Cable insulating gas detection device

Abstract

The invention relates to the field of gas detection, in particular to a cable insulation gas detection device. The system comprises a sample collection unit, and the sample collection unit is used for collecting volatile organic compounds generated during cable insulation failure due to overheating faults; the data preprocessing unit performs feature screening on an input original feature set by adopting a mutual information theory algorithm based on a feature screening module, and performs multi-layer abstract expression on features after preliminary screening by adopting a stacking automatic coding module; and the deep learning unit performs deep learning on the data after data cleaning based on an SVM integrated model, and predicts transient stability, transient stability margin and severity classification of the data. Through integration of a high-speed data processing algorithm, a K-means clustering algorithm, a mutual information theory algorithm, a stacking automatic coding module, a noise reduction cleaning module and an SVM integration model, monitoring and deep learning prediction of volatile organic compounds generated during cable insulation failure are realized.

Description

Technical Field

[0001] The invention relates to the field of gas detection, in particular to a cable insulation gas detection device. Background Art

[0002] As the scale of AC / DC interconnected power grids continues to expand, high-penetration intermittent renewable energy generation and massive flexible load response have intensified the uncertainty on both the source and load sides of the power grid. The operation mode and dynamic behavior of the power system under renewable energy are becoming increasingly complex, which puts higher requirements on the overall decision-making level and coordinated control capabilities of dispatching operations.

[0003] In traditional power systems, cable insulation gas detection devices may not be able to efficiently process and read a large amount of sample data, resulting in slow data processing speed, affecting detection efficiency, and cable insulation gas detection devices may not be able to effectively clean up noisy and unbalanced data, resulting in the accuracy and reliability of detection results being affected. Cable insulation gas detection devices may not be able to effectively integrate multi-source heterogeneous data in complex power systems, resulting in the comprehensiveness and accuracy of detection results being limited; therefore, a cable insulation gas detection device is designed. Summary of the invention

[0004] The object of the present invention is to provide a cable insulation gas detection device to solve the problem that the cable insulation gas detection device proposed in the above background technology may not be able to efficiently process and read a large amount of sample data, resulting in slow data processing speed, affecting the detection efficiency, and the cable insulation gas detection device may not be able to effectively clean up noisy and unbalanced data, resulting in the accuracy and reliability of the detection results being affected, and also includes the problem that the cable insulation gas detection device may not be able to effectively integrate multi-source heterogeneous data in complex power systems, resulting in the comprehensiveness and accuracy of the detection results being limited.

[0005] To achieve the above object, the present invention provides a cable insulation gas detection device, comprising:

[0006] A sample collection unit, the sample collection unit is used to collect volatile organic compounds generated when the cable insulation fails due to overheating;

[0007] A data analysis unit, wherein the data analysis unit generates volatile organic compounds when the cable insulation fails due to overheating, wherein the data analysis unit includes a high-throughput sensing module, and the high-throughput sensing module monitors the volatile organic compounds based on the collected volatile organic compounds based on a high-speed data processing algorithm; and establishes an original feature set based on a K-means clustering algorithm through an analysis feature module;

[0008] A data preprocessing unit, wherein the data preprocessing unit uses a mutual information theory algorithm based on a feature screening module to perform feature screening on an input original feature set. After the initial feature screening, a stacked automatic encoding module is used to perform multi-layer abstract expression on the initially screened features, and a noise reduction and cleaning module is used to clean up noisy and unbalanced data;

[0009] A deep learning unit performs deep learning on the cleaned data based on an SVM integrated model to predict its transient stability, transient stability margin and severity classification.

