Insulation defect diagnosis method and device for gas insulated switchgear

Abstract

The invention relates to an insulation defect diagnosis method and device for a gas insulated switchgear, and relates to the technical field of electric power. The method comprises the following steps: acquiring signal time sequence data of at least two physical signals of the gas insulated switchgear in an operation state; performing fragmentation processing and coding processing on each signal time sequence data to obtain a feature sequence corresponding to each signal time sequence data; the feature sequence comprises feature vectors of a plurality of signal time sequence segments of the corresponding signal time sequence data; for each feature sequence, performing enhancement processing on the feature sequence according to feature sequences except the feature sequence to obtain an enhanced feature sequence corresponding to the feature sequence; and fusing the enhanced feature sequences to obtain a target feature sequence of the gas insulated switchgear, and determining the type of the insulation defect existing in the gas insulated switchgear based on the target feature sequence. By adopting the method, the accuracy of insulation defect diagnosis of the gas insulated switchgear can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power, and in particular to an insulation defect diagnosis method and device for a gas insulated substation, a computer device, a computer readable storage medium, and a computer program product. BACKGROUND

[0002] Due to various factors in manufacturing, installation or operation, a gas insulated substation (GIS) can have insulation defects such as floating potential bodies, metal particles, insulator contamination or air gaps inside. These insulation defects can cause partial discharge under the action of an electric field, thereby threatening the safety and stability of the equipment and even the entire power grid where the gas insulated substation is located.

[0003] In related technologies, the insulation defect diagnosis of a gas insulated substation is usually based on information of a single modality. However, the diagnosis method of a single modality is one-sided in information and is easily disturbed by a specific type of noise, thereby resulting in low accuracy of the insulation defect diagnosis of the gas insulated substation. SUMMARY

[0004] Therefore, it is necessary to provide an insulation defect diagnosis method, device, computer device, computer readable storage medium and computer program product for a gas insulated substation, which can improve the accuracy of the insulation defect diagnosis of the gas insulated substation, to solve the technical problem of low accuracy of the insulation defect diagnosis of the gas insulated substation.

[0005] In a first aspect, the present application provides an insulation defect diagnosis method for a gas insulated substation, comprising:

[0006] obtaining signal time sequence data of at least two physical signals of the gas insulated substation in a running state;

[0007] performing slicing processing and encoding processing on each of the signal time sequence data to obtain a feature sequence corresponding to each of the signal time sequence data; the feature sequence comprises feature vectors of a plurality of signal time sequence segments of the corresponding signal time sequence data;

[0008] for each of the feature sequences, performing enhancement processing on the feature sequence according to the feature sequences other than the feature sequence to obtain an enhanced feature sequence corresponding to the feature sequence;

[0009] fusing each of the enhanced feature sequences to obtain a target feature sequence of the gas insulated substation, and determining a type of an insulation defect existing in the gas insulated substation based on the target feature sequence.

[0010] In one embodiment, the step of enhancing the feature sequence for each feature sequence based on feature sequences other than the original feature sequence to obtain an enhanced feature sequence corresponding to the original feature sequence includes:

[0011] For each of the first feature sequences, the corresponding second feature sequences are concatenated to obtain a concatenated feature sequence; the first feature sequence is any one of the feature sequences, and the second feature sequence is any one of the feature sequences other than the first feature sequence.

[0012] A query sequence is determined based on the first feature sequence, and a key sequence and a value sequence are determined based on the concatenated feature sequence; the query sequence includes multiple query vectors, the key sequence includes multiple key vectors, and the value sequence includes multiple value vectors.

[0013] Based on each query vector and each key vector, determine the fusion weight of each value vector under each query vector;

[0014] Based on the fusion weights of each value vector under each query vector, each query vector and each value vector are fused under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0015] In one embodiment, determining the query sequence based on the first feature sequence and determining the key sequence and value sequence based on the concatenated feature sequence includes:

[0016] Based on the query mapping weight information, each feature vector in the first feature sequence is mapped to a corresponding query vector to obtain the query sequence;

[0017] Based on the key mapping weight information, each feature vector in the concatenated feature sequence is mapped to a corresponding key vector to obtain the key sequence;

[0018] Based on the value mapping weight information, each feature vector in the concatenated feature sequence is mapped to a corresponding value vector to obtain the value sequence.

[0019] In one embodiment, the plurality of query vectors are query vectors corresponding to each feature vector in the first feature sequence; the plurality of key vectors are key vectors corresponding to each feature vector in the concatenated feature sequence; and the plurality of value vectors are value vectors corresponding to each feature vector in the concatenated feature sequence.

[0020] The step of determining the fusion weight of each value vector under each query vector based on each query vector and each key vector includes:

[0021] For each query vector, determine the matching degree between the query vector and each key vector;

[0022] Based on the matching degree between the query vector and each of the key vectors, the fusion weight of the value vector corresponding to each key vector under the query vector is determined; the value vector corresponding to each key vector is the value vector with the same feature vector as the key vector.

[0023] In one embodiment, the step of fusing each query vector and each value vector under each query vector according to the fusion weight of each value vector under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence includes:

[0024] For each query vector, according to the fusion weight of each value vector under the query vector, the value vectors are fused to obtain the fused value vector corresponding to the query vector. The query vector and the fused value vector are fused to obtain the enhanced feature vector corresponding to the query vector.

[0025] By combining the enhanced feature vectors corresponding to each query vector, an enhanced feature sequence corresponding to the first feature sequence is obtained.

[0026] In one embodiment, the feature vector in the first feature sequence is a first feature vector, and the plurality of query vectors are query vectors corresponding to each of the first feature vectors in the first feature sequence; the feature vector in the second feature sequence is a second feature vector, the plurality of key vectors are key vectors corresponding to each of the second feature vectors in the concatenated feature sequence, and the plurality of value vectors are value vectors corresponding to each of the second feature vectors in the concatenated feature sequence;

[0027] The method further includes:

[0028] For each query vector, the offset interval of the query vector is determined based on the query position identifier and offset window carried by the query vector; the query position identifier is used to characterize the position of the first feature vector corresponding to the query vector in the first feature sequence; the offset window is determined based on the gas-insulated switchgear.

