State monitoring method and system for gas insulated high-voltage switch of power distribution system

Abstract

The invention discloses a state monitoring method and system for a gas insulated high-voltage switch of a power distribution system, and the method comprises the steps: collecting state data which is related to the gas insulated high-voltage switch and comprises a plurality of data through a sensor, and determining a plurality of state types corresponding to the state data; determining a plurality of pieces of semantic information corresponding to the plurality of state categories based on the state data and determining association information among the plurality of state categories; for each piece of semantic information, determining at least one piece of data corresponding to the semantic information from the multiple pieces of data based on the associated information, the semantic information and the multiple pieces of data, and performing fusion processing on the semantic information and the at least one piece of data to obtain at least one piece of fused data; based on the fused data and the state data corresponding to the semantic information, generating fused data; and calling the neural network model based on the associated information and the fusion data to determine the state monitoring result. The state monitoring accuracy of the gas insulated high-voltage switch can be improved so as to improve the safety of a power distribution system.

Description

Technical Field

[0001] This application relates to the field of condition monitoring technology, and in particular to a condition monitoring method and system for gas-insulated high-voltage switches in power distribution systems. Background Technology

[0002] Gas-insulated high-voltage switchgear is a modular switchgear suitable for power distribution systems (such as 12kV power distribution systems). It can use insulating gases (such as SF6 gas) as the insulating medium to improve the safety performance of the power distribution system. However, if the gas-insulated high-voltage switchgear itself malfunctions and cannot operate normally, it can easily lead to a decline in the safety performance of the relevant power distribution system.

[0003] Therefore, accurately monitoring whether there are any abnormalities in the status of gas-insulated high-voltage switchgear is crucial for improving the safety performance of power distribution systems. Summary of the Invention

[0004] To address the aforementioned technical problems, this application proposes a method and system for monitoring the status of gas-insulated high-voltage switches in a power distribution system. This method and system can improve the accuracy of status monitoring of gas-insulated high-voltage switches, thereby enhancing the safety performance of the power distribution system.

[0005] In a first aspect, embodiments of this application provide a method for monitoring the condition of a gas-insulated high-voltage switch in a power distribution system, including: The status data related to the gas-insulated high-voltage switch is collected by sensors, and multiple status categories corresponding to the status data are determined. The status data includes multiple sets of data. Based on the state data, determine multiple semantic information items that correspond one-to-one with the multiple state categories, and determine the association information between the multiple state categories; For each semantic information, based on the association information, the semantic information, and the multiple data sets, at least one data set corresponding to the semantic information is determined from the multiple data sets, and the semantic information and the at least one data set are fused to obtain at least one fused data set corresponding to the semantic information. Based on the fused data corresponding to each of the semantic information and the state data, fused data is generated; Based on the associated information and the fused data, a neural network model is invoked to determine the status monitoring results that match the gas-insulated high-voltage switch.

[0006] Optionally, the step of determining at least one piece of data corresponding to the semantic information from the multiple pieces of data based on the association information, the semantic information, and the multiple pieces of data, and performing a fusion process on the semantic information and the at least one piece of data to obtain at least one fused piece of data corresponding to the semantic information, includes: Based on this semantic information and the associated information, the first cross-attention information is determined; For each of the multiple data sets, based on the semantic information and the data set, the second cross-attention information corresponding to the data set is determined, and based on the first cross-attention information and the second cross-attention information, the weight corresponding to the data set is determined; Based on a preset weight threshold and the weights corresponding to each of the multiple data sets, at least one data set is determined from the multiple data sets; Based on the weights corresponding to each of the at least one data set, the semantic information and the at least one data set are fused together to obtain at least one fused data set corresponding to the semantic information.

[0007] Optionally, determining the first cross-attention information based on the semantic information and the association information includes: Based on the semantic understanding features of the semantic information and the association features of the association information, cross-attention calculation is used to obtain a first attention feature pair. The first attention feature pair is then input into a denoising model based on a gating mechanism to obtain the first cross-attention information output by the denoising model based on the gating mechanism. And / or, The step of determining the second cross-attention information corresponding to the data based on the semantic information and the data includes: Based on the semantic understanding features of the semantic information and the data features of the data, cross-attention calculation is used to obtain a second attention feature pair. The second attention feature pair is then input into a denoising model based on a gating mechanism to obtain the second cross-attention information corresponding to the data output by the denoising model based on the gating mechanism.

[0008] Optionally, the gating-based denoising model includes a gating network, wherein the gating-based denoising model is configured as follows: In response to the input attention feature pairs, determine the original difference information between the input attention feature pairs; The input attention feature pairs are concatenated to obtain attention concatenated features; The attention-concatenated features are input into the gating network to obtain the gating vector output by the gating network; The gating vector and the original difference information are multiplied element-wise to generate and output cross-attention information corresponding to the input attention feature pair.

[0009] Optionally, the neural network model includes a first expert model and a second expert model, wherein the model parameters of the second expert model are more complex than those of the first expert model. The step of determining the status monitoring results matching the gas-insulated high-voltage switch by calling a neural network model based on the associated information and the fused data includes: The associated information and the fused data are input into the first expert model to obtain the first state prediction information output by the first expert model; The first state prediction information is input into the second expert model to obtain the second state prediction information output by the second expert model; Based on the second state prediction information, the state monitoring results matching the gas-insulated high-voltage switch are determined.

[0010] Optionally, the first expert model includes a model encoder, an input information construction unit, a first output layer, and P first expert units; The step of inputting the association information and the fused data into the first expert model to obtain the first state prediction information output by the first expert model includes: The association information and the fusion data are respectively input into the model encoder to obtain the association representation information corresponding to the association information and the fusion representation information corresponding to the fusion data output by the model encoder. The associated representation information and the fused representation information are input into the input information construction unit to obtain the expert unit input information output by the input information construction unit; The expert unit input information is input into t first expert units to obtain the output of the t first expert units, wherein the t first expert units are each first expert unit among the P first expert units that matches the expert unit input information, t≤P, and t and P are positive integers; The outputs of the t first expert units are input into the first output layer to obtain the first state prediction information output by the first output layer.

[0011] Optionally, the second expert model includes a second output layer and K second expert units, P < K, each first expert unit corresponds to at least one second expert unit, and each second expert unit corresponds to only one first expert unit; The step of inputting the first state prediction information into the second expert model to obtain the second state prediction information output by the second expert model includes: The first state prediction information is input into s second expert units to obtain the output of the s second expert units. The s second expert units are each of the T second expert units corresponding to the first state prediction information, and the T second expert units are each of the K second expert units corresponding to the t first expert units, where t≤T, s≤T, and K, s, and T are positive integers. The outputs of the s second expert units are input into the second output layer to obtain the second state prediction information output by the second output layer.

