Multi-mode-based power transmission and transformation equipment digital twin data anomaly elimination method

Abstract

The invention discloses a multi-modal-based power transmission and transformation equipment digital twin data anomaly elimination method, and relates to the technical field of data processing, and the method comprises the steps: obtaining multi-modal data of power transmission and transformation equipment, carrying out the feature outlier degree calculation of the multi-modal data according to a preset anomaly feature calculation rule, and obtaining feature outlier degree data; based on the feature outlier degree data, obtaining weight gain data of each feature in the candidate feature set by adopting an iterative feature weight calculation method, and classifying complementary features and redundant features of the features in the candidate feature set according to the weight gain data; and according to the feature outlier degree data and the feature classification result, selectively retaining features in the candidate feature set to obtain a feature subset for the digital twinning of the power transmission and transformation equipment. According to the method, feature correlation and interaction are embedded into abnormal value measurement, and whether complementary or redundant features exist in a selected feature set in a candidate feature set or not is evaluated by using a standard based on feature weight gain.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for eliminating abnormalities in digital twin data of power transmission and transformation equipment based on a multi-modal approach. Background Art

[0002] Digital twin technology has been widely used in the condition monitoring and maintenance of power equipment. By collecting and analyzing the multimodal data of the equipment and combining it with the digital model of the equipment, real-time equipment monitoring and maintenance can be achieved. However, the collected multimodal data such as vision, laser, and sensors usually contain a lot of data outliers and abnormal patterns. Data anomaly removal refers to the discovery and elimination of suspicious and rare data that are very different from most standard features or patterns. Through the data anomaly removal method, the errors and noise generated in the data acquisition process can be effectively filtered out, ensuring that the digital twin system obtains high-quality multimodal input data, and providing more accurate data support for the condition detection and fault diagnosis of power transmission and transformation equipment.

[0003] The performance of data anomaly removal depends largely on the feature set adopted. In high-dimensional datasets, on the one hand, redundant features, noisy features and their complex interactions lead to high computational cost and low removal efficiency. On the other hand, more time overhead does not lead to more accurate results, and the presence of certain features may not contribute much to the removal, or even have a negative impact. Before further determining whether there are outliers, the dataset should be reduced in dimension. Feature selection is essential for data anomaly removal, not only to eliminate irrelevant and redundant features, but also to eliminate noisy features when facing high-dimensional data. Correctly selecting an appropriate feature subset can reduce the running time and improve the accuracy of the removal results. Therefore, feature selection is usually used as a preprocessing method in data anomaly removal, and the ideal subset of original features is selected according to the evaluation criteria of the specific task.

[0004] In the data anomaly removal scenario, classification features are very important or essential information to characterize the dataset. Due to the inevitable noise, existing data anomaly removal methods are usually unable to distinguish between noisy feature values ​​and abnormal feature values. When performing feature-level subspace exploration, some relevant information may be ignored or noise / redundancy may be mixed into the relevant subspace. When there are complex interactions between noisy / redundant features and relevant features, it is difficult to identify and remove useless features.

[0005] In response to the above problems, the present invention proposes an abnormal feature data elimination algorithm that considers interaction, redundancy and complementarity. The algorithm is based on the feature selection method of dense subgraphs and the Relief algorithm drive, and combines them. According to the correlation with abnormal data, it focuses on feature interaction, feature redundancy and feature complementarity. Through algorithm improvement, the present invention will significantly enhance the digital twin system's ability to judge abnormal multimodal data, eliminate suspicious and rare data, enhance the robustness of the entire power equipment perception system for model training, and ensure the stability and safety of power transmission and transformation equipment. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the present invention provides a method for eliminating abnormalities in digital twin data of power transmission and transformation equipment based on a multi-modality, which can solve the problems mentioned in the background technology.