[0010] As a further improvement of the present technical solution, the high-speed data processing algorithm includes convolution and least squares method, which are used for data analysis and processing. The specific steps involved in the high-speed data processing algorithm are:

[0011] By inputting the original VOCs signal f(t) and the sensor response function g(t), convolution is used to simulate the interaction between the sensor response and the VOCs signal, specifically:

[0012] ;

[0013] In the formula, Indicates VOCs signal; represents the response function of the sensor; , represents the integral variable; represents the time variable; represents the signal after being processed by the sensor response function;

[0014] By inputting the preprocessed signal (f∗g)(t), the least squares method is used to fit the preprocessed signal to estimate the concentration of VOCs, specifically:

[0015] ;

[0016] In the formula, represents the signal after being processed by the sensor response function; represents the predicted value based on the parameterized model; represents the variables input into the model; Represents the total number of data points; represents the index coefficient; Represents the VOCs concentration data points after least squares fitting.

[0017] As a further improvement of the technical solution, the K-means clustering algorithm is used to establish an original feature set based on the VOCs concentration data points fitted by the least squares method. The original feature set includes system-level static features before the fault, component-level static features of the fault point and its adjacent areas, features characterizing the impact of wind and solar power generation and DC access, and fault information features; the K-means clustering algorithm is specifically:

[0018] Estimated VOCs concentration obtained by least squares fitting , the K-means algorithm is used to cluster the fitted data to identify different VOCs categories, specifically:

[0019] ;

[0020] In the formula, Represents the VOCs concentration data points after least squares fitting; represents the jth cluster center; Indicates the number of cluster centers; Represents the difference between the quantitative data point and the cluster center; Represents the total number of data points; represents the index coefficient; Represents the original feature set.

[0021] As a further improvement of the technical solution, the mutual information theory algorithm is used to eliminate features that are irrelevant to the transient stability category. The mutual information theory algorithm is specifically:

[0022] Given a transient stability class label Y and a feature set X containing multiple features, calculate the mutual information between each feature xi and the transient stability class Y, and then sort the features according to the size of the mutual information to select the features most relevant to the transient stability class:

[0023] ;

[0024] In the formula, represents the original feature set; Represents the transient stability category, that is, the classification label of the fault; Representation characteristics The probability of simultaneous occurrence of transient stability category Y; Representation characteristics Probability of occurrence; It represents the probability of transient stability category Y occurring; represents the logarithmic function; Represents the feature set after initial screening.

[0025] As a further improvement of the technical solution, the stacked automatic encoding module includes an encoding module and a decoding module; the encoding module is used to map data to a low dimension based on a dimensionality reduction encoding algorithm; the decoding module is used to reconstruct the low-dimensional data back to the original data space to obtain high-order features based on a reconstruction decoding algorithm; specifically:

[0026] Based on the dimensionality reduction coding algorithm, the input data Mapped to a low-dimensional representation h, specifically:

[0027] ;

[0028] In the formula, Represents the weight matrix of the encoding module; Represents the bias term of the encoding module; Represents the encoded low-dimensional feature representation; represents the activation function;

[0029] The low-dimensional representation h is reconstructed back to the original data space based on the reconstruction decoding algorithm, specifically:

[0030] ;

[0031] In the formula, Represents the reconstructed data; Represents the bias term of the decoding module; Represents the weight matrix of the decoding module; Represents the encoded low-dimensional feature representation; Represents the activation function.

[0032] As a further improvement of the technical solution, the noise reduction and cleaning module performs data cleaning on noisy and unbalanced data based on a data cleaning algorithm, specifically:

[0033] ;

[0034] In the formula, Represents the cleaned data; Represents the reconstructed data; represents the mean of the feature; Represents the standard deviation of the feature.

[0035] As a further improvement of the technical solution, the SVM integrated model performs deep learning and predictive analysis on the data after data cleaning, and the specific steps involved are:

[0036] S1.1, use SDAE to extract the features of each layer as the input of each sub-learner, and evaluate the transient stability based on the SVM algorithm;

[0037] S1.2, using deep learning algorithm to predict transient stability margin;

[0038] S1.3. Based on the fuzzy logic algorithm, the temporary stability margin index is used to classify the severity of the operating mode.

[0039] As a further improvement of the technical solution, in S1.1, the SVM algorithm evaluates transient stability, specifically:

[0040] ;

[0041]

[0042] In the formula, represents the weight vector; represents the bias term; represents the feature vector; represents the target value; represents the L2 norm of the weight vector; Represents the transient stability assessment index.

