Substation voiceprint recognition system based on trusted WLAN + centralized control system

Abstract

The invention relates to the technical field of voiceprint recognition and signal processing, in particular to a transformer substation voiceprint recognition system based on a trusted WLAN + centralized control system. The system comprises a voiceprint sensing unit used for collecting sound wave signals generated during operation of substation equipment; the signal processing unit adopts a fast independent component analysis technology, constructs a new objective function by combining and utilizing a complex network analysis idea, de-noising the sound wave digital signal, and extracts a feature vector by using a wavelet packet decomposition algorithm; the WLAN transmission unit transmits the feature vector to the analysis and diagnosis unit by using a trusted WLAN; and the analysis and diagnosis unit stores the voiceprint data and performs fault analysis and diagnosis on the feature vectors through a BP neural network algorithm. According to the transformer substation voiceprint recognition system based on the credible WLAN + centralized control system, the problem of noise interference is solved through complex network analysis and a rapid independent component analysis technology, and equipment fault recognition is achieved in combination with a BP neural network algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of voiceprint recognition and signal processing, and in particular to a substation voiceprint recognition system based on a trusted WLAN+centralized control system. Background Art

[0002] The substation voiceprint recognition system based on trusted WLAN + centralized control system aims to improve the accuracy and real-time performance of equipment fault identification. By real-time monitoring of the voiceprint signals generated during equipment operation, it can identify equipment anomalies and perform fault diagnosis, control the operating status and maintenance strategy of the substation, and realize remote monitoring and intelligent operation and maintenance of equipment.

[0003] The existing substation voiceprint recognition system is usually difficult to effectively eliminate the interference of external environmental noise and achieve high-precision analysis of voiceprint signals. In addition, since the traditional voiceprint recognition system requires manual patrol to diagnose equipment defects and faults, which is time-consuming and labor-intensive and has poor processing capabilities, it will lead to delayed equipment fault detection and the inability to achieve predictive maintenance of equipment. Therefore, a substation voiceprint recognition system based on trusted WLAN+centralized control system is designed. Summary of the invention

[0004] The purpose of the present invention is to provide a substation voiceprint recognition system based on a trusted WLAN + centralized control system to solve the problem raised in the above background technology that the traditional voiceprint recognition system requires manual patrol to diagnose equipment defects and faults, which is time-consuming and labor-intensive and has poor processing capabilities, resulting in delayed equipment fault detection and inability to achieve predictive maintenance of the equipment.

[0005] To achieve the above object, the present invention aims to provide a substation voiceprint recognition system based on a trusted WLAN+centralized control system, comprising: A voiceprint sensing unit, which is used to collect sound wave signals generated when substation equipment is running; The system also includes a signal processing unit, which is used to convert the acoustic signal into a sound wave digital signal, adopt a fast independent component analysis technique, and construct a new objective function in combination with a complex network analysis idea to solve a separation matrix, denoise the sound wave digital signal, and extract a feature vector using a wavelet packet decomposition algorithm; It also includes a WLAN transmission unit, which uses a trusted WLAN to transmit the feature vector to the analysis and diagnosis unit through a data transmission channel for analysis; It also includes an analysis and diagnosis unit, which is used to store voiceprint data and perform fault analysis and diagnosis on feature vectors through a BP neural network algorithm; It also includes a centralized control system, which is used for centralized remote monitoring and management of the operation of the substation.

[0006] The voiceprint sensing unit includes a microphone and a preamplifier; Among them, the microphone is used to detect the sound wave signal generated by the equipment during operation; The preamplifier is used to amplify the sound wave signal captured by the microphone and provide it to the signal processing unit.

[0007] As a further improvement of the technical solution, the signal processing unit includes a data acquisition card, a noise separation module and a frequency feature extraction module; Wherein, the data acquisition card is used to convert the sound wave signal into a sound wave digital signal; The noise separation module uses fast independent component analysis technology and complex network analysis ideas to convert the sound wave digital signal into The effective sound signal is separated from the noise signal to obtain the separated sound wave digital signal ; The frequency feature extraction module uses a wavelet packet decomposition algorithm to extract the sound wave digital signal frequency characteristics.