[0008] To solve the above technical problems, the present invention provides the following technical solutions: a method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multimodal model, comprising: obtaining multimodal data of the power transmission and transformation equipment, calculating the feature outlier degree of the multimodal data according to a preset abnormal feature calculation rule, and obtaining feature outlier degree data; based on the feature outlier degree data, using an iterative feature weight calculation method to obtain weight gain data of each feature in a candidate feature set, and classifying the features in the candidate feature set into complementary features and redundant features according to the weight gain data; based on the feature outlier degree data and the feature classification results, selectively retaining the features in the candidate feature set to obtain a feature subset for the digital twin of the power transmission and transformation equipment.

[0009] As a preferred solution of the method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal method described in the present invention, wherein: the preset abnormal feature calculation rule includes calculation of abnormal values ​​within features and calculation of abnormal values ​​between features;

[0010] Intra-feature outliers are specified by deviations in frequency from the mode;

[0011] Inter-feature outliers are specified as the co-occurrence frequency value between two features.

[0012] As a preferred solution of the method for eliminating anomaly in digital twin data of power transmission and transformation equipment based on a multimodal approach described in the present invention, the feature outlier degree data is calculated using a generic objective function, specifically, the outlier is calculated through a linear combination of the outliers within the feature and the outliers between the features.

[0013] As a preferred solution of the method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal approach described in the present invention, the iterative feature weight calculation method includes:

[0014] Dividing the multimodal data into a positive sample set and a negative sample set;

[0015] For each sample, select the similar samples and heterogeneous samples that are closest to the sample;

[0016] When the sample belongs to the positive sample set, the feature weight is subtracted from the difference value with the same type of sample and added with the difference value with the different type of sample;

[0017] When the sample belongs to the negative sample set, the feature weight is added with the difference value with the same type of sample, minus the difference value with the different type of sample.

[0018] As a preferred solution of the method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal method described in the present invention, the weight gain data is calculated according to the following formula:

[0019] FWG(f i )=W F [f i ]-W F\f [f i ];;

[0020] Among them, f i is the feature in the candidate feature set, W F [f i ] is the feature weight calculated using feature set F, W F\f [f i ] is the feature weight calculated using the feature set F\f.

[0021] As a preferred solution of the method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal method described in the present invention, the judgment rule of the feature classification result is:

[0022] When the weight gain data of feature f′ is greater than zero and J(F′ / f′)>J(F′), feature f′ is determined to be a complementary feature;

[0023] When the weight gain data of feature f′ is greater than zero and J(F′ / f′)≤J(F′), feature f′ is determined to be a redundant feature.

[0024] As a preferred solution of the method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal method described in the present invention, the features and complementary features of the data with the highest feature outlier degree are retained in the feature subset;

[0025] Redundant features are removed from the candidate feature set.

[0026] To further solve the above technical problems, the present invention provides the following technical solutions: a data anomaly elimination system for digital twins of multimodal power transmission and transformation equipment, comprising: an acquisition and processing module, used to obtain multimodal data of the power transmission and transformation equipment, and calculate the feature outlier degree of the multimodal data according to a preset abnormal feature calculation rule to obtain feature outlier degree data; a classification module, used to obtain weight gain data of each feature in a candidate feature set based on the feature outlier degree data by an iterative feature weight calculation method, and classify the features in the candidate feature set into complementary features and redundant features according to the weight gain data; a screening module, used to selectively retain the features in the candidate feature set according to the feature outlier degree data and the feature classification results, and obtain a feature subset for the digital twins of the power transmission and transformation equipment.

[0027] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method for eliminating anomalies in digital twin data of power transmission and transformation equipment based on a multi-modal method as described above are implemented.

[0028] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the method for eliminating anomalies in digital twin data of power transmission and transformation equipment based on a multi-modal method as described above are implemented.

