GIS opening and closing state analysis method and system for one-button sequential control

By applying the Transformer model based on the differential attention mechanism in GIS devices, the time domain feature information of the three-phase knife switch is extracted, and the problem of insufficient adaptability and robustness of noise data in the prior art is solved, and the more accurate three-dimensional Phase Blade Status Detection.

CN118861692BActive Publication Date: 2024-11-26HEFEI UNIV OF TECH +5
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411337193.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-25
Publication Date
2024-11-26
Estimated Expiration
2044-09-25

AI Technical Summary

Technical Problem

The three-phase knife switches have low adaptability and robustness to noise data in existing GIS equipment, resulting in low judgment accuracy in complex environments and safety risks.

Method used

Using the Transformer model based on the differential attention mechanism, the training data set is collected and constructed, and the time domain feature information of the three-phase knife switch is extracted using the central differential attention mechanism layer, and combined with a multi-layer perceptron for analysis, outputting the prediction label results. Determine the opening and closing status.

Benefits of technology

It improves the processing capability of noise data and the robustness in complex environments, ensures accurate judgment of the three-phase knife switch status, and improves detection accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118861692B_ABST
    Figure CN118861692B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for judging the opening and closing state of a GIS switch for one-button sequential control, including: collecting ground current data when the switch of the GIS device is opened and closed in various states to construct a training data set; constructing a transformer model based on a differential attention mechanism and using the training data set for training to obtain a target model, and using the target model to obtain time domain feature information of the ground current data of the three-phase switch of the GIS device; analyzing the time domain feature information by using a multi-layer perceptron to output a predicted label result of the three-phase switch of the GIS device; further processing is performed through the predicted label result to output the opening and closing state of the three-phase switch of the GIS device. The present invention uses a differential attention mechanism and central difference to remove non-stationary characteristics, extract the local current characteristics of the three-phase switch, and can better handle complex data and noise interference, ensuring the ability to accurately judge the opening and closing state of the three-phase switch.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of GIS knife switch fracture state detection, and in particular to a GIS opening and closing state analysis method and system for one-button sequential control. Background Art

[0002] GIS equipment has many advantages in the power system, but its sealed structure makes fault location difficult. Traditionally, GIS equipment uses micro switches and indicator lights to confirm the switch status, but it is more complicated to judge the opening and closing of three-phase multi-knife switches. When the mechanical position of the GIS knife switch is not in place, it will cause serious safety risks. In serious cases, it will lead to vicious safety accidents such as live earthing. Therefore, it is very important to detect and control the opening and closing status of the three-phase knife switch in the GIS equipment. The GIS equipment is factory-set with an electric shock detection function to judge the break of the knife switch, which is also an effective method to judge the status of the GIS knife switch, but as the equipment ages, this method is inevitably wrong, so an intelligent alternative is needed to improve the accuracy of the judgment result. Summary of the invention

[0003] The present invention aims to solve the problems of low adaptability of existing state detection models to noise data and low robustness in complex environments.

[0004] The present invention solves the above technical problems through the following technical means:

[0005] The GIS opening and closing state analysis method for one-button sequential control includes the following steps:

[0006] Step 1: Collect the ground current data of the GIS equipment when the knife switch is opened and closed in various states to form a data set. By recording the equipment status and the corresponding ground current value, a comprehensive training data set is constructed;

[0007] Step 2: construct a transformer model based on the differential attention mechanism and use the training data set for training to obtain a target model, and use the target model to obtain the time domain feature information of the grounding current data of the three-phase switch of the GIS equipment; the transformer model based on the differential attention mechanism includes N layers of encoder, multi-layer perceptron layer and Softmax layer in sequence according to the data input; each layer of encoder includes a central differential attention mechanism layer (CDA), a normalization layer and a feedforward layer;

[0008] Step three: Analyze the time domain feature information by using a multi-layer perceptron to output the predicted label result of the three-phase switch of the GIS equipment; further process the predicted label result to output the opening and closing state of the three-phase switch of the GIS equipment.

[0009] Furthermore, the step 1 specifically includes:

[0010] S11. Collect the grounding current data of the three-phase switch of the GIS equipment and build a three-phase switch current data set , represents the kth sampled three-phase current data, and ;in, represents the A phase current data of the kth sampling, represents the B-phase current data of the kth sample, represents the kth sampled C-phase current data, and , x k,A,t represents the A-phase current data sampled at the t-th time in the k-th line; M={A, B, C}, 1≤k≤K, 1≤t≤T, K represents the total number of sampled data; M represents the three-phase current; T represents the total sampling time;

[0011] S12. Construct the label information set of the three-phase current data set X , where y k Indicates the three-phase current data of the kth sampling The label value of y k ∈[1,Z], Z is the state type;

[0012] S13, the three-phase current dataset with labels As the training set T tr .