[0012] Optionally, determining the state monitoring result matching the gas-insulated high-voltage switch based on the second state prediction information includes: Based on the second state prediction information, the state monitoring auxiliary database associated with the gas-insulated high-voltage switch is matched to obtain at least one state monitoring auxiliary information. Based on the at least one state monitoring auxiliary information and the second state prediction information, the first pre-trained model is invoked to determine the state prediction result; Based on the at least one state monitoring auxiliary information, the second state prediction information and the state prediction result, the second pre-trained model is invoked to perform effective information content analysis on the at least one state monitoring auxiliary information to obtain the effective information content analysis result. The state monitoring result is determined based on the preset effective information content conditions, the effective information content analysis results, the state monitoring auxiliary database, and the state prediction results.

[0013] Optionally, determining the state monitoring result based on preset effective information content conditions, the effective information content analysis results, the state monitoring auxiliary database, and the state prediction results includes: If the effective information content analysis result does not meet the effective information content condition, based on the effective information content analysis result, or based on the effective information content analysis result and the state monitoring auxiliary database, the at least one state monitoring auxiliary information is updated, and the step of calling the first pre-trained model to determine the state prediction result based on the at least one state monitoring auxiliary information and the second state prediction information is returned. If the effective information content analysis results meet the effective information content conditions, the state monitoring results are determined based on the state prediction results.

[0014] Secondly, embodiments of this application provide a condition monitoring system for gas-insulated high-voltage switches in a power distribution system, including: The data acquisition and processing module is used to acquire status data related to the gas-insulated high-voltage switch through sensors and determine multiple status categories corresponding to the status data, wherein the status data includes multiple data sets. The state data processing module is used to determine, based on the state data, multiple semantic information pieces that correspond one-to-one with the multiple state categories, and to determine the association information between the multiple state categories; The first fusion module is used to determine at least one piece of data corresponding to the semantic information from the multiple pieces of data based on the association information, the semantic information, and the multiple pieces of data for each semantic information, and to perform fusion processing on the semantic information and the at least one piece of data to obtain at least one fused data corresponding to the semantic information. The second fusion module is used to generate fused data based on the fused data corresponding to each of the semantic information and the state data; The status prediction module is used to determine the status monitoring results that match the gas-insulated high-voltage switch based on the associated information and the fused data by calling a neural network model.

[0015] In summary, the embodiments of this application have at least the following beneficial effects: In this embodiment, status data related to a gas-insulated high-voltage switch is collected by sensors, and multiple status categories corresponding to the status data are determined. The status data includes multiple sets of data. Based on the status data, multiple semantic information items corresponding one-to-one with the multiple status categories are determined, and correlation information between the multiple status categories is determined. For each semantic information item, based on the correlation information, the semantic information, and the multiple sets of data, at least one set of data corresponding to the semantic information is determined from the multiple sets of data. The semantic information and the at least one set of data are then fused to obtain at least one fused set of data corresponding to the semantic information. Based on the fused set of data corresponding to each semantic information item and the status data, fused data is generated. Based on the correlation information and the fused data, a neural network model is invoked to determine the status monitoring result matching the gas-insulated high-voltage switch. This improves the accuracy of status monitoring of the gas-insulated high-voltage switch, thereby enhancing the safety performance of the power distribution system. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the status monitoring method for gas-insulated high-voltage switches in a power distribution system provided in an embodiment of this application. Figure 2 This is a schematic diagram of the status monitoring system for gas-insulated high-voltage switches in a power distribution system provided in an embodiment of this application; Figure 3This is a schematic diagram of the structure of the computer device provided in the embodiments of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments / examples are only a part of the embodiments / examples of this application, and not all of the embodiments / examples. Based on the embodiments / examples in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0018] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "multiple" means two or more. In the description of this application, the term "comprising" and its variations are open-ended, meaning "including but not limited to." The term "based on" means "at least partially based on." The term "according to" means "at least partially according to." The term "one embodiment / example" means "at least one embodiment / example"; the term "another embodiment / example" means "at least one additional embodiment / example"; the term "some embodiments / examples" means "at least some embodiments / examples."

[0019] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0020] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing specific embodiments only and is not intended to limit the application. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0021] It should be noted first that the "similarity" described in any one or more embodiments of this application can be calculated using at least one of the following: cosine similarity, Euclidean distance, Manhattan distance, Pearson correlation coefficient, etc. It should be understood that the implementation of this similarity method is merely illustrative and not intended to limit the scope of this application.

[0022] Firstly, see [the following] Figure 1 The diagram shows a schematic flowchart of a state monitoring method for a gas-insulated high-voltage switch in a power distribution system according to an embodiment of this application. This state monitoring method for a gas-insulated high-voltage switch in a power distribution system can be applied to a computer device with data processing capabilities. The method includes steps S101-S105, as detailed below.

[0023] S101, collect status data related to the gas-insulated high-voltage switch through sensors, and determine multiple status categories corresponding to the status data, wherein the status data includes multiple data sets.

[0024] In some examples, the sensor may include at least one of the following: an electrical sensor (e.g., a voltage sensor, a current sensor, etc.), a magnetic sensor, a temperature sensor, a humidity sensor, a gas sensor, a mechanical sensor, a speed / accelerometer, etc.

[0025] In some examples, at least some sensors may be installed on the gas-insulated high-voltage switch, and the sensors installed on the gas-insulated high-voltage switch may be used to detect the gas-insulated high-voltage switch to collect at least some state data (e.g., at least one of the multiple data sets), and / or, at least some sensors may be installed in the power distribution system in which the gas-insulated high-voltage switch is located, and the sensors installed in the power distribution system in which the gas-insulated high-voltage switch is located may be used to detect the power distribution system to collect at least some state data (e.g., at least one of the multiple data sets).

[0026] In some examples, since the sensors used to acquire the state data can be predetermined, the multiple state categories corresponding to the state data can be determined based on the sensor category to which each sensor belongs.

[0027] In some other examples, the state data and state category identification prompts can be input into a general model to obtain multiple state categories corresponding to the state data, output by the general model. The state category identification prompts are suitable for instructing the general model to perform category identification on the state data. Here, the multiple identified state categories can be the state categories corresponding to sub-items in the state data that the general model initially identifies as abnormal. The general model can be any general model commonly used in the field, and this application does not specifically limit it.

[0028] In some examples, the multiple state categories may include at least one of the following: electrical operating state corresponding to electrical sensors and / or magnetic sensors, mechanical action state corresponding to mechanical sensors and / or velocity / acceleration sensors, environmental adaptation state corresponding to temperature sensors and / or humidity sensors and / or gas sensors, insulation health state, etc.