[0029] Beneficial effects of the invention: The present invention proposes a new feature selection method for abnormal data in classified data, which takes into account feature interaction, redundancy and complementarity. The method embeds feature correlation and interaction into the outlier metric, and uses a criterion based on feature weight gain to evaluate whether there are complementary or redundant features in the selected feature set in the candidate feature set. In addition, feature complementarity and feature redundancy are redefined to find a feature subset that contains as much important information as possible. The application of this technology can improve the effectiveness of multimodal data training of power equipment, improve the accuracy of deep learning monitoring of the status of power equipment, and better ensure the long-term operation of power transmission and transformation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0031] Figure 1 This is a schematic diagram of the overall process of a method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal approach proposed by the present invention;

[0032] Figure 2A flow chart of an algorithm for eliminating abnormal characteristic data in a method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal approach proposed by the present invention;

[0033] Figure 3 This is a computer equipment diagram in the method for eliminating abnormalities in digital twin data of power transmission and transformation equipment based on a multi-modality proposed by the present invention. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0035] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0036] Example 1, reference Figure 1 and Figure 2 , which is an embodiment of the present invention, provides a method for eliminating abnormalities in digital twin data of power transmission and transformation equipment based on a multi-modal approach.

[0037] S1: Acquire multimodal data of power transmission and transformation equipment, calculate the feature outlier degree of the multimodal data according to preset abnormal feature calculation rules, and obtain feature outlier degree data.

[0038] First, some symbols used in the present invention are summarized in Table 1.

[0039] Table 1 Symbols and definitions

[0040]

[0041] Specifically, the dense subgraph-based feature selection algorithm simultaneously integrates intra-feature outliers and inter-feature outliers to measure the outlier value of each feature. Therefore, the preset abnormal feature calculation rules in the present invention include intra-feature outlier calculation and inter-feature outlier calculation.

[0042] Where: Intra-feature outliers are specified by the deviation of their frequencies from the mode:

[0043]

[0044] Among them, f is the feature, m is the mode, v is the eigenvalue of feature f, dom(f) is the set of possible eigenvalues ​​of the feature, and ∈ is the inverse of the number of samples.

[0045] The inter-feature outlier is specified as the co-occurrence frequency value between feature f and feature f′:

[0046]

[0047] Among them, f and f′ are two features, and v′ are the feature values ​​of features f and f′ respectively.

[0048] The generic objective function J(F') calculates the outliers by linearly combining the intra-feature outliers and inter-feature outliers calculated above:

[0049]

[0050] Among them, F′ is the feature subset, and |F′| is the cardinality of the feature subset. The feature selection algorithm based on dense subgraph aims to find the feature subset with the highest correlation with abnormal data, that is, the maximum value of J(F′).

[0051] S2: Based on the feature outlier data, an iterative feature weight calculation method is used to obtain the weight gain data of each feature in the candidate feature set, and the features in the candidate feature set are classified into complementary features and redundant features according to the weight gain data.

[0052] The present invention uses deep sparse feature selection to calculate the degree of outliers for each feature. Therefore, the same objective function J(·) is constructed for the original feature set F and the selected feature subset F', respectively. The goal is to find an optimal feature subset for abnormal data detection, retaining the complementarity between features as much as possible and reducing redundancy. In general, the higher the degree of outliers of a feature, the stronger its correlation with abnormal data detection, and the more preferentially it is selected. It is assumed that complementary features can improve the accuracy of abnormal data detection when used together, while redundant features are not.

[0053] The Relief algorithm calculates the ability of each feature to distinguish different categories by randomly learning a certain number of instances, thereby obtaining a ranking of feature weights. As features are selected one by one, the weights of the remaining features will change. In other words, the weight of a single feature will be affected by the presence or absence of other features, that is, feature complementarity. By measuring the changes in the Relief algorithm to obtain feature weight gains, possible complementary features can be identified. Complementary features are put into F' together with features that are highly relevant to abnormal data detection, while redundant features are directly removed from F.