[0008] As a further improvement of the technical solution, the noise separation module is based on the fast independent component analysis technology, combined with the use of complex network analysis ideas, to separate the effective sound signal and the noise signal of the sound wave digital signal to obtain the separated sound wave digital signal , the specific steps are as follows: S2.1.1. Maximizing Negative Entropy As the objective function: ; in, is the number of independent components; is a constant coefficient; is the expected value; is a non-quadratic optimization function; It is an independent signal source; is a standard Gaussian distribution variable; is the index of the independent component; S2.1.2. Select the logarithmic hyperbolic cosine function And the parameters As an optimization function: ; in, is the logarithmic hyperbolic cosine function; S2.1.3. Use complex network analysis to process frequency feature correlation and construct frequency segment correlation matrix , use the Louvain algorithm to cluster the frequency segments into different association communities according to their association ; S2.1.4. Based on the community division result matrix And the objective function Construct a new objective function: ; in, For the Communities No. frequency bands; For the Communities No. frequency bands; For frequency band and The strength of the association between is the separation matrix; is the total number of communities; S2.1.5. New objective function by gradient descent method Optimize and solve the separation matrix , the specific steps are as follows: Initialize the separation matrix to the identity matrix , represents the separation matrix at the zeroth iteration; For the new objective function Iterate until the new objective function Converge and get the final separation matrix : ; in, Indicates The separation matrix of the iteration; Indicates The separation matrix of the iteration; is the learning rate; The new objective function exist The gradient at S2.1.6. According to the separation matrix , calculate the separated sound wave digital signal : ; in, is a sound wave digital signal; It is the separated sound wave digital signal.

[0009] As a further improvement of this technical solution, in S2.1.3, the frequency feature correlation processing is performed using the complex network analysis idea to construct the frequency segment correlation matrix , use the Louvain algorithm to cluster the frequency segments into different association communities according to their association , as follows: The incidence matrix Each element of Defined as Communities Frequency band and frequency band The strength of the association between: ; in, For the Communities Frequency band and frequency band The strength of the association between For frequency band and frequency band The frequency difference between is the total number of frequency bands; For the Communities No. frequency band, ; is the sum of the absolute values ​​of frequencies in all frequency bands; For frequency band The signal energy of For frequency band The signal energy of Use Louvain's algorithm to calculate the incidence matrix Clustering is performed to obtain groups of frequency bands with strong correlation: Define modularity as : ; in, For elements half the strength of the association; For frequency band degree; For frequency band degree; Kronecker delta function, used to determine the frequency band and frequency band Whether they belong to the same community; Repeatedly adjust the modularity Optimize and get the community division result matrix , where the community partition result matrix Elements Indicates frequency band Belong to the community : .

[0010] As a further improvement of the present technical solution, the frequency feature extraction module uses a wavelet packet decomposition algorithm to extract the sound wave digital signal The frequency characteristics of the , the specific steps are as follows: S2.1.7. Sound wave digital signal conduct Layer wavelet packet decomposition: ; ; in, For the Low frequency coefficients after layer wavelet packet decomposition; For the High frequency coefficients after layer wavelet packet decomposition; For the The low-frequency components of the layer; is the filter coefficient of the low-pass filter; is the filter coefficient of the high-pass filter; is the time variable; is the filter order; is the number of wavelet packet decomposition layers; S2.1.8. Reconstruct the decomposed coefficients to obtain the low-frequency and high-frequency reconstructed signals of the signal: ; ; in, For the Low frequency components during layer reconstruction; For the High frequency components during layer reconstruction; is the coefficient of the low-pass filter used for low-frequency signal reconstruction in the wavelet packet reconstruction process; is the coefficient of the high-pass filter used for high-frequency signal reconstruction in the wavelet packet reconstruction process; S2.1.9. Calculation Frequency band signal Energy , … : ;in, For the Tier Signals in frequency bands; For the Tier The discrete point amplitude of each frequency band; For frequency band signal energy; S2.1.10, normalize the energy of each frequency band to construct a feature vector : ; in, is the normalized energy eigenvector; is the total number of frequency bands; is the current frequency band index.