[0029] Based on the offset interval, multiple target key vectors corresponding to the query vector are determined from the key sequence, and multiple target value vectors corresponding to the query vector are determined from the value sequence; the target key vector is a key vector carrying a key position identifier located within the offset interval, the key position identifier being used to characterize the position of the second feature vector corresponding to the key vector in its respective second feature sequence; the target value vector is a value vector carrying a value position identifier located within the offset interval, the value position identifier being used to characterize the position of the second feature vector corresponding to the value vector in its respective second feature sequence;

[0030] Based on the query vector and each of the target key vectors, determine the fusion weight of each target value vector under the query vector;

[0031] Based on the fusion weights of each target key vector corresponding to each query vector under each query vector, the query vector and the corresponding target key vectors are fused under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0032] In one embodiment, the step of performing segmentation and encoding processing on each of the signal time-series data to obtain the feature sequence corresponding to each of the signal time-series data includes:

[0033] For each of the signal timing data, the signal timing data is segmented into multiple signal timing segments.

[0034] Each of the signal time segments is encoded to obtain a feature vector corresponding to each of the signal time segments;

[0035] According to the order of the signal time sequence segments corresponding to each feature vector in the signal time sequence data, the feature vectors are combined to obtain the feature sequence corresponding to the signal time sequence data.

[0036] Secondly, this application also provides a defect diagnosis device for gas-insulated switchgear, comprising:

[0037] The signal acquisition module is used to acquire the signal timing data of at least two physical signals of the gas-insulated switchgear during operation.

[0038] The signal processing module is used to perform segmentation and encoding processing on each of the signal time-series data to obtain a feature sequence corresponding to each of the signal time-series data; the feature sequence includes feature vectors of multiple signal time-series segments of the corresponding signal time-series data.

[0039] The feature enhancement module is used to enhance each feature sequence based on other feature sequences besides the original feature sequence, to obtain an enhanced feature sequence corresponding to the original feature sequence.

[0040] The defect diagnosis module is used to fuse the enhanced feature sequences to obtain the target feature sequence of the gas-insulated switchgear, and to determine the type of insulation defect in the gas-insulated switchgear based on the target feature sequence.

[0041] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0042] Acquire the signal timing data of at least two physical signals of a gas-insulated switchgear under operating conditions;

[0043] The time-series data of each signal are segmented and encoded to obtain a feature sequence corresponding to each time-series data of the signal; the feature sequence includes feature vectors of multiple time-series segments of the corresponding time-series data.

[0044] For each of the feature sequences, the feature sequences are enhanced based on the feature sequences other than the feature sequence itself, to obtain the enhanced feature sequence corresponding to the feature sequence.

[0045] By fusing the enhanced feature sequences, a target feature sequence for the gas-insulated switchgear is obtained. Based on the target feature sequence, the type of insulation defect present in the gas-insulated switchgear is determined.

[0046] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0047] Acquire the signal timing data of at least two physical signals of a gas-insulated switchgear under operating conditions;

[0048] The time-series data of each signal are segmented and encoded to obtain a feature sequence corresponding to each time-series data of the signal; the feature sequence includes feature vectors of multiple time-series segments of the corresponding time-series data.

[0049] For each of the feature sequences, the feature sequences are enhanced based on the feature sequences other than the feature sequence itself, to obtain the enhanced feature sequence corresponding to the feature sequence.

[0050] By fusing the enhanced feature sequences, a target feature sequence for the gas-insulated switchgear is obtained. Based on the target feature sequence, the type of insulation defect present in the gas-insulated switchgear is determined.

[0051] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0052] Acquire the signal timing data of at least two physical signals of a gas-insulated switchgear under operating conditions;

[0053] The time-series data of each signal are segmented and encoded to obtain a feature sequence corresponding to each time-series data of the signal; the feature sequence includes feature vectors of multiple time-series segments of the corresponding time-series data.

[0054] For each of the feature sequences, the feature sequences are enhanced based on the feature sequences other than the feature sequence itself, to obtain the enhanced feature sequence corresponding to the feature sequence.

[0055] By fusing the enhanced feature sequences, a target feature sequence for the gas-insulated switchgear is obtained. Based on the target feature sequence, the type of insulation defect present in the gas-insulated switchgear is determined.

[0056] The aforementioned method, apparatus, computer equipment, computer-readable storage medium, and computer program product for diagnosing insulation defects in gas-insulated switchgear can enhance each feature sequence by using the remaining feature sequences corresponding to each feature sequence, thereby obtaining an enhanced feature sequence corresponding to each feature sequence. Based on the fusion processing of each enhanced feature sequence, a target feature sequence containing the interaction between various physical signals can be obtained. Based on the target feature sequence containing the interaction between various physical signals, the type of insulation defect existing in the gas-insulated switchgear can be diagnosed. Based on the above process, the enhanced feature sequence of the gas-insulated switchgear insulation defect diagnosis method can characterize the result of mutual absorption between features of different modes. Therefore, based on the target feature sequence, the lifetime prediction can be made by comprehensively considering the result of mutual absorption between features of different modes, thereby improving the accuracy of insulation defect diagnosis of gas-insulated switchgear. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is an application environment diagram of the insulation defect diagnosis method for gas-insulated switchgear in one embodiment;

[0059] Figure 2 This is a flowchart illustrating a method for diagnosing insulation defects in a gas-insulated switchgear in one embodiment.

[0060] Figure 3 This is a flowchart illustrating the steps of enhancing a feature sequence based on a feature sequence other than the feature sequence in one embodiment to obtain an enhanced feature sequence corresponding to the feature sequence.

[0061] Figure 4 This is a flowchart illustrating a multimodal insulation defect diagnosis method in one embodiment;

[0062] Figure 5 This is a structural block diagram of an insulation defect diagnosis device for a gas-insulated switchgear in one embodiment;

[0063] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0066] The insulation defect diagnosis method for gas-insulated combined electrical appliances provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, server 102 communicates with terminal 104 via a network. A data storage system can store the data that server 102 needs to process. This data storage system can be integrated onto server 102 or located in the cloud or on other network servers. Server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Terminal 104 can be, but is not limited to, various sensors, personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices.