[0029] S102, based on the state data, determine multiple semantic information items that correspond one-to-one with the multiple state categories, and determine the association information between the multiple state categories.

[0030] In some examples, the above-mentioned determination of multiple semantic information corresponding one-to-one with the multiple state categories based on the state data may include: determining multiple category feature information corresponding one-to-one with the multiple state categories based on the state data; and identifying semantic information matching each category feature information based on the state data. In this way, multiple semantic information corresponding one-to-one with the multiple state categories can be obtained.

[0031] Following the previous example, the category feature information can be generated based on at least one data set in the state data that matches the corresponding state category. For example, the at least one data set can be fused / merged, and the features of the fused / merged data can be extracted to generate the category feature information. Alternatively, feature information can be extracted from each of the at least one data set and then fused to generate the category feature information. This application embodiment does not specifically limit this. Since each state category has some related characteristics, the information of each related characteristic in the state data corresponding to each state category can be constructed into a corresponding category feature information. In this case, each related characteristic corresponding to each state category can correspond one-to-one with at least one data set matching that state category. For example, the relevant characteristics of the insulation health status may include the insulation dielectric loss factor (tanδ), leakage current value, partial discharge amplitude and frequency and / or insulation resistance value; the relevant characteristics of the electrical operating status may include the three-phase voltage amplitude, three-phase current RMS value, power factor, load rate and / or short-circuit current peak value; the relevant characteristics of the mechanical operation status may include the opening and closing coil current waveform, contact travel displacement, opening and closing time, operating mechanism vibration frequency and / or contact contact pressure; and the relevant characteristics of the environmental adaptability status may include the cabin temperature, humidity, gas concentration and / or dust content of the AC metal ring main switch.

[0032] In addition, for example, the state data or multiple data can be divided according to the relevant characteristics of each state category to obtain characteristic data corresponding to each state category, and the characteristic data corresponding to each state category can be encoded to obtain the category feature information corresponding to each state category.

[0033] Furthermore, for example, since the aforementioned category feature information can characterize feature values ​​(or their ranges), this semantic information can be used to indicate the relevant semantics of the corresponding category feature information. For instance, when the category feature information characterizes a deviation in the effective value of the three-phase current greater than a preset threshold (e.g., 15%), the corresponding semantic information can be used to indicate an imbalance in the three-phase current, thus posing a risk of uneven load distribution. This semantic information can also be used to indicate the changing trend of the corresponding category feature information.

[0034] Furthermore, exemplarily, the feature semantic recognition prompt, the state data, and the feature information of each category can be input into a general large language model to obtain the semantic information corresponding to each category feature information output by the large language model. The feature semantic recognition prompt is adapted to instruct the large language model to perform semantic recognition / semantic understanding on each category feature information using the state data as a global feature, thereby generating the corresponding semantic information. The large language model can be any large model commonly used in the field, and this application embodiment is not unique in it. It is also understood that this embodiment, by using state data as a global feature, can assist the large language model in more accurately generating semantic information corresponding to the category feature information.

[0035] Furthermore, exemplarily, the category association recognition prompt, the state data, and the multiple state categories can be input into a general large language model to obtain the association information between the multiple state categories output by the large language model. The category association recognition prompt is suitable for instructing the large language model to use the state data as a global feature to identify the category association between the multiple state categories. The large language model can be any large model commonly used in the field, and this application embodiment is not unique in it. It is also understood that this embodiment, by using state data as a global feature, can assist the large language model in more accurately mining the association relationships between multiple state categories.

[0036] S103, for each semantic information, based on the association information, the semantic information and the multiple data, at least one data corresponding to the semantic information is determined from the multiple data, and the semantic information and the at least one data are fused to obtain at least one fused data corresponding to the semantic information.

[0037] In some examples, data filtering prompts, the associated information, the semantic information, and the multiple sets of data can be input into a general large model to obtain at least one set of data corresponding to the semantic information output by the large model. The data filtering prompts are adapted to instruct the large model to supplement the semantic information with the associated information and filter out at least one set of data that matches the supplemented semantic information from the multiple sets of data.

[0038] In some examples, the above-described fusion process of the semantic information and the at least one piece of data to obtain at least one fused data corresponding to the semantic information may include: fusing the semantic information with each piece of data in the at least one piece of data to obtain a fused data corresponding to each piece of data in the at least one piece of data. That is, there can be a one-to-one correspondence between the at least one piece of data and the at least one fused data.

[0039] S104, based on the fused data corresponding to each of the semantic information and the state data, generate fused data.

[0040] In this embodiment, the generated fused dataset can be divided into at least one fused dataset obtained independently from different semantic information. This allows semantic content to be added to the state data using the fused dataset corresponding to the semantic information, so that subsequent related models can more efficiently and accurately identify effective information when recognizing the generated fused data.

[0041] In some examples, the fused data corresponding to each of the semantic information can be fused into the corresponding positions in the state data (e.g., the fused data can be appended to the data tail and / or data head of the corresponding position in the state data) to generate fused data.

[0042] S105, based on the associated information and the fused data, call the neural network model to determine the status monitoring result that matches the gas-insulated high-voltage switch.

[0043] In some examples, a pre-trained neural network model can be used to determine the state monitoring result based on the association information and the fused data. This neural network model can be a trained model capable of predicting using the association information and the fused data as input and the state monitoring result as output. During training, sample association information and sample fused data can be used as sample data (which also carries the expected corresponding state label, representing the expected state). The predicted state generated by the model based on this sample data is obtained. Based on the difference between the predicted state and the expected state represented by the label, a general loss function is used to calculate the loss value. A general training algorithm (e.g., gradient descent) is then used to train the model based on this loss value, so that the trained model possesses the aforementioned capabilities. For example, the model may include an input representation layer, a feature fusion and representation layer, and a prediction output layer. The input representation layer can receive data from the input model and convert the data into the desired feature vector form. For example, the input representation layer can use word embeddings or pre-trained language models (such as bidirectional language representation models based on the Transformer architecture) to generate semantic vectors, and / or use embedding layers to generate dense vectors. The feature fusion and representation layer can be used to fuse the feature vectors converted by the input representation layer. For example, the feature fusion and representation layer can implement the fusion through fully connected layers or attention mechanism layers. The prediction output layer can be used to generate prediction results based on the fused features. For example, the prediction output layer can use a softmax layer to output the probability distribution for different categories, and then output the prediction result based on the probability (for example, it can output the top one or more classification results with the highest probability as the prediction result).