[0011] As a further improvement of the technical solution, the WLAN transmission unit includes a data transmission channel, which is used to transmit the feature vector to the analysis and diagnosis unit through the data transmission channel. The specific steps are as follows: S3.1. The feature vector Packed into binary data stream; S3.2, using the encryption mechanism of the trusted WLAN to encrypt the binary data stream; S3.3. Transmit the binary data stream to the analysis and diagnosis unit via the trusted WLAN communication protocol.

[0012] As a further improvement of the technical solution, the analysis and diagnosis unit includes a station-side database and a fault analysis and diagnosis module; Wherein, the station-side database is used to store equipment operation data, historical fault data, and feature vector information; The fault analysis and diagnosis module uses a BP neural network algorithm to perform fault analysis and diagnosis on the feature vector.

[0013] As a further improvement of the present technical solution, the fault analysis and diagnosis module uses a BP neural network algorithm to perform fault analysis and diagnosis on the feature vector, and the specific steps are as follows: S4.1. Initialize the weights between the input layer and the hidden layer , and the weights between the hidden layer and the output layer , set the learning factor to ; S4.2. Determine the number of hidden layer nodes based on empirical formula : ; in, is the number of hidden layer nodes, is the number of input layer nodes, is the number of nodes in the output layer; S4.3, the input layer selects a linear activation function, then the node Output for: ; in, For Node Input; The hidden layer selects the logarithmic S-type function as the activation function, then the output for: ; ; in, is a hidden layer node Total input of is the hidden layer node The output value of is the Sigmoid activation function; The input layer node The output value of is the hidden layer node The activation value of The output layer selects the S-type function, then the node Output value for: ; in, For Node The output value of For Node The activation value of is the S-type activation function; For Node Total input of S4.4. The feature vector The input layer is passed to the hidden layer, the total input and output of each hidden layer node are calculated, and the output of the hidden layer is used as input and passed to the output layer to calculate the output of each node in the output layer. ; S4.5. Calculate the output layer nodes Output value With the expected output The total error : ; in, is the total error; is the expected output; For Node The output value of S4.6. Calculate the output layer error : ; in, The total error For actual output The partial derivative of The actual output For total input The partial derivative of is the derivative of the output layer activation function; Calculate the hidden layer error : ; in, is the derivative of the hidden layer activation function; is the accumulated error propagated back from the output layer; S4.7. Output layer error according to error and the hidden layer error , update the weights along the negative gradient direction of the error and weights : ; ; in, is the weight adjustment between the output layer and the hidden layer; is the weight adjustment between the hidden layer and the input layer; S4.8. Update the weights from the input layer to the hidden layer and from the hidden layer to the output layer: ; ; in, From the hidden layer to the output layer The weight at the iteration; From the hidden layer to the output layer The weight at the iteration; From the input layer to the hidden layer The weight at the iteration; From the input layer to the hidden layer The weight at the iteration; S4.9. Finally, we get the nodes of the output layer The identification result of the corresponding fault.

[0014] As a further improvement of the technical solution, the centralized control system is used to perform centralized remote monitoring and management of the substation, as follows: S5.1. The centralized control system remotely monitors the operating status of multiple substations and centralizes the equipment information of each site into a control center for management; S5.2. The centralized control system integrates fault monitoring and alarm functions, and works in conjunction with the voiceprint recognition system to perform abnormal analysis by detecting the sounds emitted during equipment operation, and combines the analysis and diagnosis unit to diagnose equipment faults in real time.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. In the substation voiceprint recognition system based on the trusted WLAN + centralized control system, based on the combination of complex network analysis and fast independent component analysis technology, it can accurately separate the effective voiceprint signal and environmental noise in the operation of the equipment, significantly improve the accuracy of signal processing, and eliminate noise interference in complex electromagnetic mechanical environments.

[0016] 2. In the substation voiceprint recognition system based on the trusted WLAN + centralized control system, the extracted voiceprint features are analyzed and intelligently classified through the BP neural network algorithm, realizing the automatic identification, accurate classification and predictive maintenance of substation equipment faults, ensuring timely diagnosis and early warning when abnormalities occur during the operation of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is the overall flow chart of the present invention; The meaning of each number in the figure is: 1. Voiceprint sensing unit; 2. Signal processing unit; 3. WLAN transmission unit; 4. Analysis and diagnosis unit; 5. Centralized control system. DETAILED DESCRIPTION

[0018] 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.