[0043] As a further improvement of the technical solution, in S1.2, a deep learning algorithm is used to predict the transient stability margin, specifically:

[0044] ;

[0045] In the formula, It represents the transient stability margin prediction; represents the weight matrix; represents the output of the hidden layer; represents the bias term; Represents the activation function.

[0046] As a further improvement of the technical solution, in S1.3, the fuzzy logic algorithm performs severity classification on the operation mode based on the temporary stability margin index, specifically:

[0047] ;

[0048] In the formula, represents the membership function; represents the slope parameter of the membership function; represents the intercept parameter of the membership function; Represents the transient stability assessment index.

[0049] Severity rating, specifically:

[0050] Determine different severity levels of the transient stability margin index, define them as "safe", "critical" and "unsafe" fuzzy sets, design membership functions for each fuzzy set, these functions map the value of the index to between 0 and 1, indicating the degree to which the index belongs to the fuzzy set; according to the fuzzy logic algorithm, combine the membership function with the fuzzy rules to form a fuzzy reasoning system; these rules define which fuzzy set it should belong to under different transient stability margin index values; convert the value of the transient stability margin index into the membership of the fuzzy set through the membership function; perform fuzzy reasoning based on the fuzzy rules and the fuzzified membership to obtain the membership of each fuzzy set; convert the membership of the fuzzy set into a specific severity level.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1. This cable insulation gas detection device integrates high-speed data processing algorithm, K-means clustering algorithm, mutual information theory algorithm, stacked automatic encoding module, noise reduction and cleaning module and SVM integrated model to achieve monitoring and deep learning prediction of volatile organic compounds generated when cable insulation fails. It can effectively prevent cable failures and ensure the safe and stable operation of the power system.

[0053] 2. In this cable insulation gas detection device, a comprehensive assessment of the cable insulation status is achieved by using deep learning algorithm and fuzzy logic algorithm to predict the transient stability margin and grade the severity. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is the overall flow chart of the present invention.

[0055] The meaning of each number in the figure is:

[0056] 1. Sample collection unit; 2. Data analysis unit; 3. Data preprocessing unit; 4. Deep learning unit. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0058] Example

[0059] See also Figure 1 As shown, a cable insulation gas detection device is provided, comprising

[0060] A sample collection unit 1, the sample collection unit 1 is used to collect volatile organic compounds generated when the cable insulation fails due to overheating;

[0061] The data analysis unit 2 includes a high-throughput sensing module, which is based on a high-speed data processing algorithm and monitors volatile organic compounds according to the collected volatile organic compounds; and establishes an original feature set based on a K-means clustering algorithm through an analysis feature module; the high-speed data processing algorithm includes convolution and least squares method, which are used for data analysis and processing, and the specific steps involved in the high-speed data processing algorithm are:

[0062] By inputting the original VOCs signal f(t) and the sensor response function g(t), convolution is used to simulate the interaction between the sensor response and the VOCs signal, specifically:

[0063] ;

[0064] In the formula, Indicates VOCs signal; represents the response function of the sensor; , represents the integral variable; represents the time variable; represents the signal after being processed by the sensor response function;

[0065] By inputting the preprocessed signal (f∗g)(t), the least squares method is used to fit the preprocessed signal to estimate the concentration of VOCs, specifically:

[0066] ;

[0067] In the formula, represents the signal after being processed by the sensor response function; represents the predicted value based on the parameterized model; represents the variables input into the model; Represents the total number of data points; represents the index coefficient; Represents the VOCs concentration data points after least squares fitting.

[0068] Based on the VOCs concentration data points fitted by the least squares method, the K-means clustering algorithm is used to establish the original feature set, which includes the system-level static features before the fault, the component-level static features of the fault point and its adjacent areas, the features characterizing the impact of wind and solar power generation and DC access, and the fault information features; the K-means clustering algorithm is specifically as follows:

[0069] Estimated VOCs concentration obtained by least squares fitting , the K-means algorithm is used to cluster the fitted data to identify different VOCs categories, specifically:

[0070] ;

[0071] In the formula, Represents the VOCs concentration data points after least squares fitting; represents the jth cluster center; Indicates the number of cluster centers; Represents the difference between the quantitative data point and the cluster center; Represents the total number of data points; represents the index coefficient; Represents the original feature set.