[0013] Furthermore, the step 2 specifically includes:

[0014] S21, the central differential attention mechanism layer uses equations (1)-(6) to extract features from the k-th GIS three-phase switch data to obtain a multi-resolution time-frequency feature S k , from which arrive The process is represented as CDA, i.e., the central differential attention mechanism layer;

[0015]

[0016]

[0017]

[0018]

[0019]

[0020]

[0021] In formula (1) to formula (6), Q k, 、V k They are The query value and truth value of and are respectively the query value linear transformation matrix and the true value linear transformation matrix, is the difference factor, with the initial value being a random value between 0 and 1. is the eigenvector obtained by central difference, for The attention weight, express The dimension of is the weight matrix of the differential factor, and softmax represents the activation function;

[0022] S22, the N-layer encoding layer uses equations (7)-(8) to extract features from the k-th GIS three-phase switch data, thereby obtaining the local features Y' of the three-phase switch current data.

[0023]

[0024]

[0025] in Represents the output of the nth layer of the kth data, the initial input S k The input X of the model k , is the hidden representation of the nth layer, () represents the normalization function, represents a feed-forward function.

[0026] Furthermore, the step three specifically includes:

[0027] S31, output of the N-layer encoding layer Use formula (9) to get the predicted label value of the kth data ;

[0028]

[0029] Where MLP() represents multi-layer perceptron;

[0030] S32: The last layer of the N-layer encoder layer outputs the predicted label value , and use formula (10) to get the three-phase current data x k,m,t The probability value P(x) corresponding to all states k,m,t |N);

[0031]

[0032] In formula (10), W y and b y Respectively The probability weight matrix and probability deviation vector are as follows; softmax represents the activation function; and the final output is the state of the GIS three-phase switch being opened and closed.

[0033] The present invention also provides a GIS opening and closing state analysis and judgment system for one-button sequential control, comprising:

[0034] The training data construction module collects the grounding current data of the GIS equipment when the knife switch is opened and closed in various states to form a data set. By recording the equipment status and the corresponding grounding current value, a comprehensive training data set is constructed;

[0035] The target model training module constructs a transformer model based on the differential attention mechanism and uses the training data set for training to obtain the target model, and uses the target model to obtain the time domain feature information of the grounding current data of the three-phase switch of the GIS equipment; the transformer model based on the differential attention mechanism includes N layers of encoder, multi-layer perceptron layer and Softmax layer in sequence according to the data input; each layer of encoder includes a central differential attention mechanism layer (CDA), a normalization layer and a feedforward layer;

[0036] The prediction module analyzes the time domain feature information by using a multi-layer perceptron to output the predicted label result of the three-phase knife switch of the GIS equipment; further processes the predicted label result to output the opening and closing state of the three-phase knife switch of the GIS equipment.

[0037] Furthermore, the training data construction module specifically includes:

[0038] S11. Collect the grounding current data of the three-phase switch of the GIS equipment and build a three-phase switch current data set , represents the kth sampled three-phase current data, and ;in, represents the A phase current data of the kth sampling, represents the B-phase current data of the kth sample, represents the kth sampled C-phase current data, and , x k,A,t represents the A-phase current data sampled at the t-th time in the k-th line; M={A, B, C}, 1≤k≤K, 1≤t≤T, K represents the total number of sampled data; M represents the three-phase current; T represents the total sampling time;

[0039] S12. Construct the label information set of the three-phase current data set X , where y k Indicates the three-phase current data of the kth sampling The label value of y k ∈[1,Z], Z is the state type;

[0040] S13, the three-phase current dataset with labels As the training set T tr .

[0041] Furthermore, the target model training module specifically includes:

[0042] S21, the central differential attention mechanism layer uses equations (1)-(6) to extract features from the k-th GIS three-phase switch data to obtain a multi-resolution time-frequency feature S k ; Among them, arrive The process is represented as CDA, i.e., the central differential attention mechanism layer;

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] In formula (1) to formula (6), Q k 、V k They are The query value and truth value of and are respectively the query value linear transformation matrix and the true value linear transformation matrix, is the difference factor, with the initial value being a random value between 0 and 1. is the eigenvector obtained by central difference, for The attention weight, express The dimension of is the weight matrix of the differential factor, and softmax represents the activation function;

[0050] S22, the N-layer encoding layer uses equations (7)-(8) to extract features from the k-th GIS three-phase switch data, thereby obtaining the local features Y' of the three-phase switch current data.