[0054] When the feature f with the highest degree of outlier is moved to F', the changes in the weights of the remaining features, i.e., the feature weight gains, are compared. The features with the largest weight reduction are singled out. The process of calculating feature weights by the Relief algorithm is shown in Algorithm 1. To avoid contingency and instability, the Relief algorithm is set to traverse all instances instead of randomly selecting a certain number of instances.

[0055] Algorithm 1: Improved Relief algorithm

[0056]

[0057] Inspired by the Relief algorithm, for a given feature f i (1≤i≤D),

[0058] Examples and examples The difference between is defined in the present invention as:

[0059]

[0060] in, and They are the features f i The eigenvalues ​​in samples s and z, feature f i is a nominal variable, z is z + or - Since the more categories of feature values ​​there are, the greater the possibility of inconsistency between two feature values, the frequency shown in the above formula is used for measurement to ensure that the feature weights are as fair and reasonable as possible.

[0061] In the feature selection method based on the understanding of feature interaction and intrinsic value interaction, the degree of outliers is based on the understanding of complex feature interactions and intrinsic value interactions, and it mines the interactions between features. When a single feature is paired with other features, the information gain relative to the category changes from weak to strong, which is regarded as feature interaction. In the present invention, we distinguish feature interaction from feature complementarity. Feature complementarity refers to the supplementary classification information provided by new features related to the selected features. Next, some new criteria for selecting potential complementary features or redundant features will be implemented. In the present invention, complementarity and redundancy are redefined.

[0062] The calculation of weight gain data is, given a feature set F and a feature f i , feature f i (1≤i≤D) The feature weight gain is defined as:

[0063] FWG(f i )=W F [f i]-W F\f [f i ];

[0064] In the formula, f i ∈F and f i ∈F\f,W F [f i ] and W F\f [f i ] represent the feature f calculated by the Relief algorithm using the feature set F and F\f respectively i In the feature selection process, the weights of the remaining features are variable and will be affected by the absence of the selected features. The present invention regards the feature with the highest positive weight gain as the most likely complementary feature of the selected feature subset and uses it to decide whether to proceed to the next step. In short, FWG(f i ) measures the extent to which a feature complements the previously selected feature set F'.

[0065] Complementarity and redundancy can be distinguished because redundant features do not bring additional information to abnormal data detection. The goal of the proposed method is to find the feature subset with the highest outlier degree (i.e., the highest feature weight gain). Therefore, the maximum complementarity and minimum redundancy objective functions are defined as:

[0066]

[0067] The objective function aims to search for the feature f' with the highest correlation with abnormal data detection from the candidate feature set F\f. Through this objective function, features that are more complementary to each other rather than redundant can be found.

[0068] Feature complementarity is that the feature f with the highest degree of outlier is selected into F' and feature f'∈F. If FWG(f')>0 and J(F' / f')>J(F'), then the complementarity of feature f' and feature f should satisfy the conditions specified by the function MCMR(f').

[0069] Complementarity needs to meet two conditions at the same time: First, it needs to supplement the information of the selected feature subset F'. The feature f' calculated by the above formula is most affected by F'. In the absence of F', the performance of feature f' on abnormal data detection decreases most significantly. Another condition is that the supplementary information should be beneficial to OD, that is, J(F' / f')>J(F').

[0070] Feature redundancy is that the feature f with the highest degree of outlier is selected into F' and feature f'∈F. If FWG(f')>0 and J(F' / f')≤J(F'), then the redundancy of feature f' relative to feature f should also satisfy the conditions specified by the function MCMR(f').

[0071] Even though the redundant feature f' is affected by the selected feature subset F', its combination with F' does not bring additional information to OD, that is, J(F' / f')≤J(F').

[0072] From another perspective, complementarity and redundancy refer to the relationship between two features and the target category. When one of the features is selected, the feature weight gain of the other feature will show a decrease, that is, FWG(f')>0, which indicates either a complementary relationship or a redundant relationship. The essential difference between the two lies in the information they provide. Redundant features bring repeated information, while complementary features provide fresh and effective information.