[0067] For example, firstly, server 102 acquires signal timing data of at least two physical signals of the gas-insulated switchgear during operation through a terminal 104 deployed inside or near the gas-insulated switchgear, such as a physical signal sensor; then, server 102 performs segmentation and encoding processing on each signal timing data to obtain corresponding feature sequences; each feature sequence corresponds to a signal timing data, including feature vectors of multiple signal timing segments of the signal timing data; next, server 102 enhances each feature sequence based on feature sequences other than the feature sequence to obtain an enhanced feature sequence corresponding to the feature sequence; then, server 102 fuses the enhanced feature sequences to obtain the target feature sequence of the gas-insulated switchgear, determines the type of insulation defect existing in the gas-insulated switchgear based on the target feature sequence, and sends the diagnosed insulation defect type to the terminal 104 corresponding to the maintenance personnel, such as a personal computer, laptop, smartphone, tablet, IoT device, and portable wearable device.

[0068] In some embodiments, such as Figure 2 As shown, a method for diagnosing insulation defects in bulk-insulated switchgear is provided, which can be applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps:

[0069] Step S202: Obtain the signal timing data of at least two physical signals of the gas-insulated switchgear in operation.

[0070] Each signal timing data corresponds to a physical signal; for example, the physical signal includes at least two of the following: ultrasonic signal, ultra-high frequency electromagnetic wave signal, and high frequency pulse current signal. In practical applications, the physical signal includes ultrasonic signal, ultra-high frequency electromagnetic wave signal, and high frequency pulse current signal. It is easy to understand that the physical signal may also include other types of physical signals, such as optical signals, which are not limited in this application.

[0071] Among them, ultrasonic signals refer to mechanical wave signals with frequencies higher than the upper limit of human hearing (usually ≥20kHz); ultra-high frequency electromagnetic wave signals refer to electromagnetic wave signals with frequencies between 300MHz and 3GHz; and high frequency pulse current signals refer to pulse current signals with frequencies in the kHz to MHz range.

[0072] The signal timing data of the physical signal includes the signal value of the physical signal at different times; the signal values ​​in the signal timing data are arranged in chronological order according to their corresponding times.

[0073] In this step, physical signal sensors are deployed inside or near the gas-insulated switchgear to acquire physical signals of the gas-insulated switchgear during operation; the server acquires the signal timing data of various physical signals of the gas-insulated switchgear during operation based on the signal acquisition of the physical signal sensors.

[0074] In practical applications, physical signal sensors report the raw signal time-series data they collect to a server. The server then performs preprocessing on the raw signal time-series data, such as denoising and normalization, to obtain the corresponding signal time-series data. For example, wavelet denoising is used, and Z-Score normalization is preferred for normalization.

[0075] Step S204: Perform segmentation and encoding processing on the time-series data of each signal to obtain the feature sequence corresponding to the time-series data of each signal.

[0076] The signal time series data includes feature vectors of multiple signal time series segments corresponding to the signal time series data.

[0077] In this step, for each signal timing data, the server divides the signal timing data into a preset number of signal timing segments of equal length. Then, the server encodes each signal timing segment to obtain a feature vector corresponding to each signal timing segment. Next, the server arranges the signal timing segments corresponding to each feature vector in the order of arrangement in the signal timing data to obtain the feature sequence corresponding to the signal timing data.

[0078] In practical applications, for each signal time sequence segment, the server first performs preliminary encoding on the signal time sequence segment through a linear projection layer to obtain the corresponding initial feature vector. Then, the server adds a position identifier and a modal identifier to the beginning or end of the initial feature vector to obtain the corresponding feature vector with added identifiers. The position identifier is used to represent the position of the corresponding signal time sequence segment in the signal time sequence data, such as which segment it is, and the modal identifier is used to represent the physical signal to which the corresponding signal time sequence segment belongs. Next, the server encodes the feature vector with added identifiers through the Transformer encoder corresponding to the physical signal to which the signal time sequence segment belongs, to obtain the corresponding feature vector. The Transformer encoder does not encode the position identifier and modal identifier in the feature vector with added identifiers.

[0079] Step S206: For each feature sequence, enhance the feature sequence based on the feature sequences other than the feature sequence itself to obtain the enhanced feature sequence corresponding to the feature sequence.

[0080] In this step, for each first feature sequence, the server searches for information related to the first feature sequence from each corresponding second feature sequence, and enhances the first feature sequence based on the information related to the first feature sequence to obtain the enhanced feature sequence corresponding to the first feature sequence. Here, the first feature sequence is any one of the feature sequences, and the second feature sequence is any one of the feature sequences other than the first feature sequence.

[0081] Step S208: Fuse the enhanced feature sequences to obtain the target feature sequence of the gas-insulated switchgear. Based on the target feature sequence, determine the type of insulation defect in the gas-insulated switchgear.

[0082] In this step, the server concatenates the enhanced feature sequences to obtain the target feature sequence of the gas-insulated switchgear, and inputs the target feature sequence into the pre-trained insulation defect diagnosis model. Based on the target feature sequence, the insulation defect diagnosis model determines the probability of different insulation defect types, and identifies the insulation defect type with the highest probability as the type of insulation defect present in the gas-insulated switchgear.

[0083] For example, the insulation defect diagnosis model is trained based on the sample signal time series data of the sample gas-insulated switchgear.

[0084] For example, the insulation defect diagnosis model is a multilayer perceptron (MLP) classifier, which determines the probability of different insulation defect types based on the softmax function.

[0085] In the above-mentioned method for diagnosing insulation defects in gas-insulated switchgear, each feature sequence can be enhanced by the remaining feature sequences corresponding to each feature sequence, thereby obtaining an enhanced feature sequence corresponding to each feature sequence. Based on the fusion processing of each enhanced feature sequence, a target feature sequence containing the interaction between various physical signals can be obtained. Based on the target feature sequence containing the interaction between various physical signals, the type of insulation defect in the gas-insulated switchgear can be diagnosed. In the insulation defect diagnosis method for gas-insulated switchgear based on the above process, the enhanced feature sequence can characterize the result of mutual absorption between features of different modes. Therefore, based on the target feature sequence, the result of mutual absorption between features of different modes can be comprehensively considered for lifetime prediction, thereby improving the accuracy of insulation defect diagnosis of gas-insulated switchgear.