[0044] In one optional implementation, the step of determining at least one piece of data corresponding to the semantic information from the multiple pieces of data based on the association information, the semantic information, and the multiple pieces of data, and performing a fusion process on the semantic information and the at least one piece of data to obtain at least one fused piece of data corresponding to the semantic information, includes: Based on this semantic information and the associated information, the first cross-attention information is determined; For each of the multiple data sets, based on the semantic information and the data set, the second cross-attention information corresponding to the data set is determined, and based on the first cross-attention information and the second cross-attention information, the weight corresponding to the data set is determined; Based on a preset weight threshold and the weights corresponding to each of the multiple data sets, at least one data set is determined from the multiple data sets; Based on the weights corresponding to each of the at least one data set, the semantic information and the at least one data set are fused together to obtain at least one fused data set corresponding to the semantic information.

[0045] In some examples, a third cross-attention information can be obtained by weighting the first cross-attention information and the second cross-attention information, and the degree of relevance between the semantic information and the data can be determined based on the third cross-attention information (e.g., it can be characterized by similarity), so as to map the degree of relevance to the weight corresponding to the data (e.g., it can be based on the mapping relationship between the degree of relevance and the weight).

[0046] In some examples, the at least one data point can be any of the multiple data points whose weight is greater than the preset weight threshold. The preset weight threshold can be adjusted by the user according to the required monitoring accuracy and is not uniquely defined here.

[0047] In some examples, the above-mentioned fusion processing of the semantic information and the at least one piece of data based on the weights corresponding to each of the at least one piece of data to obtain at least one fused data corresponding to the semantic information may include: performing weighted calculations based on the at least one piece of data and its corresponding weights to obtain at least one weighted data corresponding to the at least one piece of data; and fusing the at least one weighted data with the semantic information to obtain at least one fused data corresponding to the at least one piece of data.

[0048] In one optional implementation, determining the first cross-attention information based on the semantic information and the association information includes: Based on the semantic understanding features of the semantic information and the association features of the association information, cross-attention calculation is used to obtain a first attention feature pair. The first attention feature pair is then input into a denoising model based on a gating mechanism to obtain the first cross-attention information output by the denoising model based on the gating mechanism. And / or, The step of determining the second cross-attention information corresponding to the data based on the semantic information and the data includes: Based on the semantic understanding features of the semantic information and the data features of the data, cross-attention calculation is used to obtain a second attention feature pair. The second attention feature pair is then input into a denoising model based on a gating mechanism to obtain the second cross-attention information corresponding to the data output by the denoising model based on the gating mechanism.

[0049] In some examples, cross-attention computation on two types of features (semantic understanding features and related features, or semantic understanding features and data features) can allow the information of the two features to be adaptively fused and interacted, thereby generating a pair of attention features (i.e., two attention features, such as a first attention feature pair or a second attention feature pair).

[0050] In one optional implementation, the gating-based denoising model includes a gating network, wherein the gating-based denoising model is configured as follows: In response to the input attention feature pairs, determine the original difference information between the input attention feature pairs; The input attention feature pairs are concatenated to obtain attention concatenated features; The attention-concatenated features are input into the gating network to obtain the gating vector output by the gating network; The gating vector and the original difference information are multiplied element-wise to generate and output cross-attention information corresponding to the input attention feature pair. This element-wise multiplication can achieve noise reduction.

[0051] It should be noted that, in this embodiment, the input attention feature pairs refer to the first attention feature pair and the second attention feature pair. Accordingly, when the input attention feature pair is the first attention feature pair, the corresponding cross-attention information is the first cross-attention information; when the input attention feature pair is the second attention feature pair, the corresponding cross-attention information is the second cross-attention information.

[0052] In some examples, this raw difference information can be used to represent the absolute difference between two attention features in any dimension, for example, by calculating the cosine distance between the two attention features.

[0053] In some examples, this raw difference information can be represented using the corresponding raw difference vector.

[0054] In some examples, the gated network can be built on a neural network, which may include a hidden layer and an output layer that are electrically connected in sequence, and the output layer may be configured with a sigmoid activation function.

[0055] In this embodiment, the element-wise multiplication is equivalent to weighting the gated vector and the original difference vector, thereby weighting the original difference information. Compared with denoising done by using fixed original difference information, this embodiment can selectively retain important differences and suppress irrelevant noise, thereby improving the accuracy of subsequent status monitoring and improving the safety performance of the power distribution system.

[0056] In some cases, the features of all state data can be simply concatenated or averaged and then input into the model. However, this approach may suffer from problems such as important but sparse information being overwhelmed by a large amount of irrelevant or noisy data, inability to dynamically select relevant data based on the current state of the switching equipment, and high computational overhead due to the model processing all data simultaneously.

[0057] In some cases, a standard cross-attention layer can be used, where semantic information is the query and all data is the key / value pair, and then the weights are directly calculated and fused. However, this approach may suffer from problems such as ignoring context (failing to utilize related information to enhance semantic representation, resulting in isolated semantics for each state), sensitivity to noise (standard attention is easily affected by irrelevant data, leading to false attention), and singular decision-making (weights are based solely on the similarity of semantic-data pairs, lacking multi-dimensional considerations).

[0058] In one optional implementation, the neural network model includes a first expert model and a second expert model, wherein the model parameters of the second expert model are more complex than those of the first expert model. The step of determining the status monitoring results matching the gas-insulated high-voltage switch by calling a neural network model based on the associated information and the fused data includes: The associated information and the fused data are input into the first expert model to obtain the first state prediction information output by the first expert model; The first state prediction information is input into the second expert model to obtain the second state prediction information output by the second expert model; Based on the second state prediction information, the state monitoring results matching the gas-insulated high-voltage switch are determined.

[0059] In some examples, this second state prediction information can be directly used as the state monitoring result matched with the gas-insulated high-voltage switch.

[0060] In one optional implementation, the first expert model includes a model encoder, an input information construction unit, a first output layer, and P first expert units; The step of inputting the association information and the fused data into the first expert model to obtain the first state prediction information output by the first expert model includes: The association information and the fusion data are respectively input into the model encoder to obtain the association representation information corresponding to the association information and the fusion representation information corresponding to the fusion data output by the model encoder. The associated representation information and the fused representation information are input into the input information construction unit to obtain the expert unit input information output by the input information construction unit; The expert unit input information is input into t first expert units to obtain the output of the t first expert units, wherein the t first expert units are each first expert unit among the P first expert units that matches the expert unit input information, t≤P, and t and P are positive integers; The outputs of the t first expert units are input into the first output layer to obtain the first state prediction information output by the first output layer.

