wherein the program executes the method for integrating multimodal features of heterogeneous data for power grid safety production review as described above.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention introduces Pearson correlation coefficient analysis to determine feature correlation and combines it with normalization processing to complete mixed attribute feature matching. It relies on an adaptive feature selection model based on channel attention to achieve accurate integration of multimodal features. On this basis, it uses the maximum matching method to eliminate redundant information and combines the normalization processing of the trust function to optimize the integration results, which significantly improves the efficiency and accuracy of feature integration and provides efficient data support for the audit of the power grid safety production risk management system. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 This is a schematic diagram of the multimodal feature integration model structure; Figure 2 This is a schematic diagram illustrating the distribution of data features. Figure 3 This is a diagram illustrating the tightness of the match. Figure 4 This is a diagram showing the comparison of data integration time. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] In one embodiment of the present invention, a method for integrating multimodal features of heterogeneous data in power grid safety production review is provided, comprising the following steps: Feature extraction: The heterogeneous data of power grid safety production review to be processed are classified into independent datasets, and then feature extraction is performed and the feature values ​​are normalized. Hybrid attribute feature matching: Calculate the correlation coefficient between multimodal features, and enhance the heterogeneous data of power grid safety production review based on the correlation coefficient to form a feature matching matrix. Then, save the correlation information stored in the feature matching matrix into the spatial basis of the corresponding two modes, and determine whether the feature vectors are assigned to the same feature set based on the central cosine similarity. Multimodal feature integration: After directly concatenating the features, a channel attention mechanism is used to obtain the weight vector. Then, a residual connection is used to add the original features and the weighted features to obtain the integrated features. Redundancy processing of multi-source heterogeneous data: The maximum matching method is used to remove redundant information in the original power grid safety production data, and a trust function is introduced to quantify the credibility and correlation between different modal data. The normalized trust function is integrated with the redundant multimodal data features to obtain the final integrated result of the multimodal data features.

[0023] The following is a detailed explanation of the four steps above: 1. Feature Extraction of Heterogeneous Data for Power Grid Safety Production In the actual operation of the power grid, the data for power grid safety production review comes from multiple power subsystems. Due to the differences in data encoding rules and data storage formats used by each subsystem, the collected power grid safety production review data exhibits significant multimodal characteristics. This type of multimodal data not only mixes multiple attributes such as numerical and categorical data, but also has high-dimensional features. Therefore, dimensionality reduction processing is required before feature extraction to improve data processing efficiency and accuracy.

[0024] First, the heterogeneous data of power grid safety production review to be processed is divided into two categories: numerical data and categorical data, which form independent data sets respectively. The preliminary feature measurement is achieved by calculating the distance between the data, and the calculation formula is shown in Equation (1).

[0025] (1) In equation (1), Indicates the distance between safety production review data. and Representing different attributes of data, Indicates the number of data sets.

[0026] After dimensionality reduction, the multimodal power grid safety production review data can be represented as a sample containing n attributes. For this sample, the present invention uses a deep learning model based on an encoder-decoder structure for feature extraction. This structure can effectively filter and retain key features in the data through bidirectional processing of encoding compression and decoding reconstruction. The loss function of its feature extraction process is defined as shown in formula (2).

[0027] (2) In equation (2), This is the loss function, used to measure the error in feature extraction. This refers to the category information to which the actual feature belongs. The probability of the feature category output by the model.

[0028] Considering that the feature values ​​of different safety production review data have large differences in dimensions and numerical ranges, the extracted features need to be normalized to eliminate the influence of dimensions, as shown in formula (3).

[0029] (3) In formula (3), For the input safety production review data feature values, The mean of the input features. The variance values ​​of the input features. To avoid constants with a denominator of 0 (the value is 1e-8).

[0030] Through the above normalization process, feature values ​​of different dimensions can be unified into a relatively consistent numerical range, ensuring that the selected features are representative and comparable, and laying a reliable data foundation for the subsequent mixed attribute feature matching process.

[0031] 2. Matching Heterogeneous Data with Mixed Attribute Features in Power Grid Safety Production Review After extracting features from the power grid safety production review data, it is necessary to establish the correlation between different modal features through feature matching, laying the foundation for subsequent feature integration. This step achieves accurate matching of mixed attribute features by analyzing feature correlation, constructing a matching matrix, and calculating similarity.

[0032] First, the Pearson correlation coefficient is obtained by analyzing the correlation between multimodal features, and a multimodal feature matching matrix is ​​constructed based on this coefficient. The feature matching module uses the normalized Pearson correlation coefficient to calculate the correlation magnitude of the multimodal features of the safety production review data, as shown in formula (4).