[0072] The high-throughput sensing module uses a high-speed data processing algorithm to monitor volatile organic compounds (VOCs) generated when cable insulation fails. This process first uses convolution operations to simulate the interaction between VOCs signals and sensor response functions to obtain processed signals, and then fits these signals through the least squares method to estimate the concentration of VOCs. Next, the K-means clustering algorithm is used to cluster the fitted VOCs concentration data points to establish an original feature set containing system-level static features, component-level static features, new energy generation impact features, and fault information features to identify different VOCs categories.

[0073] The data preprocessing unit 3 uses the mutual information theory algorithm based on the feature screening module to perform feature screening on the input original feature set. After the initial feature screening, the stacked automatic encoding module is used to perform multi-layer abstract expression on the initially screened features, and the noise reduction and cleaning module is used to clean the noisy and unbalanced data; the mutual information theory algorithm is used to eliminate features that are not related to the transient stability category. The mutual information theory algorithm is specifically:

[0074] Given a transient stability class label Y and a feature set X containing multiple features, calculate the mutual information between each feature xi and the transient stability class Y, and then sort the features according to the size of the mutual information to select the features most relevant to the transient stability class:

[0075] ;

[0076] In the formula, represents the original feature set; Represents the transient stability category, that is, the classification label of the fault; Representation characteristics The probability of simultaneous occurrence of transient stability category Y; Representation characteristics Probability of occurrence; It represents the probability of transient stability category Y occurring; represents the logarithmic function; Represents the feature set after initial screening.

[0077] The stacked automatic encoding module includes an encoding module and a decoding module; the encoding module is used to map data to low dimensions based on a dimensionality reduction encoding algorithm; the decoding module is used to reconstruct low-dimensional data back to the original data space to obtain high-order features based on a reconstruction decoding algorithm; specifically:

[0078] Based on the dimensionality reduction coding algorithm, the input data Mapped to a low-dimensional representation h, specifically:

[0079] ;

[0080] In the formula, Represents the weight matrix of the encoding module; Represents the bias term of the encoding module; Represents the encoded low-dimensional feature representation; represents the activation function;

[0081] The low-dimensional representation h is reconstructed back to the original data space based on the reconstruction decoding algorithm, specifically:

[0082] ;

[0083] In the formula, Represents the reconstructed data; Represents the bias term of the decoding module; Represents the weight matrix of the decoding module; Represents the encoded low-dimensional feature representation; Represents the activation function.

[0084] Activation function, specifically:

[0085] ;

[0086] In the formula, Represents the input to the activation function; this function sets all negative inputs to 0, while positive inputs remain unchanged;

[0087] The key to the stacked autoencoder module lies in its unsupervised learning process. During training, the autoencoder tries to learn an identity function, that is, the input and output are as close as possible, but by limiting the number of nodes in the hidden layer, the network is forced to learn an effective low-dimensional representation of the input data. This process can be seen as a process of feature extraction and dimensionality reduction.

[0088] The stacked autoencoder is pre-trained layer by layer, and each autoencoder is only responsible for learning the weights and biases of the current layer, and then fixes these parameters as the input of the next layer. Finally, the performance of the entire model can be optimized by fine-tuning the parameters of the entire network.

[0089] The noise reduction and cleaning module performs data cleaning on noisy and unbalanced data based on the data cleaning algorithm, specifically:

[0090] ;

[0091] In the formula, Represents the cleaned data; Represents the reconstructed data; represents the mean of the feature; Represents the standard deviation of the feature.

[0092] The noise reduction and cleaning module uses a data cleaning algorithm to process noisy and unbalanced data. By subtracting the feature mean and dividing by the standard deviation, it achieves data noise reduction and scale unification, thereby improving data quality and providing a more reliable data foundation for subsequent modeling.