[0051]

[0052]

[0053] in Represents the output of the nth layer of the kth data, the initial input S k The input X of the model k , is the hidden representation of the nth layer, () represents the normalization function, represents a feed-forward function.

[0054] Furthermore, the prediction module specifically includes:

[0055] S31, output of the N-layer encoding layer Use formula (9) to get the predicted label value of the kth data ;

[0056]

[0057] Where MLP() represents multi-layer perceptron;

[0058] S32: The last layer of the N-layer encoder layer outputs the predicted label value , and use formula (10) to get the three-phase current data x k,m,t The probability value P(x) corresponding to all states k,m,t |N);

[0059]

[0060] In formula (10), W y and b y Respectively The probability weight matrix and probability deviation vector are as follows; softmax represents the activation function; and the final output is the state of the GIS three-phase switch being opened and closed.

[0061] The present invention also provides a processing device, comprising at least one processor and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor can execute the above method by calling the program instructions.

[0062] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the above method.

[0063] The advantages of the present invention are:

[0064] The existing state detection models have low adaptability to noise data and low robustness in complex environments. The present invention uses the differential attention mechanism and central difference to remove non-stationary characteristics and extract the local characteristics of the current of the three-phase knife switch, which can better handle complex data and noise interference and ensure the ability to accurately judge the opening and closing states of the three-phase knife switch. The integration of the differential attention mechanism into the existing Transformer model requires modifying the encoding layer of the model to ensure that the new mechanism can work in conjunction with other parts of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a network structure diagram of a transformer model based on a differential attention mechanism in an embodiment of the present invention;

[0066] Figure 2 This is a structural diagram of the central differential attention mechanism in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] The GIS opening and closing state analysis method for one-button sequential control described in this embodiment includes the following steps:

[0069] Step 1: Collect the ground current data of the GIS equipment when the knife switch is opened and closed in various states to form a data set. By recording the equipment status and the corresponding ground current value, a comprehensive training data set is constructed;

[0070] Step 2: Construct a transformer model based on the differential attention mechanism and use the training data set for training to obtain a target model, and use the target model to obtain the time domain feature information of the grounding current data of the three-phase switch of the GIS equipment; Figure 1 , Figure 2 As shown, the transformer model based on the differential attention mechanism includes N layers of encoder, multi-layer perceptron layer and Softmax layer in sequence according to data input; each encoder layer includes a central differential attention mechanism layer (CDA), a normalization layer and a feedforward layer;

[0071] Step three: Analyze the time domain feature information by using a multi-layer perceptron to output the predicted label result of the three-phase switch of the GIS equipment; further process the predicted label result to output the opening and closing state of the three-phase switch of the GIS equipment.

[0072] Wherein, the step 1 specifically includes:

[0073] S11. Collect the grounding current data of the three-phase switch of the GIS equipment and build a three-phase switch current data set , represents the kth sampled three-phase current data, and ;in, represents the A phase current data of the kth sampling, represents the B-phase current data of the kth sample, represents the kth sampled C-phase current data, and , x k,A,t represents the A-phase current data sampled at the t-th time in the k-th line; M={A, B, C}, 1≤k≤K, 1≤t≤T, K represents the total number of sampled data; M represents the three-phase current; T represents the total sampling time;

[0074] S12. Construct the label information set of the three-phase current data set X , where y k Indicates the three-phase current data of the kth sampling The label value of y k ∈[1,Z], Z is the state type;

[0075] S13, the three-phase current dataset with labels As the training set T tr .

[0076] The step 2 specifically includes:

[0077] S21, the central differential attention mechanism layer uses equations (1)-(6) to extract features from the k-th GIS three-phase switch data to obtain a multi-resolution time-frequency feature S k ; Among them, arrive The process is represented as CDA, i.e., the central differential attention mechanism layer;

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084] In formula (1) to formula (6), Q k, 、V k They are The query value and truth value of and are respectively the query value linear transformation matrix and the true value linear transformation matrix, is the difference factor, with the initial value being a random value between 0 and 1. is the eigenvector obtained by central difference, for The attention weight, express The dimension of is the weight matrix of the differential factor, and softmax represents the activation function;

[0085] S22, the N-layer encoding layer uses equations (7)-(8) to extract features from the k-th GIS three-phase switch data, thereby obtaining the local features Y' of the three-phase switch current data.