[0019] See also Figure 1 As shown, a substation voiceprint recognition system based on a trusted WLAN + centralized control system is provided, including: A voiceprint sensor unit 1, which is used to collect sound wave signals generated when substation equipment is running; In this embodiment, the voiceprint sensing unit 1 includes a microphone and a preamplifier; Among them, the microphone is used to detect the sound wave signal generated by the equipment during operation; The microphone is installed on the housing of the substation equipment or the nearby structure by contact or non-contact method to capture mechanical waves and vibration sounds; the sensor can be installed on the main transformer equipment of the substation in the following ways: Three contact voiceprint collectors are installed on the long side; one contact voiceprint collector is installed on the short side; The preamplifier is used to amplify the sound wave signal captured by the microphone and provide it to the signal processing unit 2.

[0020] The system also includes a signal processing unit 2, which is used to convert the acoustic signal into a sound wave digital signal, use a fast independent component analysis technique, and construct a new objective function in combination with a complex network analysis idea to solve a separation matrix, denoise the sound wave digital signal, and use a wavelet packet decomposition algorithm to extract a feature vector; In this embodiment, the signal processing unit 2 includes a data acquisition card, a noise separation module and a frequency feature extraction module; Wherein, the data acquisition card is used to convert the sound wave signal into a sound wave digital signal; The data acquisition card samples the sound wave signal, and the sampling rate of each channel is not less than 48kHz, ensuring that sufficiently detailed voiceprint information is captured at a higher resolution; The noise separation module uses fast independent component analysis technology and complex network analysis ideas to convert the sound wave digital signal into The effective sound signal is separated from the noise signal to obtain the separated sound wave digital signal ; The frequency feature extraction module uses a wavelet packet decomposition algorithm to extract the sound wave digital signal frequency characteristics.

[0021] The noise separation module is based on the fast independent component analysis technology, combined with the use of complex network analysis ideas, to separate the effective sound signal of the sound wave digital signal from the noise signal, and obtain the separated sound wave digital signal , the specific steps are as follows: S2.1.1. Maximizing Negative Entropy As the objective function: ; in, is the number of independent components; is a constant coefficient; is the expected value; is a non-quadratic optimization function; It is an independent signal source; is a standard Gaussian distribution variable; is the index of the independent component; S2.1.2. Select the logarithmic hyperbolic cosine function And the parameters As an optimization function: ; in, is the logarithmic hyperbolic cosine function; S2.1.3. Use complex network analysis to process frequency feature correlation and construct frequency segment correlation matrix , use the Louvain algorithm to cluster the frequency segments into different association communities according to their association , as follows: The incidence matrix Each element of Defined as Communities Frequency band and frequency band The strength of the association between: ; in, For the Communities Frequency band and frequency band The strength of the association between For frequency band and frequency band The frequency difference between is the total number of frequency bands; For the frequency bands; is the sum of the absolute values ​​of frequencies in all frequency bands; For frequency band The signal energy of For frequency band The signal energy of Use Louvain's algorithm to calculate the incidence matrix Clustering is performed to obtain groups of frequency bands with strong correlation: Define modularity as : ; in, For elements half the strength of the association; For frequency band degree; For frequency band degree; Kronecker delta function, used to determine the frequency band and frequency band Whether they belong to the same community; In this embodiment, The Kronecker delta function is specifically defined as: ; Repeatedly adjust the modularity Optimize and get the community division result matrix , where the community partition result matrix Elements Indicates frequency band Belong to the community : ; S2.1.4. Based on the community division result matrix And the objective function Construct a new objective function: ; in, For the Communities No. frequency bands; For the Communities No. frequency bands; For frequency band and The strength of the association between is the separation matrix; is the total number of communities; S2.1.5. New objective function by gradient descent method Optimize and solve the separation matrix , the specific steps are as follows: Initialize the separation matrix to the identity matrix , represents the separation matrix at the zeroth iteration; For the new objective function Iterate until the new objective function Converge and get the final separation matrix : ; in, Indicates The separation matrix of the iteration; Indicates The separation matrix of the iteration; is the learning rate; The new objective function exist The gradient at S2.1.6. According to the separation matrix , calculate the separated sound wave digital signal : ; in, is a sound wave digital signal; It is the separated sound wave digital signal.