[0093] A deep learning unit 4 performs deep learning on the cleaned data based on the SVM integrated model to predict its transient stability, transient stability margin and severity classification.

[0094] The SVM integrated model performs deep learning and predictive analysis on the cleaned data. The specific steps involved are:

[0095] SDAE is used to extract the features of each layer as the input of each sub-learner, and the transient stability is evaluated based on the SVM algorithm; the SVM algorithm evaluates the transient stability, specifically:

[0096] ;

[0097]

[0098] In the formula, represents the weight vector; represents the bias term; represents the feature vector; represents the target value; represents the L2 norm of the weight vector; Represents the transient stability assessment index.

[0099] The deep learning algorithm is used to predict the transient stability margin; The deep learning algorithm is used to predict the transient stability margin, specifically:

[0100] ;

[0101] In the formula, It represents the transient stability margin prediction; represents the weight matrix; represents the output of the hidden layer; represents the bias term; Represents the activation function.

[0102] Represents the activation function, specifically:

[0103] ;

[0104] In the formula, Represents the input of the activation function; this function compresses the input to the interval (0,1);

[0105] Based on the fuzzy logic algorithm, the temporary stability margin index is used to classify the severity of the operation mode; the fuzzy logic algorithm is used to classify the severity of the operation mode based on the temporary stability margin index, specifically:

[0106] ;

[0107] In the formula, represents the membership function; represents the slope parameter of the membership function; represents the intercept parameter of the membership function; Represents the transient stability assessment index.

[0108] Severity rating, specifically:

[0109] Determine different severity levels of the transient stability margin index, define them as "safe", "critical" and "unsafe" fuzzy sets, design membership functions for each fuzzy set, these functions map the value of the index to between 0 and 1, indicating the degree to which the index belongs to the fuzzy set; according to the fuzzy logic algorithm, combine the membership function with the fuzzy rules to form a fuzzy reasoning system; these rules define which fuzzy set it should belong to under different transient stability margin index values; convert the value of the transient stability margin index into the membership of the fuzzy set through the membership function; perform fuzzy reasoning based on the fuzzy rules and the fuzzified membership to obtain the membership of each fuzzy set; convert the membership of the fuzzy set into a specific severity level.

[0110] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A cable insulation gas detection device, characterized in that: include A sample collection unit (1), the sample collection unit (1) being used to collect volatile organic matter generated when cable insulation fails due to overheating; A data analysis unit (2), the data analysis unit (2) comprising a high-throughput sensing module, the high-throughput sensing module monitoring volatile organic compounds based on the collected volatile organic compounds based on a high-speed data processing algorithm; and establishing an original feature set based on a K-means clustering algorithm through a feature analysis module; A data preprocessing unit (3), wherein the data preprocessing unit (3) uses a mutual information theory algorithm based on a feature screening module to perform feature screening on an input original feature set, and after the initial feature screening, uses a stacked automatic encoding module to perform multi-layer abstract expression on the initially screened features, and uses a noise reduction and cleaning module to clean up noisy and unbalanced data; A deep learning unit (4) is used to perform deep learning on the cleaned data based on an SVM integrated model to predict its transient stability, transient stability margin and severity classification.

2. The cable insulation gas detection device according to claim 1, characterized in that: The high-speed data processing algorithm includes convolution and least squares method, which are used for data analysis and processing. The specific steps involved in the high-speed data processing algorithm are: By inputting the original VOCs signal f(t) and the sensor response function g(t), convolution is used to simulate the interaction between the sensor response and the VOCs signal, specifically: ； In the formula, Indicates VOCs signal; represents the response function of the sensor; , represents the integral variable; represents the time variable; represents the signal after being processed by the sensor response function; By inputting the preprocessed signal (f∗g)(t), the least squares method is used to fit the preprocessed signal to estimate the concentration of VOCs, specifically: ； In the formula, represents the signal after being processed by the sensor response function; represents the predicted value based on the parameterized model; represents the variables input into the model; Represents the total number of data points; represents the index coefficient; Represents the VOCs concentration data points after least squares fitting.