[0086] Furthermore, by fusing multimodal features during the feature extraction stage, this embodiment can learn the deep, nonlinear interaction relationships between different physical signals during degradation, avoiding the loss of rich low-level features and temporal dynamic information of physical signals during processing, and improving the accuracy and generalization ability of insulation defect diagnosis of gas-insulated switchgear.

[0087] In some embodiments, such as Figure 3 As shown, step S206 above, for each feature sequence, enhances the feature sequence based on feature sequences other than the feature sequence itself, to obtain the enhanced feature sequence corresponding to the feature sequence, including the following steps:

[0088] Step S302: For each first feature sequence, concatenate the corresponding second feature sequences to obtain a concatenated feature sequence; the first feature sequence is any one of the feature sequences, and the second feature sequence is any one of the feature sequences other than the first feature sequence.

[0089] Step S304: Determine the query sequence based on the first feature sequence, and determine the key sequence and value sequence based on the concatenated feature sequence.

[0090] Step S306: Determine the fusion weight of each value vector under each query vector based on each query vector and each key vector.

[0091] Step S308: Based on the fusion weights of each value vector under each query vector, fuse each query vector and each value vector under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0092] The query sequence includes multiple query vectors, the key sequence includes multiple key vectors, and the value sequence includes multiple value vectors.

[0093] In this embodiment, for each first feature sequence, firstly, the server concatenates all second feature sequences other than the first feature sequence to obtain a concatenated feature sequence; then, the server determines the query vector corresponding to each feature vector in the first feature sequence, and arranges the corresponding query vectors according to the order of the feature vectors in the first feature sequence to obtain a query sequence; it also determines the key vector and value vector corresponding to each feature vector in the concatenated feature sequence, and arranges the corresponding key vectors according to the order of the feature vectors in the concatenated feature sequence to obtain a key sequence; and it arranges the corresponding value vectors according to the order of the feature vectors in the concatenated feature sequence to obtain a value sequence; then, the server determines the fusion weight of each value vector under each query vector based on the matching degree between each query vector and each key vector; next, the server performs fusion processing on the query vector and each value vector under each query vector based on the fusion weight of each value vector under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0094] In practical applications, the server adjusts the proportion of each value vector in the enhanced feature sequence by fusing weights, thereby achieving the process of softly filtering out information related to the first feature sequence from the spliced ​​feature sequence (i.e., each second feature sequence).

[0095] Taking the enhancement processing of ultrasonic feature sequences as an example, which includes physical signals such as ultrasonic signals, ultra-high frequency electromagnetic wave signals, and high frequency pulse current signals, the server combines the ultra-high frequency electromagnetic wave feature sequences and the high frequency pulse current feature sequences to obtain the ultra-high frequency electromagnetic wave-high frequency pulse current feature sequences. Then, the server uses the ultrasonic feature sequences as the query to determine the query sequence, and uses the ultra-high frequency electromagnetic wave-high frequency pulse current feature sequences as the key and value to determine the key sequence and value sequence. Through the query-key-value mechanism, the server softly filters out information related to the ultrasonic feature sequences from the ultra-high frequency electromagnetic wave-high frequency pulse current feature sequences.

[0096] In this embodiment, the enhancement processing based on the remaining feature sequences corresponding to each feature sequence enables the features of each modality to fully absorb the features of the remaining modalities.

[0097] In some embodiments, step S304 above, determining the query sequence based on the first feature sequence and determining the key sequence and value sequence based on the concatenated feature sequence, includes the following steps: mapping each feature vector in the first feature sequence to a corresponding query vector based on query mapping weight information to obtain the query sequence; mapping each feature vector in the concatenated feature sequence to a corresponding key vector based on key mapping weight information to obtain the key sequence; and mapping each feature vector in the concatenated feature sequence to a corresponding value vector based on value mapping weight information to obtain the value sequence.

[0098] Among them, the query mapping weight information, key mapping weight information, and value mapping weight information are all weight matrices that include multiple weights; the number of weights in the weight matrix is ​​the same as the number of feature vectors in the feature sequence.

[0099] In this embodiment, the server multiplies the query mapping weight information and the first feature sequence to map each feature vector in the first feature sequence to the corresponding query vector, thereby obtaining a query sequence; the server multiplies the key mapping weight information and the concatenated feature sequence to map each feature vector in the concatenated feature sequence to the corresponding key vector, thereby obtaining a key sequence; the server multiplies the value mapping weight information and the concatenated feature sequence to map each feature vector in the concatenated feature sequence to the corresponding value vector, thereby obtaining a value sequence.

[0100] For example, the process of determining the query sequence, key sequence, and value sequence is shown in Formula 1:

[0101] (Formula 1)

[0102] Wherein, S1 is the first feature sequence, S2 is the second feature sequence, and the number of S2 sequences is at least one; W Q To query mapping weight information, W K W maps weight information to keys. V Values ​​are mapped to weight information; Q is the query sequence, K is the key sequence, and V is the value sequence; Contat(·) is the concatenation operation.

[0103] In this embodiment, the server can determine the corresponding query sequence, key sequence, and value sequence for each modality, thereby facilitating subsequent enhancement processing of the features of each modality based on the Query-Key-Value mechanism.

[0104] In some embodiments, the multiple query vectors are query vectors corresponding to each feature vector in the first feature sequence; the multiple key vectors are key vectors corresponding to each feature vector in the concatenated feature sequence; and the multiple value vectors are value vectors corresponding to each feature vector in the concatenated feature sequence.

[0105] Step S306 above, which determines the fusion weight of each value vector under each query vector based on each query vector and each key vector, includes the following steps: for each query vector, determining the matching degree between the query vector and each key vector; and based on the matching degree between the query vector and each key vector, determining the fusion weight of the value vector corresponding to each key vector under the query vector.