[0022] The frequency feature extraction module uses a wavelet packet decomposition algorithm to extract the sound wave digital signal The frequency characteristics of the , the specific steps are as follows: S2.1.7. Sound wave digital signal conduct Layer wavelet packet decomposition: ; ; in, For the Low frequency coefficients after layer wavelet packet decomposition; For the High frequency coefficients after layer wavelet packet decomposition; For the The low-frequency components of the layer; is the filter coefficient of the low-pass filter; is the filter coefficient of the high-pass filter; is the time variable; is the filter order; is the number of wavelet packet decomposition layers; S2.1.8. Reconstruct the decomposed coefficients to obtain the low-frequency and high-frequency reconstructed signals of the signal: ; ; in, For the Low frequency components during layer reconstruction; For the High frequency components during layer reconstruction; is the coefficient of the low-pass filter used for low-frequency signal reconstruction in the wavelet packet reconstruction process; is the coefficient of the high-pass filter used for high-frequency signal reconstruction in the wavelet packet reconstruction process; S2.1.9. Calculation Frequency band signal Energy , … : ;in, For the Tier Signals in frequency bands; For the Tier The discrete point amplitude of each frequency band; For frequency band signal energy; S2.1.10, normalize the energy of each frequency band to construct a feature vector : ; in, is the normalized energy eigenvector; is the total number of frequency bands; is the current frequency band index.

[0023] In this embodiment, the Meyer wavelet basis is selected to perform a 5-layer decomposition on the signal, dividing the acoustic wave signal into 32 frequency bands, and the bandwidth of each frequency band is about 750 Hz; It also includes a WLAN transmission unit 3, which uses a trusted WLAN to transmit the feature vector to an analysis and diagnosis unit 4 through a data transmission channel for analysis; The WLAN transmission unit 3 includes a data transmission channel, which is used to transmit the feature vector to the analysis and diagnosis unit 4 through the data transmission channel. The specific steps are as follows: S3.1. The feature vector Packed into binary data stream; S3.2, using the encryption mechanism of the trusted WLAN to encrypt the binary data stream; S3.3. Transmit the binary data stream to the analysis and diagnosis unit 4 via the trusted WLAN communication protocol.

[0024] In this embodiment, the trusted WLAN communication protocol is the WPA3 encryption protocol and the 802.1X authentication protocol. The trusted WLAN has a data integrity and anti-tampering mechanism, and ensures the security of data during transmission through an additional message integrity check code MIC; It also includes an analysis and diagnosis unit 4, which is used to store voiceprint data and perform fault analysis and diagnosis on feature vectors through a BP neural network algorithm; The analysis and diagnosis unit 4 includes a station-side database and a fault analysis and diagnosis module; Wherein, the station-side database is used to store equipment operation data, historical fault data, and feature vector information; The fault analysis and diagnosis module uses a BP neural network algorithm to perform fault analysis and diagnosis on the feature vector.

[0025] In this embodiment, the fault analysis and diagnosis module uses the BP neural network algorithm to perform fault analysis and diagnosis on the feature vector. The specific steps are as follows: S4.1. Initialize the weights between the input layer and the hidden layer , and the weights between the hidden layer and the output layer , set the learning factor to ; S4.2. Determine the number of hidden layer nodes based on empirical formula : ; in, is the number of hidden layer nodes, is the number of input layer nodes, is the number of nodes in the output layer; S4.3, the input layer selects a linear activation function, then the node Output for: ; in, For Node Input; The hidden layer selects the logarithmic S-type function as the activation function, then the output for: ; ; in, is a hidden layer node Total input of is the hidden layer node The output value of is the Sigmoid activation function; The input layer node The output value of is the hidden layer node The activation value of The output layer selects the S-type function, then the node Output value for: ; in, For Node The output value of For Node The activation value of is the S-type activation function; For Node Total input of S4.4. The feature vector The input layer is passed to the hidden layer, the total input and output of each hidden layer node are calculated, and the output of the hidden layer is used as input and passed to the output layer to calculate the output of each node in the output layer. ; S4.5. Calculate the output layer nodes Output value With the expected output The total error : ; in, is the total error; is the expected output; For Node The output value of S4.6. Calculate the output layer error : ; in, The total error For actual output The partial derivative of The actual output For total input The partial derivative of is the derivative of the output layer activation function; Calculate the hidden layer error : ; in, is the derivative of the hidden layer activation function; is the accumulated error propagated back from the output layer; S4.7. Output layer error according to error and the hidden layer error , update the weights along the negative gradient direction of the error and weights : ; ; in, is the weight adjustment between the output layer and the hidden layer; is the weight adjustment between the hidden layer and the input layer; S4.8. Update the weights from the input layer to the hidden layer and from the hidden layer to the output layer: ; ; in, From the hidden layer to the output layer The weight at the iteration; From the hidden layer to the output layer The weight at the iteration; From the input layer to the hidden layer The weight at the iteration; From the input layer to the hidden layer The weight at the iteration; S4.9. Finally, we get the nodes of the output layer The identification result of the corresponding fault.