[0061] In some examples, at least some of the first expert units and / or at least some of the second expert units may employ a hybrid expert network.

[0062] In some examples, the model encoder can be used to encode association information to obtain association representation information, or it can be used to transform association information into a specific space (e.g., latent space) through encoding processing to obtain association representation information. Accordingly, the model encoder can be used to encode fused data to obtain fused representation information, or it can be used to transform fused data into a specific space (e.g., latent space) through encoding processing to obtain fused representation information.

[0063] In some examples, the input information construction unit can be used to fuse (e.g., splice) the associated representation information and the fused representation information to obtain expert unit input information; or, the input information construction unit can be used to fuse (e.g., splice) the associated representation information and the fused representation information to obtain fused information, and also to perform self-attention calculation on the fused information, thereby adjusting the fused information to obtain expert unit input information.

[0064] In some examples, the first output layer can be used to map the outputs of the t first expert units to the corresponding first state prediction information through a preset activation function (such as the Softmax function); or, the first output layer can be used to fuse the outputs of the t first expert units and then map them to the corresponding first state prediction information through a preset activation function.

[0065] In one optional implementation, the second expert model includes a second output layer and K second expert units, P < K, each first expert unit corresponds to at least one second expert unit, and each second expert unit corresponds to only one first expert unit. The step of inputting the first state prediction information into the second expert model to obtain the second state prediction information output by the second expert model includes: The first state prediction information is input into s second expert units to obtain the output of the s second expert units. The s second expert units are each of the T second expert units corresponding to the first state prediction information, and the T second expert units are each of the K second expert units corresponding to the t first expert units, where t≤T, s≤T, and K, s, and T are positive integers. The outputs of the s second expert units are input into the second output layer to obtain the second state prediction information output by the second output layer.

[0066] In some examples, the input information of the expert unit can be fed into each of the P first expert units to calculate the probability distribution information (e.g., through softmax), thereby obtaining the probability corresponding to each of the P first expert units. Then, based on the comparison between this probability and a preset probability threshold, t first expert units are determined from the P first expert units. Here, the t first expert units can be expert units with probabilities greater than the preset probability threshold. This probability represents the degree of correlation between the input information of the expert unit and the corresponding first expert unit (i.e., the processing ability of the first expert unit for the input information of the expert unit). In this way, t first expert units with strong processing ability for the input information of the expert unit can be determined, so as to improve the accuracy and reliability of the first state analysis information obtained by processing.

[0067] Then, T second expert units corresponding to t first expert units can be determined from the K second expert units to further process the first state prediction information. However, before processing, it is easy to understand that the above example can be used to determine s second expert units with strong processing capabilities for the first state prediction information from the T second expert units, so as to process the first state prediction information and improve the accuracy and reliability of the final second state prediction information.

[0068] In some examples, the second output layer can be used to map the outputs of the s second expert units to the corresponding second state prediction information through a preset activation function (such as the Softmax function); or, the second output layer can be used to fuse the outputs of the s second expert units and then map them to the corresponding second state prediction information through a preset activation function.

[0069] In some examples, the corresponding first expert unit and second expert unit can be expert networks trained with knowledge data from the same domain. At least some second expert units can use a larger amount of knowledge data and / or have a higher level of data annotation during training compared to their corresponding first expert units, so that the processing power of the second expert unit is stronger than that of the first expert unit.

[0070] In some cases, certain related technologies can utilize a single trained large neural network, such as DNN (Deep Neural Networks), CNN (Convolutional Neural Networks), or Transformer, to construct the aforementioned artificial intelligence model. This model is then used to process correlated information and fused data to obtain the state monitoring results corresponding to the state data. However, compared to the "expert network" in the embodiments of this application, the technical solution in this case generally has the following problems.

[0071] 1. Lack of specialization: A neural network attempts to learn all types of analysis patterns, which can easily lead to a lack of specialization and poor performance on specific complex tasks. In contrast, the embodiments of this application have stronger specialization.

[0072] 2. High computational cost: Each analysis requires activating all parameters of the entire neural network, resulting in high computational overhead, especially when the model is large (e.g., a model built from the DNN, CNN, Transformer, etc.). In contrast, the embodiments of this application only activate specific related modules for computation each time, resulting in low computational cost.

[0073] 3. Poor interpretability: It is difficult to understand how the model makes decisions and it is difficult to trace which part is responsible for which analysis. In contrast, the embodiments of this application are processed by different expert modules, and each decision can be traced back to which specific module is responsible for the analysis.

[0074] 4. Insufficient scalability: When adding new features, the entire network needs to be retrained, while the embodiments of this application only require adding a new expert module and training it to add new features.

[0075] In one optional implementation, determining the state monitoring result matching the gas-insulated high-voltage switch based on the second state prediction information includes: Based on the second state prediction information, the state monitoring auxiliary database associated with the gas-insulated high-voltage switch is matched to obtain at least one state monitoring auxiliary information. Based on the at least one state monitoring auxiliary information and the second state prediction information, the first pre-trained model is invoked to determine the state prediction result; Based on the at least one state monitoring auxiliary information, the second state prediction information and the state prediction result, the second pre-trained model is invoked to perform effective information content analysis on the at least one state monitoring auxiliary information to obtain the effective information content analysis result. The state monitoring result is determined based on the preset effective information content conditions, the effective information content analysis results, the state monitoring auxiliary database, and the state prediction results.

[0076] In some examples, the above-mentioned matching based on the second state prediction information in a state monitoring auxiliary database associated with the gas-insulated high-voltage switch yields at least one state monitoring auxiliary information, which may include: Based on the second state prediction information, a matching is performed in the state monitoring auxiliary database to obtain M state monitoring auxiliary information that match the second state prediction information. Among these M state monitoring auxiliary information, the state monitoring auxiliary information with the higher the matching degree with the second state prediction information is ranked higher. For example, the matching degree can be obtained by calculating the cosine similarity of information features between the state monitoring auxiliary information and the second state prediction information. Calculate the ratio between the matching degree of each of the M state monitoring auxiliary information (excluding the first one) and the matching degree of the previous state monitoring auxiliary information to obtain the matching degree drop ratio of each state monitoring auxiliary information. If the matching degree drop ratio is greater than or equal to the corresponding threshold (e.g., 90% or 85%), continue to calculate the matching degree drop ratio of the next state monitoring auxiliary information. If the matching degree drop ratio is less than the corresponding threshold, combine the previous state monitoring auxiliary information and all the previous state monitoring auxiliary information to form the at least one state monitoring auxiliary information.