[0106] Here, the value vector corresponding to each key vector is the value vector of the same feature vector as the key vector. In some embodiments, the order of each key vector in the key sequence is the same as the order of the feature vectors corresponding to each key vector in the concatenated feature sequence, and the order of each value vector in the value sequence is the same as the order of the feature vectors corresponding to each value vector in the concatenated feature sequence. Therefore, there is a correspondence between the key vector and the value vector corresponding to the m-th feature vector in the concatenated feature sequence, or in other words, there is a correspondence between the m-th key vector in the key sequence and the m-th value vector in the value sequence.

[0107] In this embodiment, for each query vector, the server determines the matching degree between the query vector and each key vector; then, based on the matching degree between the query vector and each key vector, the server determines the fusion weight of the value vector corresponding to each key vector under the query vector. For example, based on the matching degree between the query vector and the m-th key vector, the server determines the fusion weight of the m-th value vector under the query vector.

[0108] In this embodiment, the server can determine the matching degree between each query vector and each key vector through the Query-Mey-Value mechanism, and determine the fusion weight of the value vector corresponding to each key vector under each query vector based on the matching degree between each query vector and each key vector.

[0109] In some embodiments, step S308 above, which involves fusing each value vector under each query vector according to the fusion weight of each value vector under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence, includes the following steps: for each query vector, fusing each value vector according to the fusion weight of each value vector under the query vector to obtain the fused value vector corresponding to the query vector; fusing the query vector and the fused value vector to obtain the enhanced feature vector corresponding to the query vector; and combining the enhanced feature vectors corresponding to each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0110] In this embodiment, for each query vector, the server employs a cross-modal attention mechanism, using the fusion weights of each value vector under the query vector as adjustment coefficients for each value vector. By multiplying each value vector with its corresponding adjustment coefficient, the server achieves fusion processing of each value vector, thereby obtaining the fused value vector corresponding to the query vector. Next, the server fuses the query vector and its corresponding fused value vector through residual connection to obtain the enhanced feature vector corresponding to the query vector. Finally, the server arranges the enhanced feature vectors corresponding to each query vector according to their order of arrangement in the query sequence, thereby obtaining the enhanced feature sequence corresponding to the first feature sequence.

[0111] For example, the process by which the server enhances the first feature sequence is shown in Formula 2:

[0112] (Formula 2)

[0113] Where d is the dimension of the feature vector in the feature sequence; This represents the process of obtaining the fusion weights. This indicates the fusion process; S1' is the enhanced feature sequence corresponding to the first feature sequence; LayerNorm represents the normalization layer.

[0114] In this embodiment, the server can softly filter out information related to the first feature sequence through the Query-Key-Value mechanism, and enhance the first feature sequence through the cross-modal attention mechanism and the information related to the first feature sequence that has been softly filtered out, thereby enabling the first feature sequence to absorb features from other modalities.

[0115] In some embodiments, the feature vector in the first feature sequence is a first feature vector, and the multiple query vectors are query vectors corresponding to each first feature vector in the first feature sequence; the feature vector in the second feature sequence is a second feature vector, the multiple key vectors are key vectors corresponding to each second feature vector in the concatenated feature sequence, and the multiple value vectors are value vectors corresponding to each second feature vector in the concatenated feature sequence.

[0116] Since the physical signals in this application are multi-modal physical signals, and there may be time delays between different modal physical signals (e.g., electromagnetic wave signals arrive first, and ultrasonic signals arrive later), in order to reduce the query volume and improve query accuracy in the Query-Key-Value mechanism, steps S306 to S308 above determine the fusion weight of each value vector under each query vector based on each query vector and each key vector, and perform fusion processing on each value vector under each query vector based on the fusion weight of each value vector under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence, including the following steps. Step 1: For each query vector, determine the offset interval of the query vector based on the query position identifier and offset window carried by the query vector; based on the offset interval, determine multiple target key vectors corresponding to the query vector from the key sequence, and determine multiple target value vectors corresponding to the query vector from the value sequence; based on the query vector and each target key vector, determine the fusion weight of each target value vector under the query vector; based on the fusion weight of each target key vector corresponding to each query vector under each query vector, perform fusion processing on each query vector and its corresponding target key vectors under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0117] The query position identifier is used to characterize the position of the first feature vector corresponding to the query vector in the first feature sequence. For example, the query position identifier carried by the query vector corresponding to the nth feature vector in the first feature sequence is n.

[0118] The offset window is determined based on the gas-insulated switchgear; further, the offset window is calculated based on the size of the gas-insulated switchgear and the speed of sound propagation; for example, the offset window is [△ Min , △ Max ].

[0119] In some embodiments, the offset interval is centered on the query location identifier, and the length of the offset interval is the same as the length of the offset window. For example, assuming the query location identifier is n, the offset window is [△ Min , △ Max If ], then the offset interval is [n-△ Min ,n+△ Max In other embodiments, the offset interval starts from the query location identifier, and the length of the offset interval is the same as the length of the offset window. For example, assuming the query location identifier is n, the offset window is [△]. Min , △ Max If ], then the offset interval is [n+△ Min ,n+△ Max In some other embodiments, the offset interval ends with the query position identifier, and the length of the offset interval is the same as the length of the offset window. For example, assuming the query position identifier is n, the offset window is [△]. Min , △ Max If ], then the offset interval is [n-△ Max , n-△ Min ].

[0120] The target key vector is a key vector carrying a key position identifier located within the offset interval. The key position identifier is used to characterize the position of the second feature vector corresponding to the key vector in its respective second feature sequence.

[0121] The target value vector is a value vector carrying a value position identifier located within the offset interval. The value position identifier is used to characterize the position of the second feature vector corresponding to the value vector in the second feature sequence to which it belongs.