[0026] In this embodiment, the number of feature vectors is 32, so the number of input layer nodes is 32; there are 8 types of faults, so the number of output layer nodes is 8; the number of hidden layer nodes calculated is 8; In this embodiment, the expected output is the actual fault type of the transformer, represented as a binary vector, where: If the node The corresponding fault type is the fault type of the current sample, then the expected output is ; Expected output of the remaining nodes .

[0027] It also includes a centralized control system 5, which is used for centralized remote monitoring and management of the operation of the substation.

[0028] In this embodiment, the centralized control system 5 is used to perform centralized remote monitoring and management of the substation, as follows: S5.1. The centralized control system remotely monitors the operating status of multiple substations and centralizes the equipment information of each site into a control center for management; S5.2. The centralized control system integrates fault monitoring and alarm functions, and is linked with the voiceprint recognition system. It performs abnormal analysis by detecting the sound emitted during the operation of the equipment, and combines the analysis and diagnosis unit 4 to diagnose equipment faults in real time.

[0029] In this embodiment, the centralized control system has the ability to remotely control substation equipment. Operation and maintenance personnel can remotely control the start and stop of circuit breakers and switchgear through the system, perform load switching, power regulation and other operations, thereby achieving rapid response and flexible scheduling of substations; The centralized control system supports long-term storage and analysis of operating data, helping operation and maintenance personnel to understand the historical operation status of the substation and assist in decision-making; The centralized control system provides an intuitive user interface, through which operation and maintenance personnel can view equipment operating status, alarm information, historical data, fault diagnosis results, etc. The interface can support multi-level data display, including the overall operation status of the entire substation and the detailed status of individual equipment.

[0030] 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. The substation voiceprint recognition system based on trusted WLAN + centralized control system is characterized by: include: A voiceprint sensing unit (1), the voiceprint sensing unit (1) being used to collect sound wave signals generated during operation of substation equipment; A signal processing unit (2), the signal processing unit (2) is used to convert the acoustic signal into a sound wave digital signal, use a fast independent component analysis technique, and use a complex network analysis idea to construct a new objective function to solve a separation matrix, denoise the sound wave digital signal, and use a wavelet packet decomposition algorithm to extract a feature vector; A WLAN transmission unit (3), wherein the WLAN transmission unit (3) uses a trusted WLAN to transmit the feature vector to an analysis and diagnosis unit (4) through a data transmission channel for analysis; An analysis and diagnosis unit (4), the analysis and diagnosis unit (4) being used to store voiceprint data and perform fault analysis and diagnosis on feature vectors using a BP neural network algorithm; A centralized control system (5), the centralized control system (5) is used for centralized remote monitoring and management of the operation of the substation.

2. The substation voiceprint recognition system based on trusted WLAN + centralized control system according to claim 1 is characterized by: The voiceprint sensing unit (1) comprises a microphone and a preamplifier; Among them, the microphone is used to detect the sound wave signal generated by the equipment during operation; The preamplifier is used to amplify the sound wave signal captured by the microphone and provide it to the signal processing unit (2).