[0077] In some examples, the first pre-trained model described above can adopt a large language model. For example, the state prediction prompt, the at least one state monitoring auxiliary information, and the second state prediction information can be input into the first pre-trained model to obtain the state prediction result output by the first pre-trained model. The state prediction prompt can be used to instruct the first pre-trained model to update the second state prediction information with the at least one state monitoring auxiliary information as an aid to obtain the state prediction result.

[0078] In some examples, the second pre-trained model described above can employ a large language model. For instance, the effective information content analysis prompt, the at least one state monitoring auxiliary information, the second state prediction information, and the state prediction result can be input into the second pre-trained model to obtain the effective information content analysis result output by the second pre-trained model. The effective information content analysis prompt can be used to instruct the second pre-trained model to perform effective information content analysis on the at least one state monitoring auxiliary information based on the second state prediction information and the state prediction result to generate an effective information content analysis result. This effective information content analysis result can include at least one of the following: a quantity analysis result of the state monitoring auxiliary information, and a correlation analysis result of the state monitoring auxiliary information.

[0079] Thus, in this embodiment of the application, after the corresponding second state prediction information is predicted by the expert model, the advantages of the large model in terms of knowledge comprehensiveness can be further utilized to adjust the second state prediction information, thereby combining the advantages of both the expert model and the large model to improve the accuracy of state monitoring of gas-insulated high-voltage switches.

[0080] In one optional implementation, determining the state monitoring result based on preset effective information content conditions, the effective information content analysis results, the state monitoring auxiliary database, and the state prediction results includes: If the effective information content analysis result does not meet the effective information content condition, based on the effective information content analysis result, or based on the effective information content analysis result and the state monitoring auxiliary database, the at least one state monitoring auxiliary information is updated, and the step of calling the first pre-trained model to determine the state prediction result based on the at least one state monitoring auxiliary information and the second state prediction information is returned. If the effective information content analysis results meet the effective information content conditions, the state monitoring results are determined based on the state prediction results.

[0081] In some examples, the aforementioned effective information content analysis results may include the state monitoring auxiliary information quantity analysis results. The effective information content condition may include an auxiliary information quantity condition, which may require the state monitoring auxiliary information quantity analysis results to indicate that the quantity of at least one state monitoring auxiliary information is reasonable. Further, if the state monitoring auxiliary information quantity analysis results indicate that the quantity of at least one state monitoring auxiliary information is unreasonable, for example, if the quantity is too large, the quantity of at least one state monitoring auxiliary information can be reduced (starting from the last-ranked state monitoring auxiliary information in the at least one state monitoring auxiliary information list, a preset number or preset proportion of state monitoring auxiliary information can be removed in each update); or, for example, if the effective information content analysis results indicate that the quantity of at least one state monitoring auxiliary information is too small, new state monitoring auxiliary information corresponding to the second state prediction information can be matched from the state monitoring auxiliary database to supplement the quantity of at least one state monitoring auxiliary information.

[0082] In some examples, the above-mentioned effective information content analysis results may include the status monitoring auxiliary information correlation analysis results. The effective information content condition may include a correlation condition, which may require that the status monitoring auxiliary information correlation analysis results indicate that there is no missing status monitoring auxiliary information among the above at least one status monitoring auxiliary information. If there is a missing status monitoring auxiliary information, i.e. the correlation condition is not met, then the missing status monitoring auxiliary information can be obtained from the status monitoring auxiliary database based on the missing status monitoring auxiliary information indicated by the status monitoring auxiliary information correlation analysis results, thereby supplementing the missing status monitoring auxiliary information back into the above at least one status monitoring auxiliary information.

[0083] Secondly, correspondingly, the embodiments of this application also provide a status monitoring system for gas-insulated high-voltage switches in a power distribution system, which can realize all the processes of the status monitoring method for gas-insulated high-voltage switches in a power distribution system provided in the above embodiments.

[0084] See Figure 2 This illustration shows a schematic diagram of the condition monitoring system for a gas-insulated high-voltage switch in a power distribution system provided in an embodiment of this application. The condition monitoring system 200 includes: The data acquisition and processing module 201 is used to acquire status data related to the gas-insulated high-voltage switch through sensors and determine multiple status categories corresponding to the status data, wherein the status data includes multiple data sets. The state data processing module 202 is used to determine, based on the state data, multiple semantic information items corresponding one-to-one with the multiple state categories, and to determine the association information between the multiple state categories; The first fusion module 203 is used to determine at least one piece of data corresponding to the semantic information from the multiple pieces of data based on the association information, the semantic information, and the multiple pieces of data for each semantic information, and to perform fusion processing on the semantic information and the at least one piece of data to obtain at least one fused data corresponding to the semantic information. The second fusion module 204 is used to generate fused data based on the fused data corresponding to each of the semantic information and the state data; The status prediction module 205 is used to determine the status monitoring results that match the gas-insulated high-voltage switch by calling a neural network model based on the associated information and the fused data.

[0085] In one optional implementation, the step of determining at least one piece of data corresponding to the semantic information from the multiple pieces of data based on the association information, the semantic information, and the multiple pieces of data, and performing a fusion process on the semantic information and the at least one piece of data to obtain at least one fused piece of data corresponding to the semantic information, includes: Based on this semantic information and the associated information, the first cross-attention information is determined; For each of the multiple data sets, based on the semantic information and the data set, the second cross-attention information corresponding to the data set is determined, and based on the first cross-attention information and the second cross-attention information, the weight corresponding to the data set is determined; Based on a preset weight threshold and the weights corresponding to each of the multiple data sets, at least one data set is determined from the multiple data sets; Based on the weights corresponding to each of the at least one data set, the semantic information and the at least one data set are fused together to obtain at least one fused data set corresponding to the semantic information.

[0086] In one optional implementation, determining the first cross-attention information based on the semantic information and the association information includes: Based on the semantic understanding features of the semantic information and the association features of the association information, cross-attention calculation is used to obtain a first attention feature pair. The first attention feature pair is then input into a denoising model based on a gating mechanism to obtain the first cross-attention information output by the denoising model based on the gating mechanism. And / or, The step of determining the second cross-attention information corresponding to the data based on the semantic information and the data includes: Based on the semantic understanding features of the semantic information and the data features of the data, cross-attention calculation is used to obtain a second attention feature pair. The second attention feature pair is then input into a denoising model based on a gating mechanism to obtain the second cross-attention information corresponding to the data output by the denoising model based on the gating mechanism.

[0087] In one optional implementation, the gating-based denoising model includes a gating network, wherein the gating-based denoising model is configured as follows: In response to the input attention feature pairs, determine the original difference information between the input attention feature pairs; The input attention feature pairs are concatenated to obtain attention concatenated features; The attention-concatenated features are input into the gating network to obtain the gating vector output by the gating network; The gating vector and the original difference information are multiplied element-wise to generate and output cross-attention information corresponding to the input attention feature pair.