[0073] In order to distinguish complementarity from redundancy, the present invention compares the impact of adding each feature on the objective function. Specifically, the present invention compares the values ​​of J(F' / f') and J(F'). It is important to note that it is impractical for each feature to have a complementary or redundant feature that meets the conditions. If not, this step is skipped. Instead, this is an iterative search process to select a feature from F and either add it to F' as a complementary feature or remove it from F as a redundant feature.

[0074] In summary, the present invention proposes a new feature selection algorithm (FSIRC), which takes into account the interaction, complementarity and redundancy between features, as shown in Algorithm 2.

[0075] Algorithm 2: FSIRC algorithm

[0076]

[0077]

[0078] The FSIRC algorithm can be divided into two main parts. In the first part, inspired by the deep sparse feature selection algorithm, it measures the outlier degree of each feature based on the training of the algorithm, which in part builds a value map based on the correlation of anomaly data detection to infer deeper feature interactions and value interactions. In the second part, it combines the outlier degree with a new feature selection strategy. In the process of selecting features with higher outlier degree, the weights of the remaining features in the candidate feature set are iteratively calculated by the Relief algorithm. The feature weight gain is used as a criterion to find possible complementary features for the selected feature subset. Therefore, the FSIRC algorithm is able to identify potential feature combinations with feature weight gains that have a significant impact on the selected feature subset. Complementary features will be selected, redundant features will be eliminated, and features that are highly correlated with anomaly data will also be selected. The overall algorithm process is as follows Figure 2 shown.

[0079] S3: Based on the feature outlier data and feature classification results, the features in the candidate feature set are selectively retained to obtain a feature subset for the digital twin of the power transmission and transformation equipment.

[0080] In summary, the present invention proposes a new feature selection method for abnormal data in classified data, which takes into account feature interaction, redundancy and complementarity. This method embeds feature correlation and interaction into the outlier metric, and uses a criterion based on feature weight gain to evaluate whether there are complementary or redundant features in the selected feature set in the candidate feature set. In addition, feature complementarity and feature redundancy are redefined to find feature subsets that contain as much important information as possible. The application of this technology can improve the effectiveness of multimodal data training of power equipment, improve the accuracy of deep learning monitoring of the status of power equipment, and better ensure the long-term operation of power transmission and transformation equipment. These innovations constitute the core technology and competitive advantages that the present invention intends to protect.

[0081] Embodiment 2 is an embodiment of the present invention, which provides a multi-modal digital twin data anomaly elimination system for power transmission and transformation equipment, including:

[0082] The acquisition and processing module is used to obtain multimodal data of the power transmission and transformation equipment, calculate the characteristic outlier degree of the multimodal data according to the preset abnormal characteristic calculation rules, and obtain the characteristic outlier degree data;

[0083] A classification module is used to obtain weight gain data of each feature in the candidate feature set based on feature outlier data by using an iterative feature weight calculation method, and classify the features in the candidate feature set into complementary features and redundant features according to the weight gain data;

[0084] The screening module is used to selectively retain the features in the candidate feature set based on the feature outlier data and feature classification results, so as to obtain a feature subset for the digital twin of the power transmission and transformation equipment.

[0085] Example 3, reference Figure 3, is an embodiment of the present invention, which is different from the previous embodiment in that: if the function is implemented in the form of a software function 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, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0086] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0087] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0088] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0089] Example 4 is an embodiment of the present invention, which provides a method for eliminating anomaly in digital twin data of power transmission and transformation equipment based on a multi-modal approach. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0090] In order to verify the effect of this algorithm, the abnormal feature data removal algorithm is applied to 4 widely used multimodal classification datasets covering different fields. The datasets are selected according to the following criteria: the dataset must contain outliers or have a highly unbalanced class distribution in order to reduce the sampling rate when converting classified data into abnormal data. In these datasets, Probe and U2R use Probe and user-to-root attacks as outliers for common classes, respectively, derived from KDDCUP99; Penglung1 is derived from PenglungEW and uses different classes as outliers. The basic statistics of these datasets and the results of common abnormal feature removal algorithms are shown in Table 2.