3. The substation voiceprint recognition system based on trusted WLAN + centralized control system according to claim 1 is characterized by: The signal processing unit (2) comprises a data acquisition card, a noise separation module and a frequency feature extraction module; Wherein, the data acquisition card is used to convert the sound wave signal into a sound wave digital signal; The noise separation module uses fast independent component analysis technology and complex network analysis ideas to convert the sound wave digital signal into The effective sound signal is separated from the noise signal to obtain the separated sound wave digital signal ; The frequency feature extraction module uses a wavelet packet decomposition algorithm to extract the sound wave digital signal frequency characteristics.

4. The substation voiceprint recognition system based on trusted WLAN + centralized control system according to claim 3 is characterized by: The noise separation module is based on the fast independent component analysis technology, combined with the use of complex network analysis ideas, to separate the effective sound signal of the sound wave digital signal from the noise signal, and obtain the separated sound wave digital signal , the specific steps are as follows: S2.1.

1. Maximizing Negative Entropy As the objective function: ； in, is the number of independent components; is a constant coefficient; is the expected value; is a non-quadratic optimization function; It is an independent signal source; is a standard Gaussian distribution variable; is the index of the independent component; S2.1.

2. Select the logarithmic hyperbolic cosine function And the parameters As an optimization function: ； in, is the logarithmic hyperbolic cosine function; S2.1.

3. Use complex network analysis to process frequency feature correlation and construct frequency segment correlation matrix , use the Louvain algorithm to cluster the frequency segments into different association communities according to their association ; S2.1.

4. Based on the community division result matrix And the objective function Construct a new objective function: ； in, For the Communities No. frequency bands; For the Communities No. frequency bands; For frequency band and The strength of the association between is the separation matrix; is the total number of communities; S2.1.

5. New objective function by gradient descent method Optimize and solve the separation matrix , the specific steps are as follows: Initialize the separation matrix to the identity matrix , represents the separation matrix at the zeroth iteration; For the new objective function Iterate until the new objective function Converge and get the final separation matrix : ； in, Indicates The separation matrix of the iteration; Indicates The separation matrix of the iteration; is the learning rate; The new objective function exist The gradient at S2.1.

6. According to the separation matrix , calculate the separated sound wave digital signal : ； in, is a sound wave digital signal; It is the separated sound wave digital signal.

5. The substation voiceprint recognition system based on trusted WLAN+centralized control system according to claim 4 is characterized by: In S2.1.3, the frequency feature correlation processing is performed using the complex network analysis idea to construct the frequency segment correlation matrix , use the Louvain algorithm to cluster the frequency segments into different association communities according to their association , as follows: The incidence matrix Each element of Defined as Communities Frequency band and frequency band The strength of the association between: ； in, For the Communities Frequency band and frequency band The strength of the association between For frequency band and frequency band The frequency difference between is the total number of frequency bands; For the Communities No. frequency band, ; is the sum of the absolute values ​​of frequencies in all frequency bands; For frequency band The signal energy of For frequency band The signal energy of Use Louvain's algorithm to calculate the incidence matrix Clustering is performed to obtain groups of frequency bands with strong correlation: Define modularity as : ； in, For elements half the strength of the association; For frequency band degree; For frequency band degree; Kronecker delta function, used to determine the frequency band and frequency band Whether they belong to the same community; Repeatedly adjust the modularity Optimize and get the community division result matrix , where the community partition result matrix Elements Indicates frequency band Belong to the community : 。 6. The substation voiceprint recognition system based on trusted WLAN + centralized control system according to claim 5 is characterized by: The frequency feature extraction module uses a wavelet packet decomposition algorithm to extract the sound wave digital signal The frequency characteristics of the , the specific steps are as follows: S2.1.

7. Sound wave digital signal conduct Layer wavelet packet decomposition: ； ； in, For the Low frequency coefficients after layer wavelet packet decomposition; For the High frequency coefficients after layer wavelet packet decomposition; For the The low-frequency components of the layer; is the filter coefficient of the low-pass filter; is the filter coefficient of the high-pass filter; is the time variable; is the filter order; is the number of wavelet packet decomposition layers; S2.1.

8. Reconstruct the decomposed coefficients to obtain the low-frequency and high-frequency reconstructed signals of the signal: ； ； in, For the Low frequency components during layer reconstruction; For the High frequency components during layer reconstruction; is the coefficient of the low-pass filter used for low-frequency signal reconstruction in the wavelet packet reconstruction process; is the coefficient of the high-pass filter used for high-frequency signal reconstruction in the wavelet packet reconstruction process; S2.1.