[0033] (4) In formula (4), To determine the correlation between the features of multimodal safety production review data, This is a sample of data for multimodal safety production review. The average value of the data. Let S be the standard deviation of the data. Here, S represents two data samples, and Formula 4 calculates the correlation between these two data samples. i is the average value of data S1. i is the standard deviation of data S1; j is the average value of data S2. j is the standard deviation of data S2. Based on the feature correlation magnitude calculated above, the power grid safety production review data is enhanced to strengthen the correlation features between modes and form a feature matching matrix. The calculation formula is shown in formula (5).

[0034] (5) In equation (5), To enhance the feature matching matrix obtained after processing, the feature matching matrix contains the correlation of multimodal features of safety production data and its guiding information on clustering.

[11] Subsequently, the relevance information stored in the feature matching matrix is ​​saved into the spatial basis of the corresponding two modalities.

[12] Furthermore, the accurate matching of mixed attribute features is achieved by calculating the central cosine similarity of the power grid safety production review data, and the calculation formula is shown in Equation (6).

[0035] (6) In equation (6), Represents the central cosine similarity. and They represent different spatial bases. This indicates the number of feature vectors.

[0036] It should be noted that the central cosine similarity calculation result is used to determine the degree of correlation between feature vectors: when the result is close to 1, it indicates that the two feature vectors are highly correlated and can be classified into the same feature set; when the result is far from 1, it indicates that the feature vectors are not highly correlated and need to be stored separately. In actual power grid safety production review data processing, a baseline value needs to be set for the target feature, and inconsistent attribute values ​​in the feature should be iteratively updated until the central cosine similarity is close to 1 to ensure the accuracy of feature matching.

[0037] By constructing a multimodal feature matching matrix, calculating feature correlation using the normalized Pearson correlation coefficient, enhancing data based on correlation, and then using the central cosine similarity to quantify the correlation degree of feature vectors, the matching process of mixed attribute features is completed, providing a reliable basis for the initial integration of multimodal features of heterogeneous data.

[0038] 3. Preliminary integration of multimodal features from heterogeneous data To effectively utilize heterogeneous data from power grid safety production reviews, it is necessary to integrate the extracted modal features to form a unified and efficient feature representation. The model structure is as follows: Figure 1 As shown.

[0039] The specific integration process is divided into two stages: feature splicing and adaptive optimization.

[0040] In the first stage, power grid safety production review data from different modalities are directly spliced ​​together in the early stage to form a joint feature containing multimodal information. For the integration scenario of similar data, this invention designs an adaptive feature selection model based on channel attention, which focuses on strengthening the integrated features common to the modalities, as shown in formula (7).

[0041] (7) In formula (7), The weight vectors generated for the channel attention mechanism are used to represent the importance of different features. For fully connected layer operations, For average pooling operation, For feature cascading operations, As a unique integration feature of modality, This model identifies common features shared by all modalities. It adaptively filters and highlights common features that contribute significantly to the integration result, while suppressing redundant information.

[0042] In the second stage, after obtaining the channel weight vector, in order to further optimize the integration effect, it is necessary to reweight the integrated power grid safety production review data features based on the weight vector to achieve dynamic adjustment of different features. At the same time, the residual connection method is used to add the original features and the weighted features to retain the original feature information and alleviate the gradient vanishing problem of deep networks, thereby obtaining the integrated features. The calculation formula is shown in formula (8).

[0043] (8) In formula (8), These are the initial integrated features after optimization.

[0044] Through the above operations, the multimodal characteristics of two different types of power grid safety production review data are organically integrated to form a preliminary multimodal integration feature that is both comprehensive and targeted, laying the foundation for subsequent redundancy processing and final integration.

[0045] 4. Redundancy processing of multi-source heterogeneous data After the initial multimodal feature integration, the resulting integrated data may still contain a large amount of redundant information. This redundant information can interfere with the accuracy of subsequent analysis and increase the computational burden. Therefore, it is necessary to further optimize the integration results through redundancy processing. In this step, the maximum matching method is used to remove redundant information from the original power grid safety production data, as shown in formula (9).

[0046] (9) In formula (9), This is a set of multimodal data features after redundancy processing. To assess the correlation between features in multimodal data, This is the criterion function used to select the optimal subset of features (achieving optimal selection by balancing feature relevance with minimizing redundancy).