[0088] In one optional implementation, the neural network model includes a first expert model and a second expert model, wherein the model parameters of the second expert model are more complex than those of the first expert model. The step of determining the status monitoring results matching the gas-insulated high-voltage switch by calling a neural network model based on the associated information and the fused data includes: The associated information and the fused data are input into the first expert model to obtain the first state prediction information output by the first expert model; The first state prediction information is input into the second expert model to obtain the second state prediction information output by the second expert model; Based on the second state prediction information, the state monitoring results matching the gas-insulated high-voltage switch are determined.

[0089] In one optional implementation, the first expert model includes a model encoder, an input information construction unit, a first output layer, and P first expert units; The step of inputting the association information and the fused data into the first expert model to obtain the first state prediction information output by the first expert model includes: The association information and the fusion data are respectively input into the model encoder to obtain the association representation information corresponding to the association information and the fusion representation information corresponding to the fusion data output by the model encoder. The associated representation information and the fused representation information are input into the input information construction unit to obtain the expert unit input information output by the input information construction unit; The expert unit input information is input into t first expert units to obtain the output of the t first expert units, wherein the t first expert units are each first expert unit among the P first expert units that matches the expert unit input information, t≤P, and t and P are positive integers; The outputs of the t first expert units are input into the first output layer to obtain the first state prediction information output by the first output layer.

[0090] In one optional implementation, the second expert model includes a second output layer and K second expert units, P < K, each first expert unit corresponds to at least one second expert unit, and each second expert unit corresponds to only one first expert unit. The step of inputting the first state prediction information into the second expert model to obtain the second state prediction information output by the second expert model includes: The first state prediction information is input into s second expert units to obtain the output of the s second expert units. The s second expert units are each of the T second expert units corresponding to the first state prediction information, and the T second expert units are each of the K second expert units corresponding to the t first expert units, where t≤T, s≤T, and K, s, and T are positive integers. The outputs of the s second expert units are input into the second output layer to obtain the second state prediction information output by the second output layer.

[0091] In one optional implementation, determining the state monitoring result matching the gas-insulated high-voltage switch based on the second state prediction information includes: Based on the second state prediction information, the state monitoring auxiliary database associated with the gas-insulated high-voltage switch is matched to obtain at least one state monitoring auxiliary information. Based on the at least one state monitoring auxiliary information and the second state prediction information, the first pre-trained model is invoked to determine the state prediction result; Based on the at least one state monitoring auxiliary information, the second state prediction information and the state prediction result, the second pre-trained model is invoked to perform effective information content analysis on the at least one state monitoring auxiliary information to obtain the effective information content analysis result. The state monitoring result is determined based on the preset effective information content conditions, the effective information content analysis results, the state monitoring auxiliary database, and the state prediction results.

[0092] In one optional implementation, determining the state monitoring result based on preset effective information content conditions, the effective information content analysis results, the state monitoring auxiliary database, and the state prediction results includes: If the effective information content analysis result does not meet the effective information content condition, based on the effective information content analysis result, or based on the effective information content analysis result and the state monitoring auxiliary database, the at least one state monitoring auxiliary information is updated, and the step of calling the first pre-trained model to determine the state prediction result based on the at least one state monitoring auxiliary information and the second state prediction information is returned. If the effective information content analysis results meet the effective information content conditions, the state monitoring results are determined based on the state prediction results.

[0093] Thirdly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any of the above-mentioned embodiments.

[0094] Fourthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the method described in any of the above-described embodiments.

[0095] Fifthly, embodiments of this application provide a computer device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the steps of the method described in any of the preceding claims.

[0096] See Figure 3 The computer device in this embodiment includes a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 301, such as a status monitoring program for gas-insulated high-voltage switches in a power distribution system. When the processor 301 executes the computer program, it implements the steps in the aforementioned embodiments of the status monitoring methods for gas-insulated high-voltage switches in power distribution systems, for example... Figure 1 The steps S101-S105 are shown.

[0097] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 302 and executed by the processor 301 to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the computer device.

[0098] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that the schematic diagram is merely an example of a computer device and does not constitute a limitation on the computer device. It may include more or fewer components than shown, or combine certain components, or different components. For example, the computer device may also include input / output devices, network access devices, buses, etc.

[0099] The processor 301 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor 301 can be any conventional processor. The processor 301 is the control center of the computer device, connecting various parts of the entire computer device through various interfaces and lines.

[0100] The memory 302 can be used to store the computer programs and / or modules. The processor 301 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 302 and calling the data stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 302 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0101] Wherein, if the modules / units integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a non-transitory computer-readable storage medium. When the computer program is executed by the processor 301, it can implement the steps of the various method embodiments described above. Wherein, the computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0102] In summary, the embodiments of this application have at least the following beneficial effects: In this embodiment, status data related to a gas-insulated high-voltage switch is collected by sensors, and multiple status categories corresponding to the status data are determined. The status data includes multiple sets of data. Based on the status data, multiple semantic information items corresponding one-to-one with the multiple status categories are determined, and correlation information between the multiple status categories is determined. For each semantic information item, based on the correlation information, the semantic information, and the multiple sets of data, at least one set of data corresponding to the semantic information is determined from the multiple sets of data. The semantic information and the at least one set of data are then fused to obtain at least one fused set of data corresponding to the semantic information. Based on the fused set of data corresponding to each semantic information item and the status data, fused data is generated. Based on the correlation information and the fused data, a neural network model is invoked to determine the status monitoring result matching the gas-insulated high-voltage switch. This improves the accuracy of status monitoring of the gas-insulated high-voltage switch, thereby enhancing the safety performance of the power distribution system.

[0103] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary hardware platforms, or it can be implemented entirely by hardware. Based on this understanding, all or part of the technical solutions of this application that contribute to the background technology can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM (Read-Only Memory) / RAM (Random Access Memory), magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0104] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A method for monitoring the condition of a gas-insulated high-voltage switch in a power distribution system, characterized in that, include: The status data related to the gas-insulated high-voltage switch is collected by sensors, and multiple status categories corresponding to the status data are determined. The status data includes multiple sets of data. Based on the state data, determine multiple semantic information items that correspond one-to-one with the multiple state categories, and determine the association information between the multiple state categories; For each semantic information, based on the association information, the semantic information, and the multiple data sets, at least one data set corresponding to the semantic information is determined from the multiple data sets, and the semantic information and the at least one data set are fused to obtain at least one fused data set corresponding to the semantic information. Based on the fused data corresponding to each of the semantic information and the state data, fused data is generated; Based on the associated information and the fused data, a neural network model is invoked to determine the status monitoring results that match the gas-insulated high-voltage switch.