[0122] In this embodiment, for each query vector, firstly, the server determines the offset interval of the query vector based on the query position identifier and offset window carried by the query vector; then, the server determines multiple target key vectors whose corresponding key position identifiers are located within the offset interval from the key sequence, and multiple target value vectors whose corresponding value position identifiers are located within the offset interval from the value sequence; next, the server determines the fusion weight of each target value vector under the query vector based on the query vector and each target key vector; finally, the server performs fusion processing on the query vector and the corresponding target key vectors under each query vector based on the fusion weight of each target key vector under each query vector, to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0123] Taking the enhancement processing of ultrasonic feature sequences as an example, which includes physical signals such as ultrasonic signals, ultra-high frequency electromagnetic wave signals, and high frequency pulse current signals, we assume that the query vector that needs to be enhanced is the query vector corresponding to the nth ultrasonic feature vector in the ultrasonic feature sequence, and the offset interval is [n+Δ]. Min ,n+△ Max Then the server retrieves the (n+Δ)th element from the UHF electromagnetic wave characteristic sequence from the key sequence. Min up to the n+△ Max The bond vector corresponding to the characteristic vector of the ultra-high frequency electromagnetic wave, and the (n+Δ)th characteristic vector in the high frequency pulse current sequence. Min up to the n+△ Max The key vector corresponding to the characteristic vector of the high-frequency pulse current is used as the target key vector. Simultaneously, the server obtains the (n+Δ)th key vector from the ultra-high frequency electromagnetic wave characteristic sequence from the value sequence. Min up to the n+△ Max The value vector corresponding to the characteristic vector of the ultra-high frequency electromagnetic wave, and the (n+Δ)th characteristic vector in the high frequency pulse current characteristic sequence. Min up to the n+△ Max The value vector corresponding to the characteristic vector of a high-frequency pulse current is used as the target value vector.

[0124] In this embodiment, by considering the possible time delay between physical signals of different modes to determine the corresponding offset interval, and by determining the corresponding target key vector and target value vector based on the offset interval, the query volume in the Query-Key-Value mechanism can be reduced and the query accuracy can be improved.

[0125] In some embodiments, step S204 above involves segmenting and encoding each signal time-series data to obtain a feature sequence corresponding to each signal time-series data. The feature sequence includes feature vectors of multiple signal time-series segments of the corresponding signal time-series data, and includes the following steps: for each signal time-series data, segmenting the signal time-series data into multiple signal time-series segments; encoding each signal time-series segment to obtain a feature vector corresponding to each signal time-series segment; and combining the feature vectors according to the order of the signal time-series segments corresponding to each feature vector in the signal time-series data to obtain a feature sequence corresponding to the signal time-series data.

[0126] In this embodiment, for each signal timing segment, the server first performs preliminary encoding on the signal timing segment through a linear projection layer to obtain an initial feature vector. Then, the server adds a position identifier and a modal identifier to the beginning or end of the initial feature vector to obtain a corresponding initial feature vector with added identifiers. The position identifier is used to characterize the position of the corresponding signal timing segment in the signal timing data to which it belongs, such as which segment; the modal identifier is used to characterize the physical signal to which the corresponding signal timing segment belongs. Next, the server encodes the feature vector with added identifiers through the Transformer encoder corresponding to the physical signal to which the signal timing segment belongs, to obtain a corresponding feature vector. The Transformer encoder does not encode the position identifier and modal identifier in the feature vector with added identifiers.

[0127] In this embodiment, the server is able to obtain the feature sequence corresponding to each signal timing data.

[0128] To more clearly illustrate the insulation defect diagnosis method for gas-insulated switchgear provided in this application, a specific embodiment is used below to describe the method. However, it should be understood that the embodiments of this application are not limited thereto. Figure 4 As shown, in some embodiments, this application also provides a multimodal insulation defect diagnosis method, specifically including the following steps:

[0129] 1. Data acquisition and preprocessing.

[0130] The original signal timing data of ultrasonic signal, ultra-high frequency electromagnetic wave signal and high frequency pulse current signal of gas-insulated switchgear under operating conditions are acquired. The original signal timing data of each type are denoised and normalized to obtain the signal timing data of ultrasonic signal, ultra-high frequency electromagnetic wave signal and high frequency pulse current signal.

[0131] 2. Fragmentation and encoding processing.

[0132] For each signal time series data, the signal time series data is divided into multiple signal time series segments of equal length according to a preset number. For each signal time series segment, the signal time series segment is initially encoded through a linear projection layer to obtain the corresponding initial feature vector. Then, position identifiers and modal identifiers are added to the beginning or end of the initial feature vector to obtain the corresponding feature vector with added identifiers. Next, the feature vector with added identifiers is encoded through the Transformer encoder corresponding to the signal time series data to obtain the corresponding feature vector. The feature vectors of the signal time series data are combined to obtain the feature sequence corresponding to the signal time series data.

[0133] 3. Feature enhancement.

[0134] For each feature sequence, concatenate all other feature sequences to obtain a concatenated feature sequence. Using this feature sequence as the query, determine the query sequence. Using the concatenated feature sequence as the key and value, determine the key sequence and value sequence. Utilizing cross-modal attention and Query-Key-Value mechanisms, determine the fusion weights of each value vector in the value sequence under each query vector based on the query vectors in the query sequence and the key vectors in the key sequence. Based on the fusion weights of each value vector under each query vector, perform fusion processing on each value vector under each query vector to find content related to each query vector and enhance each query vector, obtaining the enhanced feature sequence corresponding to this feature sequence.

[0135] 4. Feature fusion and insulation defect diagnosis.

[0136] The enhanced feature sequences of ultrasonic signals, ultra-high frequency electromagnetic wave signals, and high frequency pulse current signals are spliced ​​together to obtain the target feature sequence of gas-insulated switchgear. The target feature sequence is then input into a multilayer sensor to diagnose the type of insulation defects present in the gas-insulated switchgear.

[0137] In this embodiment, firstly, a cross-modal attention mechanism enables deep interaction between ultrasonic signals, ultra-high frequency electromagnetic wave signals, and high-frequency pulsed current signals, allowing the multilayer perceptron to learn more complex and fundamental nonlinear relationships among the three modes. Secondly, due to more thorough fusion, the multilayer perceptron can comprehensively utilize the high sensitivity of ultra-high frequency electromagnetic wave signals and high-frequency pulsed current signals, as well as the localization characteristics of ultrasonic signals, thereby effectively suppressing noise and interference from a single mode and significantly improving the accuracy of identifying early weak defects and complex defect types. Thirdly, when a certain mode signal is severely interfered with or lacks information, the cross-modal attention mechanism allows it to obtain more effective information from other modes for compensation, thereby improving the robustness of the entire diagnostic process.