[0047] After eliminating redundant features, to further improve the reliability of the integration results, a trust function needs to be introduced to quantify the credibility and correlation between different modal data. By analyzing the characteristic attributes of each modal data, the trust function for heterogeneous data is calculated using formula (4) and the Pearson correlation coefficient. (In Formula 4, Corresponding to the trust function S1 corresponds to the multimodal feature set after redundancy processing. S2 corresponds to (Corresponding target samples). Then, the trust function is normalized to eliminate dimensional differences and ensure that it is comparable under a uniform scale. Finally, the normalized trust function is integrated with the redundant multimodal data features to obtain the final integrated result of the multimodal data features, as shown in formula (10).

[0048] (10) In formula (10), The final result of integrating multimodal data features. This is the trust function.

[0049] Through a series of processes involving redundancy handling and trust function integration, the efficient and accurate integration of multimodal features of heterogeneous data in power grid safety production review is achieved, providing high-quality feature data support for subsequent safety production risk management system audits.

[0050] Heterogeneous data on power grid safety production review with multimodal characteristics were collected experimentally. Before the experiment, the original distribution of data features was recorded. Subsequently, the multimodal feature integration method for heterogeneous data on power grid safety production review proposed in this patent was used to integrate and process these data. The results are as follows: Figure 2 As shown.

[0051] from Figure 2 (a) Before integration, it can be seen that the data on safety production review of different power grids are highly discrete, with low coverage and overlap. This means that there are many gaps in the data, failing to fully cover all aspects of power grid safety production review, which may lead to data redundancy and safety hazards. According to Figure 2 (a) After integration, it can be seen that the data distribution is more concentrated and regular after the integration method designed in this patent, the dispersion is reduced, and it can more comprehensively cover all aspects of power grid safety production and reduce the data gap area. Within the distribution range, the dispersion between nodes is small and the correlation between nodes is closer, indicating that the integration method effectively enhances the data correlation, improves the coverage of all aspects of power grid safety production review, optimizes the data feature distribution, and can more accurately reflect the actual status of power grid safety production review.

[0052] Multimodal feature integration and comparative analysis In the study of the multimodal feature integration method of heterogeneous data in power grid safety production review, feature matching and integration efficiency are the key indicators for evaluating the quality of the method. Feature matching operation is performed on the data in each dataset, and the features contained in the data are compared and analyzed with the preset feature classes. Based on the matching results, the tightness between features in the feature class is calculated. The greater the tightness, the more accurate the matching result, as shown in formula (10).

[0053] (10) In formula (10), Indicates density, and Indicates two features, Indicates the similarity between features.

[0054] To verify the performance of the designed algorithm in feature integration, three other integration methods were selected as comparison methods on the same dataset. Feature integration was performed on the dataset using each method, and the compactness results for each method were obtained. Figure 3 As shown.

[0055] from Figure 3 The results show that, compared with the other three integration methods, the proposed method performs best across various datasets, exhibiting high tightness and low fluctuation. The matching tightness across different datasets is consistently above 0.9, significantly outperforming other methods. This demonstrates that the proposed algorithm can more accurately identify feature associations between data points, making the matching results more consistent with reality. This is because a multimodal feature matching matrix is ​​constructed based on the Pearson correlation coefficient to quantify the correlation between features, and then combined with central cosine similarity to complete mixed attribute feature matching. This dual-verification approach improves the accuracy of feature association judgment and ensures the tightness of the matching results.

[0056] Besides the accuracy of feature matching, data integration efficiency is also an important indicator of the practicality of a method. In this experiment, data from a heterogeneous dataset for power grid safety production review was integrated, and the time required for different integration methods was recorded and plotted as a curve, such as... Figure 4 As shown.

[0057] according to Figure 4 The data integration time curves show that as the amount of heterogeneous data from power grid safety production review increases, the required integration time also gradually increases. However, it is evident that the integration time required by the method designed in this patent is significantly lower than that of the three comparative methods. After all data integration is completed, the integration time of the method designed in this patent is 3.7 seconds, while the integration times required by the three comparative methods are 7.8 seconds, 9.1 seconds, and 8.7 seconds, respectively. This is because the adaptive feature selection model based on channel attention reweights the features of different modalities through weight vectors, highlighting important features and suppressing irrelevant information. Residual connections retain the original feature information, avoiding feature loss during the integration process and improving the integrity of the integrated features. This achieves effective fusion of multi-source information, reduces information gaps between modalities, and makes the integration results more comprehensive.