2. The method according to claim 1, characterized in that, The step of determining at least one piece of data corresponding to the semantic information from the multiple pieces of data based on the association information, the semantic information, and the multiple pieces of data, and then fusing the semantic information and the at least one piece of data to obtain at least one fused piece of data corresponding to the semantic information, includes: Based on this semantic information and the associated information, the first cross-attention information is determined; For each of the multiple data sets, based on the semantic information and the data set, the second cross-attention information corresponding to the data set is determined, and based on the first cross-attention information and the second cross-attention information, the weight corresponding to the data set is determined; Based on a preset weight threshold and the weights corresponding to each of the multiple data sets, at least one data set is determined from the multiple data sets; Based on the weights corresponding to each of the at least one data set, the semantic information and the at least one data set are fused together to obtain at least one fused data set corresponding to the semantic information.

3. The method according to claim 2, characterized in that, The determination of the first cross-attention information based on the semantic information and the association information includes: Based on the semantic understanding features of the semantic information and the association features of the association information, cross-attention calculation is used to obtain a first attention feature pair. The first attention feature pair is then input into a denoising model based on a gating mechanism to obtain the first cross-attention information output by the denoising model based on the gating mechanism. And / or, The step of determining the second cross-attention information corresponding to the data based on the semantic information and the data includes: Based on the semantic understanding features of the semantic information and the data features of the data, cross-attention calculation is used to obtain a second attention feature pair. The second attention feature pair is then input into a denoising model based on a gating mechanism to obtain the second cross-attention information corresponding to the data output by the denoising model based on the gating mechanism.

4. The method according to claim 3, characterized in that, The gating mechanism-based denoising model includes a gating network, wherein the gating mechanism-based denoising model is configured as follows: In response to the input attention feature pairs, determine the original difference information between the input attention feature pairs; The input attention feature pairs are concatenated to obtain attention concatenated features; The attention-concatenated features are input into the gating network to obtain the gating vector output by the gating network; The gating vector and the original difference information are multiplied element-wise to generate and output cross-attention information corresponding to the input attention feature pair.

5. The method according to claim 1, characterized in that, The neural network model includes a first expert model and a second expert model, wherein the model parameters of the second expert model are more complex than those of the first expert model. The step of determining the status monitoring results matching the gas-insulated high-voltage switch by calling a neural network model based on the associated information and the fused data includes: The associated information and the fused data are input into the first expert model to obtain the first state prediction information output by the first expert model; The first state prediction information is input into the second expert model to obtain the second state prediction information output by the second expert model; Based on the second state prediction information, the state monitoring results matching the gas-insulated high-voltage switch are determined.

6. The method according to claim 5, characterized in that, The first expert model includes a model encoder, an input information construction unit, a first output layer, and P first expert units; The step of inputting the association information and the fused data into the first expert model to obtain the first state prediction information output by the first expert model includes: The association information and the fusion data are respectively input into the model encoder to obtain the association representation information corresponding to the association information and the fusion representation information corresponding to the fusion data output by the model encoder. The associated representation information and the fused representation information are input into the input information construction unit to obtain the expert unit input information output by the input information construction unit; The expert unit input information is input into t first expert units to obtain the output of the t first expert units, wherein the t first expert units are each first expert unit among the P first expert units that matches the expert unit input information, t≤P, and t and P are positive integers; The outputs of the t first expert units are input into the first output layer to obtain the first state prediction information output by the first output layer.

7. The method according to claim 6, characterized in that, The second expert model includes a second output layer and K second expert units, P < K, each first expert unit corresponds to at least one second expert unit, and each second expert unit corresponds to only one first expert unit; The step of inputting the first state prediction information into the second expert model to obtain the second state prediction information output by the second expert model includes: The first state prediction information is input into s second expert units to obtain the output of the s second expert units. The s second expert units are each of the T second expert units corresponding to the first state prediction information, and the T second expert units are each of the K second expert units corresponding to the t first expert units, where t≤T, s≤T, and K, s, and T are positive integers. The outputs of the s second expert units are input into the second output layer to obtain the second state prediction information output by the second output layer.

8. The method according to any one of claims 5-7, characterized in that, The step of determining the state monitoring result matching the gas-insulated high-voltage switch based on the second state prediction information includes: Based on the second state prediction information, the state monitoring auxiliary database associated with the gas-insulated high-voltage switch is matched to obtain at least one state monitoring auxiliary information. Based on the at least one state monitoring auxiliary information and the second state prediction information, the first pre-trained model is invoked to determine the state prediction result; Based on the at least one state monitoring auxiliary information, the second state prediction information and the state prediction result, the second pre-trained model is invoked to perform effective information content analysis on the at least one state monitoring auxiliary information to obtain the effective information content analysis result. The state monitoring result is determined based on the preset effective information content conditions, the effective information content analysis results, the state monitoring auxiliary database, and the state prediction results.

9. The method according to claim 8, characterized in that, The determination of the state monitoring result based on the preset effective information content conditions, the effective information content analysis results, the state monitoring auxiliary database, and the state prediction results includes: If the effective information content analysis result does not meet the effective information content condition, based on the effective information content analysis result, or based on the effective information content analysis result and the state monitoring auxiliary database, the at least one state monitoring auxiliary information is updated, and the step of calling the first pre-trained model to determine the state prediction result based on the at least one state monitoring auxiliary information and the second state prediction information is returned. If the effective information content analysis results meet the effective information content conditions, the state monitoring results are determined based on the state prediction results.

10. A condition monitoring system for gas-insulated high-voltage switches in a power distribution system, characterized in that, include: The data acquisition and processing module is used to acquire status data related to the gas-insulated high-voltage switch through sensors and determine multiple status categories corresponding to the status data, wherein the status data includes multiple data sets. The state data processing module is used to determine, based on the state data, multiple semantic information pieces that correspond one-to-one with the multiple state categories, and to determine the association information between the multiple state categories; The first fusion module is used to determine at least one piece of data corresponding to the semantic information from the multiple pieces of data based on the association information, the semantic information, and the multiple pieces of data for each semantic information, and to perform fusion processing on the semantic information and the at least one piece of data to obtain at least one fused data corresponding to the semantic information. The second fusion module is used to generate fused data based on the fused data corresponding to each of the semantic information and the state data; The status prediction module is used to determine the status monitoring results that match the gas-insulated high-voltage switch based on the associated information and the fused data by calling a neural network model.

Images