9. Calculation Frequency band signal Energy , … : ； in, For the Tier Signals in frequency bands; For the Tier The discrete point amplitude of each frequency band; For frequency band The signal energy of S2.1.

10. Normalize the energy of each frequency band to construct a feature vector : ； in, is the normalized energy eigenvector; is the total number of frequency bands; is the current frequency band index.

7. The substation voiceprint recognition system based on trusted WLAN + centralized control system according to claim 1 is characterized by: The WLAN transmission unit (3) comprises a data transmission channel, which is used to transmit the feature vector to the analysis and diagnosis unit (4) through the data transmission channel. The specific steps are as follows: S3.

1. The feature vector Packed into binary data stream; S3.2, using the encryption mechanism of the trusted WLAN to encrypt the binary data stream; S3.

3. Transmit the binary data stream to the analysis and diagnosis unit (4) via the trusted WLAN communication protocol.

8. The substation voiceprint recognition system based on trusted WLAN + centralized control system according to claim 1 is characterized by: The analysis and diagnosis unit (4) comprises a station-side database and a fault analysis and diagnosis module; Wherein, the station-side database is used to store equipment operation data, historical fault data, and feature vector information; The fault analysis and diagnosis module uses a BP neural network algorithm to perform fault analysis and diagnosis on the feature vector.

9. The substation voiceprint recognition system based on trusted WLAN + centralized control system according to claim 8 is characterized by: The fault analysis and diagnosis module uses the BP neural network algorithm to perform fault analysis and diagnosis on the feature vector. The specific steps are as follows: S4.

1. Initialize the weights between the input layer and the hidden layer , and the weights between the hidden layer and the output layer , set the learning factor to ; S4.

2. Determine the number of hidden layer nodes based on empirical formula : ； in, is the number of hidden layer nodes, is the number of input layer nodes, is the number of nodes in the output layer; S4.3, the input layer selects a linear activation function, then the node Output for: ； in, For Node Input; The hidden layer selects the logarithmic S-type function as the activation function, then the output for: ； ； in, is a hidden layer node Total input of is the hidden layer node The output value of is the Sigmoid activation function; The input layer node The output value of is the hidden layer node The activation value of The output layer selects the S-type function, then the node Output value for: ； in, For Node The output value of For Node The activation value of is the S-type activation function; For Node Total input of S4.

4. The feature vector The input layer is passed to the hidden layer, the total input and output of each hidden layer node are calculated, and the output of the hidden layer is used as input and passed to the output layer to calculate the output of each node in the output layer. ; S4.

5. Calculate the output layer nodes Output value With the expected output The total error : ； in, is the total error; is the expected output; For Node The output value of S4.

6. Calculate the output layer error : ； in, The total error For actual output The partial derivative of The actual output For total input The partial derivative of is the derivative of the output layer activation function; Calculate the hidden layer error : ； in, is the derivative of the hidden layer activation function; is the accumulated error propagated back from the output layer; S4.

7. Output layer error according to error and the hidden layer error , update the weights along the negative gradient direction of the error and weights : ； ； in, is the weight adjustment between the output layer and the hidden layer; is the weight adjustment between the hidden layer and the input layer; S4.

8. Update the weights from the input layer to the hidden layer and from the hidden layer to the output layer: ； ； in, From the hidden layer to the output layer The weight at the iteration; From the hidden layer to the output layer The weight at the iteration; From the input layer to the hidden layer The weight at the iteration; From the input layer to the hidden layer The weight at the iteration; S4.

9. Finally, we get the nodes of the output layer The identification result of the corresponding fault.

10. The substation voiceprint recognition system based on trusted WLAN + centralized control system according to claim 1 is characterized by: The centralized control system (5) is used to perform centralized remote monitoring and management of the substation, as follows: S5.

1. The centralized control system remotely monitors the operating status of multiple substations and centralizes the equipment information of each site into a control center for management; S5.

2. The centralized control system integrates fault monitoring and alarm functions and works in conjunction with the voiceprint recognition system to perform abnormal analysis by detecting the sounds emitted during the operation of the equipment and to diagnose equipment faults in real time in combination with the analysis and diagnosis unit (4).

Images