3. The cable insulation gas detection device according to claim 2, characterized in that: Based on the VOCs concentration data points fitted by the least squares method, the K-means clustering algorithm is used to establish the original feature set, which includes the system-level static features before the fault, the component-level static features of the fault point and its adjacent areas, the features characterizing the impact of wind and solar power generation and DC access, and the fault information features; the K-means clustering algorithm is specifically as follows: Estimated VOCs concentration obtained by least squares fitting , the K-means algorithm is used to cluster the fitted data to identify different VOCs categories, specifically: ； In the formula, Represents the VOCs concentration data points after least squares fitting; represents the jth cluster center; Indicates the number of cluster centers; Represents the difference between the quantitative data point and the cluster center; Represents the total number of data points; represents the index coefficient; Represents the original feature set.

4. The cable insulation gas detection device according to claim 1, characterized in that: The mutual information theory algorithm is used to eliminate features that are irrelevant to the transient stability category. The mutual information theory algorithm is specifically: ； In the formula, represents the original feature set; Indicates transient stability category; Representation characteristics The probability of simultaneous occurrence of transient stability category Y; Representation characteristics Probability of occurrence; It represents the probability of transient stability category Y occurring; represents the logarithmic function; Represents the feature set after initial screening.

5. The cable insulation gas detection device according to claim 1, characterized in that: The stacked automatic encoding module includes an encoding module and a decoding module; the encoding module is used to map data to a low dimension based on a dimensionality reduction encoding algorithm; the decoding module is used to reconstruct the low-dimensional data back to the original data space to obtain high-order features based on a reconstruction decoding algorithm; Specifically: Based on the dimensionality reduction coding algorithm, the input data Mapped to a low-dimensional representation h, specifically: ； In the formula, Represents the weight matrix of the encoding module; Represents the bias term of the encoding module; Represents the encoded low-dimensional feature representation; represents the activation function; The low-dimensional representation h is reconstructed back to the original data space based on the reconstruction decoding algorithm, specifically: ； In the formula, Represents the reconstructed data; Represents the bias term of the decoding module; Represents the weight matrix of the decoding module; Represents the encoded low-dimensional feature representation; Represents the activation function.

6. The cable insulation gas detection device according to claim 1, characterized in that: The noise reduction and cleaning module performs data cleaning on noisy and unbalanced data based on a data cleaning algorithm, specifically: ； In the formula, Represents the cleaned data; Represents the reconstructed data; represents the mean of the feature; Represents the standard deviation of the feature.

7. The cable insulation gas detection device according to claim 1, characterized in that: The SVM integrated model performs deep learning and predictive analysis on the cleaned data, and the specific steps involved are: S1.1, use SDAE to extract the features of each layer as the input of each sub-learner, and evaluate the transient stability based on the SVM algorithm; S1.2, using deep learning algorithm to predict transient stability margin; S1.

3. Based on the fuzzy logic algorithm, the temporary stability margin index is used to classify the severity of the operating mode.

8. The cable insulation gas detection device according to claim 7, characterized in that: In S1.1, the SVM algorithm evaluates transient stability, specifically: ； In the formula, represents the weight vector; represents the bias term; represents the feature vector; represents the target value; represents the L2 norm of the weight vector; Represents the transient stability assessment index.

9. The cable insulation gas detection device according to claim 7, characterized in that: In S1.2, a deep learning algorithm is used to predict the transient stability margin, specifically: ； In the formula, It represents the transient stability margin prediction; represents the weight matrix; represents the output of the hidden layer; represents the bias term; Represents the activation function.

10. The cable insulation gas detection device according to claim 7, characterized in that: In S1.3, the fuzzy logic algorithm classifies the severity of the operation mode based on the temporary stability margin index, specifically: ； In the formula, represents the membership function; represents the slope parameter of the membership function; represents the intercept parameter of the membership function; Represents the transient stability assessment index.

Images