[0138] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0139] Based on the same inventive concept, this application also provides an insulation defect diagnosis device for gas-insulated combined electrical appliances, used to implement the insulation defect diagnosis method for gas-insulated combined electrical appliances described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the insulation defect diagnosis device for gas-insulated combined electrical appliances provided below can be found in the limitations of the insulation defect diagnosis method for gas-insulated combined electrical appliances described above, and will not be repeated here.

[0140] In some embodiments, such as Figure 5 As shown, an insulation defect diagnosis device for gas-insulated switchgear is provided, comprising: a signal acquisition module 502, a signal processing module 504, a feature enhancement module 506, and a defect diagnosis module 508, wherein:

[0141] The signal acquisition module 502 is used to acquire the signal timing data of at least two physical signals of the gas-insulated switchgear in operation.

[0142] The signal processing module 504 is used to perform segmentation and encoding processing on each signal time-series data to obtain the feature sequence corresponding to each signal time-series data; the feature sequence includes the feature vectors of multiple signal time-series segments of the corresponding signal time-series data.

[0143] The feature enhancement module 506 is used to enhance the feature sequence for each feature sequence based on the feature sequences other than the feature sequence itself, so as to obtain the enhanced feature sequence corresponding to the feature sequence.

[0144] The defect diagnosis module 508 is used to fuse various enhanced feature sequences to obtain the target feature sequence of the gas-insulated switchgear, and to determine the type of insulation defect in the gas-insulated switchgear based on the target feature sequence.

[0145] In one embodiment, the feature enhancement module 506 is further configured to concatenate each second feature sequence corresponding to each first feature sequence to obtain a concatenated feature sequence; the first feature sequence is any one of the feature sequences, and the second feature sequence is any one of the feature sequences other than the first feature sequence; a query sequence is determined based on the first feature sequence, and a key sequence and a value sequence are determined based on the concatenated feature sequence; the query sequence includes multiple query vectors, the key sequence includes multiple key vectors, and the value sequence includes multiple value vectors; the fusion weight of each value vector under each query vector is determined based on each query vector and each key vector; and each query vector and each value vector are fused under each query vector based on the fusion weight of each value vector under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0146] In one embodiment, the feature enhancement module 506 is further configured to map each feature vector in the first feature sequence to a corresponding query vector according to the query mapping weight information to obtain a query sequence; map each feature vector in the concatenated feature sequence to a corresponding key vector according to the key mapping weight information to obtain a key sequence; and map each feature vector in the concatenated feature sequence to a corresponding value vector according to the value mapping weight information to obtain a value sequence.

[0147] In one embodiment, the multiple query vectors are query vectors corresponding to each feature vector in the first feature sequence; the multiple key vectors are key vectors corresponding to each feature vector in the concatenated feature sequence; and the multiple value vectors are value vectors corresponding to each feature vector in the concatenated feature sequence.

[0148] The feature enhancement module 506 is also used to determine the matching degree between the query vector and each key vector for each query vector; and to determine the fusion weight of the value vector corresponding to each key vector under the query vector based on the matching degree between the query vector and each key vector; the value vector corresponding to each key vector is the value vector of the same feature vector as the key vector.

[0149] In one embodiment, the feature enhancement module 506 is further configured to perform fusion processing on each value vector according to the fusion weight of each value vector under the query vector for each query vector, to obtain the fusion value vector corresponding to the query vector, and to fuse the query vector and the fusion value vector; and to combine the enhanced feature vectors corresponding to each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0150] In one embodiment, the feature vector in the first feature sequence is a first feature vector, and the multiple query vectors are query vectors corresponding to each first feature vector in the first feature sequence; the feature vector in the second feature sequence is a second feature vector, the multiple key vectors are key vectors corresponding to each second feature vector in the concatenated feature sequence, and the multiple value vectors are value vectors corresponding to each second feature vector in the concatenated feature sequence.

[0151] The feature enhancement module 506 is further configured to, for each query vector, determine the offset interval of the query vector based on the query position identifier and offset window carried by the query vector; the query position identifier is used to characterize the position of the first feature vector corresponding to the query vector in the first feature sequence; the offset window is determined based on the gas-insulated combined electrical appliances; based on the offset interval, determine multiple target key vectors corresponding to the query vector from the key sequence, and determine multiple target value vectors corresponding to the query vector from the value sequence; the target key vector is a key vector whose carried key position identifier is located within the offset interval, and the key position identifier is used to characterize the position of the second feature vector corresponding to the key vector in its respective second feature sequence; the target value vector is a value vector whose carried value position identifier is located within the offset interval, and the value position identifier is used to characterize the position of the second feature vector corresponding to the value vector in its respective second feature sequence; determine the fusion weight of each target value vector under the query vector based on the query vector and each target key vector; and perform fusion processing on each query vector and its corresponding target key vectors under each query vector based on the fusion weight of each target key vector corresponding to each query vector under each query vector to obtain the enhanced feature sequence corresponding to the first feature sequence.

[0152] In one embodiment, the signal processing module 504 is further configured to perform segmentation processing on each signal timing data, dividing the signal timing data into multiple signal timing segments; perform encoding processing on each signal timing segment to obtain a feature vector corresponding to each signal timing segment; and combine the feature vectors according to the order of the signal timing segments corresponding to each feature vector in the signal timing data to obtain a feature sequence corresponding to the signal timing data.

[0153] Each module in the aforementioned gas-insulated switchgear insulation defect diagnosis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0154] In some embodiments, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores the signal values ​​of the physical signals of the gas-insulated switchgear. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for diagnosing insulation defects in gas-insulated switchgear.