[0058] As can be seen from the above, the method designed in this application is superior to the comparative methods in terms of feature matching tightness and data integration efficiency. It can more accurately mine data feature correlations and complete the integration of a large amount of data in a shorter time.

[0059] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0060] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0061] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0062] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for integrating multimodal features of heterogeneous data in power grid safety production review, characterized in that, Includes the following steps: Feature extraction: The heterogeneous data of power grid safety production review to be processed are classified into independent datasets, and then feature extraction is performed and the feature values ​​are normalized. Hybrid attribute feature matching: Calculate the correlation coefficient between multimodal features, and enhance the heterogeneous data of power grid safety production review based on the correlation coefficient to form a feature matching matrix. Then, save the correlation information stored in the feature matching matrix into the spatial basis of the corresponding two modes, and determine whether the feature vectors are assigned to the same feature set based on the central cosine similarity. Multimodal feature integration: After directly concatenating the features, a channel attention mechanism is used to obtain the weight vector. Then, a residual connection is used to add the original features and the weighted features to obtain the integrated features.

2. The method for integrating multimodal features of heterogeneous data in power grid safety production review according to claim 1, characterized in that, The process of mixed attribute feature matching includes the following steps: The correlation magnitude of the multimodal features of safety production review data is calculated using the normalized Pearson correlation coefficient, as shown in the following formula: In the formula, To determine the correlation between the features of multimodal safety production review data, and All of these are multimodal safety production review data samples. For data The average value, For data The average value, For data standard deviation For data Standard deviation; Enhancement processing is applied to the power grid safety production review data to form a feature matching matrix. The calculation formula is shown in the following formula: In the formula, To enhance the initial feature matrix before processing, L ( X ) for The feature matching matrix obtained after performing enhancement processing; Subsequently, the relevance information stored in the feature matching matrix is ​​saved to the spatial basis of the corresponding two modalities; Calculate the center cosine similarity of the power grid safety production review data. When the difference between the calculated center cosine similarity result and 1 is less than the set minimum threshold, they are grouped into the same feature set and stored together; when the difference between the result and 1 is greater than the set maximum threshold, they are stored separately.

3. The method for integrating multimodal features of heterogeneous data in power grid safety production review according to claim 2, characterized in that, The formula for calculating central cosine similarity is as follows: In the formula, Represents the central cosine similarity. and They represent different spatial bases. This indicates the number of feature vectors.

4. The method for integrating multimodal features of heterogeneous data in power grid safety production review according to claim 1, characterized in that, The weight vector is obtained as follows: In the formula, The weight vectors generated for the channel attention mechanism are used to represent the importance of different features. For fully connected layer operations, For average pooling operation, For feature cascading operations, As a unique integration feature of modality, These are integrated features shared by all modalities.

5. The method for integrating multimodal features of heterogeneous data in power grid safety production review according to claim 4, characterized in that, The integrated features are obtained in the following ways: In the formula, These are the initial integrated features after optimization.

6. The method for integrating multimodal features of heterogeneous data in power grid safety production review according to claim 1, characterized in that, It also includes the following steps: Redundancy processing of multi-source heterogeneous data: The maximum matching method is used to remove redundant information in the original power grid safety production data, and a trust function is introduced to quantify the credibility and correlation between different modal data. The normalized trust function is integrated with the redundant multimodal data features to obtain the final integrated result of the multimodal data features.

7. The method for integrating multimodal features of heterogeneous data in power grid safety production review according to claim 6, characterized in that, The process of obtaining the final integration result is as follows: The maximum matching method is used to remove redundant information from the original power grid safety production data, as shown in the following formula: In the formula, This is a set of multimodal data features after redundancy processing. To assess the correlation between features in multimodal data, Here is the criterion function used to select the optimal feature subset. The standard deviation of the current feature distribution; The trust function for heterogeneous data is calculated using the Pearson correlation coefficient. As shown below: In the formula, Represents the trust function. This represents the set of multimodal features after redundancy processing. Indicates the corresponding target sample. express The average value, express The average value, express standard deviation express Standard deviation; The trust function is normalized, and then integrated with the redundant multimodal data features to obtain the final integrated result of the multimodal data features, as shown in the following formula: In the formula, The final result of integrating multimodal data features. This is the trust function.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the method for integrating heterogeneous data multimodal features for power grid safety production review as described in any one of claims 1 to 7.

9. A processor, characterized in that, The processor is used to run a program, wherein the program executes the method for integrating multimodal features of heterogeneous data for power grid safety production review as described in any one of claims 1 to 7.

Images