[0086]

[0087]

[0088] in Represents the output of the nth layer of the kth data, the initial input S k The input X of the model k , is the hidden representation of the nth layer, () represents the normalization function, represents a feed-forward function.

[0089] The step three specifically includes:

[0090] S31, output of the N-layer encoding layer Use formula (9) to get the predicted label value of the kth data ;

[0091]

[0092] Where MLP() represents multi-layer perceptron;

[0093] S32: The last layer of the N-layer encoder layer outputs the predicted label value , and use formula (10) to get the three-phase current data x k,m,tThe probability value P(x) corresponding to all states k,m,t |N);

[0094]

[0095] In formula (10), W y and b y Respectively The probability weight matrix and probability deviation vector are as follows; softmax represents the activation function; and the final output is the state of the GIS three-phase switch being opened and closed.

[0096] In order to better demonstrate the superiority of the transformer model based on the differential attention mechanism in this embodiment, this embodiment uses the same data to train different models, and finally inputs the same data to obtain the results shown in Table 1:

[0097] Table 1 Comparison of experimental results

[0098]

[0099] The experimental results are summarized as follows. The transformer model (CDAformer) based on the differential attention mechanism proposed in this patent performs well, with an accuracy of 84.65%, a precision of 70.19%, a recall of 76.52%, and an F1 score of 73.21%. In addition, the robustness of the model is attributed to its advanced attention mechanism, which can effectively capture the subtle temporal dynamics inherent in GIS data. It is followed by the LSTM model with an accuracy of 75.12% and an F1 score of 51.19%. The Transformer model has an accuracy of 61.47%, the lowest accuracy of all tested models, but a high recall of 61.91%, which shows that the model has the potential to identify relevant patterns, but at the cost of increased false alarms, which is consistent with its overall performance, indicating that there is room for improvement in accurate temporal prediction. Both the TM and RF models show limitations in this specific case, with an accuracy of no more than 53.47%, indicating challenges in effectively judging the complexity of the GIS switch status problem.

[0100] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A GIS opening and closing state analysis method for one-button sequential control, characterized in that: The following steps are involved: Step 1: Collect the ground current data of the GIS equipment when the knife switch is opened and closed in various states to form a data set. By recording the equipment status and the corresponding ground current value, a comprehensive training data set is constructed; Step 2: construct a transformer model based on the differential attention mechanism and use the training data set for training to obtain a target model, and use the target model to obtain the time domain feature information of the grounding current data of the three-phase switch of the GIS equipment; the transformer model based on the differential attention mechanism includes N layers of encoder, multi-layer perceptron layer and Softmax layer in sequence according to the data input; each layer of encoder includes a central differential attention mechanism layer CDA, a normalization layer and a feedforward layer; Step 3: Analyze the time domain feature information by using a multi-layer perceptron to output the predicted label result of the three-phase switch of the GIS equipment; further process the predicted label result to output the opening and closing state of the three-phase switch of the GIS equipment; The step 1 specifically includes: S11. Collect the grounding current data of the three-phase switch of the GIS equipment and build a three-phase switch current data set , represents the kth sampled three-phase current data, and ;in, represents the A phase current data of the kth sampling, represents the B-phase current data of the kth sample, represents the kth sampled C-phase current data, and , x k,A,t represents the A-phase current data sampled at the t-th time in the k-th line; M={A, B, C}, 1≤k≤K, 1≤t≤T, K represents the total number of sampled data; M represents the three-phase current; T represents the total sampling time; S12. Construct the label information set of the three-phase current data set X , where y k Indicates the three-phase current data of the kth sampling The label value of y k ∈[1,Z], Z is the state type; S13, the three-phase current dataset with labels As the training set T tr ; The step 2 specifically includes: S21, the central differential attention mechanism layer uses equations (1)-(6) to extract features from the k-th GIS three-phase switch data to obtain a multi-resolution time-frequency feature S k ; Among them, arrive The process is represented as CDA, i.e., the central differential attention mechanism layer; In formula (1) to formula (6), Q k 、V k They are The query value and truth value of and are respectively the query value linear transformation matrix and the true value linear transformation matrix, is the differential factor, with an initial value of a random value between 0 and 1. is the eigenvector obtained by central difference, for The attention weight, express The dimension of is the weight matrix of the differential factor, and softmax represents the activation function; S22, N layer The encoding layer uses equations (7)-(8) to extract features from the k-th GIS three-phase switch data, thereby obtaining the local features Y' of the three-phase switch current data. in Represents the output of the nth layer of the kth data, the initial input S k The input X of the model k , is the hidden representation of the nth layer, () represents the normalization function, represents a feedforward function; The step three specifically includes: S31, the output of the encoding layer in layer N Use formula (9) to get the predicted label value of the kth data ; Where MLP() represents multi-layer perceptron; S32: The last layer of the N-layer encoder layer outputs the predicted label value , and use formula (10) to get the three-phase current data x k,m,t The probability value P(x) corresponding to all states k,m,t |N); In formula (10), W y and b y Respectively The probability weight matrix and probability deviation vector of GIS are obtained; softmax represents the activation function; and the final output is the state of the GIS three-phase switch on and off.