[0155] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0156] In some embodiments, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0157] In some embodiments, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0158] In some embodiments, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0159] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0160] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0161] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method of diagnosing an insulation defect of a gas-insulated combined electrical apparatus, characterized by, The method comprises: acquiring signal time sequence data of at least two physical signals of a gas insulated combined electric appliance in a running state; performing slice processing and encoding processing on each of the signal time sequence data to obtain a feature sequence corresponding to each of the signal time sequence data; the feature sequence comprises feature vectors of multiple signal time sequence segments of the corresponding signal time sequence data; for each of the feature sequences, performing enhancement processing on the feature sequence according to the feature sequences other than the feature sequence to obtain an enhanced feature sequence corresponding to the feature sequence; fusing each of the enhanced feature sequences to obtain a target feature sequence of the gas insulated combined electric appliance, and determining a type of insulation defect existing in the gas insulated combined electric appliance based on the target feature sequence.

2. The method of claim 1, wherein, The method further comprises: for each of the feature sequences, performing enhancement processing on the feature sequence according to the feature sequences other than the feature sequence to obtain an enhanced feature sequence corresponding to the feature sequence, comprising: for each of the first feature sequences, splicing each of the second feature sequences corresponding to the first feature sequence to obtain a spliced feature sequence; the first feature sequence is any one of the feature sequences, and the second feature sequence is any one of the feature sequences other than the first feature sequence; determining a query sequence according to the first feature sequence, and determining a key sequence and a value sequence according to the spliced feature sequence; the query sequence comprises multiple query vectors, the key sequence comprises multiple key vectors, and the value sequence comprises multiple value vectors; determining a fusion weight of each of the value vectors under each of the query vectors according to each of the query vectors and each of the key vectors; 3. The method of claim 2, wherein, fusing each of the query vectors and each of the value vectors under each of the query vectors according to the fusion weight of each of the value vectors under each of the query vectors to obtain an enhanced feature sequence corresponding to the first feature sequence. The method further comprises: mapping each of the feature vectors in the first feature sequence into a corresponding query vector according to query mapping weight information to obtain the query sequence; mapping each of the feature vectors in the spliced feature sequence into a corresponding key vector according to key mapping weight information to obtain the key sequence; 4. The method of claim 2, wherein, mapping each of the feature vectors in the spliced feature sequence into a corresponding value vector according to value mapping weight information to obtain the value sequence. The multiple query vectors are query vectors corresponding to each of the feature vectors in the first feature sequence; the multiple key vectors are key vectors corresponding to each of the feature vectors in the spliced feature sequence; and the multiple value vectors are value vectors corresponding to each of the feature vectors in the spliced feature sequence. The method further comprises: for each of the query vectors, determining a matching degree between the query vector and each of the key vectors; and determining the fusion weight of each of the value vectors under each of the query vectors according to the matching degrees. determine a fusion weight of each value vector corresponding to each key vector under the query vector according to a matching degree between the query vector and each key vector; the value vector corresponding to each key vector is a value vector corresponding to a same feature vector as the key vector.

5. The method of claim 2, wherein, fusing each query vector and each value vector under each query vector according to the fusion weight of each value vector under each query vector, to obtain an enhanced feature sequence corresponding to the first feature sequence, including: for each query vector, fusing each value vector according to the fusion weight of each value vector under the query vector, to obtain a fusion value vector corresponding to the query vector; fusing the query vector and the fusion value vector, to obtain an enhanced feature vector corresponding to the query vector; combining the enhanced feature vectors corresponding to each query vector, to obtain an enhanced feature sequence corresponding to the first feature sequence.

6. The method according to any one of claims 2 to 4, characterized in that, the feature vectors in the first feature sequence are first feature vectors, and the plurality of query vectors are query vectors corresponding to each first feature vector in the first feature sequence; the feature vectors in the second feature sequence are second feature vectors, and the plurality of key vectors are key vectors corresponding to each second feature vector in the spliced feature sequence, and the plurality of value vectors are value vectors corresponding to each second feature vector in the spliced feature sequence; the method further includes: for each query vector, determining an offset interval of the query vector according to a query position identifier carried by the query vector and an offset window; the query position identifier is used to represent a position of a first feature vector corresponding to the query vector in the first feature sequence; and the offset window is determined according to the gas insulated combined electric appliance; determining a plurality of target key vectors corresponding to the query vector from the key sequence and a plurality of target value vectors corresponding to the query vector from the value sequence according to the offset interval; the target key vector is a key vector carrying a key position identifier located in the offset interval, and the key position identifier is used to represent a position of a second feature vector corresponding to the key vector in a second feature sequence to which the key vector belongs; and the target value vector is a value vector carrying a value position identifier located in the offset interval, and the value position identifier is used to represent a position of a second feature vector corresponding to the value vector in a second feature sequence to which the value vector belongs; determining a fusion weight of each target value vector under the query vector according to the query vector and each target key vector; fusing each query vector and each target key vector corresponding to the query vector under each query vector according to the fusion weight of each target key vector corresponding to each query vector under each query vector, to obtain an enhanced feature sequence corresponding to the first feature sequence.

7. The method of claim 1, wherein, the signal time sequence data are divided into fragments and encoded, to obtain a feature sequence corresponding to each signal time sequence data, including: The signal timing data is processed in pieces for each signal timing data, and the signal timing data is divided into multiple signal timing pieces; Each signal timing piece is encoded to obtain a feature vector corresponding to each signal timing piece; The feature vectors are combined according to the order of the signal timing pieces corresponding to the feature vectors in the signal timing data to obtain a feature sequence corresponding to the signal timing data.

8. A defect diagnosis device for gas-insulated combined electrical appliances, characterized in that, The device comprises: A signal acquisition module is configured to acquire signal timing data of at least two physical signals of a gas insulated switchgear in a running state; A signal processing module is configured to process each signal timing data in pieces and encode each signal timing data to obtain a feature sequence corresponding to each signal timing data; the feature sequence comprises feature vectors of multiple signal timing pieces of the corresponding signal timing data; A feature enhancement module is configured to, for each feature sequence, enhance the feature sequence according to the feature sequences other than the feature sequence to obtain an enhanced feature sequence corresponding to the feature sequence; A defect diagnosis module is configured to fuse each enhanced feature sequence to obtain a target feature sequence of the gas insulated switchgear, and determine a type of insulation defect existing in the gas insulated switchgear based on the target feature sequence. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 7.

Images