[0091] Table 2 Dataset summary

[0092]

[0093]

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal model, characterized in that: include: Acquire multimodal data of the power transmission and transformation equipment, and calculate the characteristic outlier degree of the multimodal data according to a preset abnormal characteristic calculation rule to obtain characteristic outlier degree data; Based on the feature outlier data, an iterative feature weight calculation method is used to obtain weight gain data of each feature in the candidate feature set, and the features in the candidate feature set are classified into complementary features and redundant features according to the weight gain data; According to the feature outlier data and feature classification results, the features in the candidate feature set are selectively retained to obtain a feature subset for the digital twin of the power transmission and transformation equipment.

2. The method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal approach as claimed in claim 1 is characterized in that: The preset abnormal feature calculation rules include intra-feature abnormal value calculation and inter-feature abnormal value calculation; Intra-feature outliers are specified by deviations in frequency from the mode; Inter-feature outliers are specified as the co-occurrence frequency value between two features.

3. The method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal approach as claimed in claim 2 is characterized in that: The feature outlier degree data is calculated using a generic objective function, specifically, the outlier is calculated through a linear combination of the intra-feature outlier and the inter-feature outlier.

4. The method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal approach as claimed in claim 3 is characterized by: The iterative feature weight calculation method comprises: Dividing the multimodal data into a positive sample set and a negative sample set; For each sample, select the similar samples and heterogeneous samples that are closest to the sample; When the sample belongs to the positive sample set, the feature weight is subtracted from the difference value with the same type of sample and added with the difference value with the different type of sample; When the sample belongs to the negative sample set, the feature weight is added with the difference value with the same type of sample, minus the difference value with the different type of sample.

5. The method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal approach as claimed in claim 4 is characterized in that: The weight gain data is calculated according to the following formula: FWG(f i )=W F [f i ]-W F\f [f i ];; Among them, f i is the feature in the candidate feature set, W F [f i ] is the feature weight calculated using feature set F, W F\f [f i ] is the feature weight calculated using the feature set F\f.

6. The method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal approach as claimed in claim 5, characterized in that: The judgment rule of the feature classification result is: When the weight gain data of feature f′ is greater than zero and J(F′ / f′)>J(F′), feature f′ is determined to be a complementary feature; When the weight gain data of feature f′ is greater than zero and J(F′ / f′)≤J(F′), feature f′ is determined to be a redundant feature.

7. The method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modality according to claim 6 is characterized in that: retaining the features and complementary features having the highest feature outlier data in the feature subset; Redundant features are removed from the candidate feature set.

8. A system for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal system, based on a method for eliminating abnormal data of digital twins of power transmission and transformation equipment based on a multi-modal system according to any one of claims 1 to 7, characterized in that: include, An acquisition and processing module is used to obtain multimodal data of the power transmission and transformation equipment, and calculate the characteristic outlier degree of the multimodal data according to a preset abnormal characteristic calculation rule to obtain characteristic outlier degree data; A classification module, used to obtain weight gain data of each feature in the candidate feature set by an iterative feature weight calculation method based on the feature outlier data, and classify the features in the candidate feature set into complementary features and redundant features according to the weight gain data; A screening module is used to selectively retain the features in the candidate feature set according to the feature outlier data and the feature classification results, so as to obtain a feature subset for the digital twin of the power transmission and transformation equipment.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, it implements the steps of the method for eliminating abnormalities in digital twin data of power transmission and transformation equipment based on a multi-modal method described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for eliminating abnormalities in digital twin data of power transmission and transformation equipment based on a multi-modality according to any one of claims 1 to 7 are implemented.

Images