2. The GIS opening and closing status analysis system for one-button sequential control is characterized by: include: The training data construction module collects the grounding current data of the GIS equipment when the knife switch is opened and closed in various states to form a data set. By recording the equipment status and the corresponding grounding current value, a comprehensive training data set is constructed; The target model training module constructs a transformer model based on the differential attention mechanism and uses the training data set for training to obtain the target model, and uses the target model to obtain the time domain feature information of the grounding current data of the three-phase switch of the GIS equipment; the transformer model based on the differential attention mechanism includes N layers of encoder, multi-layer perceptron layer and Softmax layer in sequence according to the data input; each layer of encoder includes a central differential attention mechanism layer CDA, a normalization layer and a feedforward layer; The prediction module uses a multi-layer perceptron to analyze the time domain feature information and outputs the predicted label result of the three-phase switch of the GIS equipment; further processes the predicted label result to output the opening and closing state of the three-phase switch of the GIS equipment; The training data construction module specifically includes: S11. Collect the grounding current data of the three-phase switch of the GIS equipment and build a three-phase switch current data set , represents the kth sampled three-phase current data, and ;in, represents the A phase current data of the kth sampling, represents the B-phase current data of the kth sample, represents the kth sampled C-phase current data, and , x k,A,t represents the A-phase current data sampled at the t-th time in the k-th line; M={A, B, C}, 1≤k≤K, 1≤t≤T, K represents the total number of sampled data; M represents the three-phase current; T represents the total sampling time; S12. Construct the label information set of the three-phase current data set X , where y k Indicates the three-phase current data of the kth sampling The label value of y k ∈[1,Z], Z is the state type; S13, the three-phase current dataset with labels As the training set T tr ; The target model training module specifically includes: S21, the central differential attention mechanism layer uses equations (1)-(6) to extract features from the k-th GIS three-phase switch data to obtain a multi-resolution time-frequency feature S k ; Among them, arrive The process is represented as CDA, i.e., the central differential attention mechanism layer; In formula (1) to formula (6), Q k 、V k They are The query value and truth value of and are respectively the query value linear transformation matrix and the true value linear transformation matrix, is the differential factor, with an initial value of a random value between 0 and 1. is the eigenvector obtained by central difference, for The attention weight, express The dimension of is the weight matrix of the differential factor, and softmax represents the activation function; S22, N layer The encoding layer uses equations (7)-(8) to extract features from the k-th GIS three-phase switch data, thereby obtaining the local features Y' of the three-phase switch current data. in Represents the output of the nth layer of the kth data, the initial input S k The input X of the model k , is the hidden representation of the nth layer, () represents the normalization function, represents a feedforward function; The prediction module specifically includes: S31, the output of the encoding layer in layer N Use formula (9) to get the predicted label value of the kth data ; Where MLP() represents multi-layer perceptron; S32: The last layer of the N-layer encoder layer outputs the predicted label value , and use formula (10) to get the three-phase current data x k,m,t The probability value P(x) corresponding to all states k,m,t |N); In formula (10), W y and b y Respectively The probability weight matrix and probability deviation vector of GIS are obtained; softmax represents the activation function; and the final output is the state of the GIS three-phase switch on and off.

3. A processing device, characterized in that: The method comprises at least one processor and at least one memory in communication with the processor, wherein the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the method as claimed in claim 1.

4. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions that cause the computer to execute the method of claim 1 .