500kV GIS multi-mode working condition abnormity comprehensive monitoring system and method based on voiceprint recognition

Abstract

The invention provides a 500kV GIS multi-mode working condition abnormity comprehensive monitoring system and method based on voiceprint recognition. A data acquisition module is used for collecting multi-mode original data from various sensors and devices; the network transmission module is used for transmitting the data to the voiceprint scoring module and the multi-modal analysis module through a network; the voiceprint scoring module is used for extracting sound features from the multi-modal original data, comparing the sound features with a preset voiceprint model, and giving a similarity or anomaly score; the multi-modal analysis module is used for comprehensively analyzing data from different modals, fusing various data and performing data analysis and pattern recognition by applying a deep learning algorithm; the fault judgment module performs fault detection and diagnosis by applying a preset rule or model based on an analysis result of the multi-modal analysis module, and judges whether the system or the equipment has a fault or not; and the result output module is used for displaying the final analysis and judgment result in the form of a chart, a report or an alarm.

Description

Technical Field

[0001] The present invention belongs to the technical field of safe operation of power systems, and specifically relates to a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring system and method based on voiceprint recognition. Background Art

[0002] GIS, the full name of which is Gas Insulated Switchgear, is a complete set of power distribution equipment that combines circuit breakers, disconnectors, grounding switches, transformers, lightning arresters, busbars, connectors and outgoing line terminals in a metal shell. As its core component, the GIS circuit breaker system plays an important role in automatically controlling and protecting circuits.

[0003] The circuit breaker is the core component of the GIS system, which is used to automatically cut off the circuit when the current is overloaded or short-circuited to prevent damage to electrical equipment and lines. The circuit breaker is usually composed of an operating mechanism (101), a contact system (102) and an arc extinguishing device (103). The isolating switch (104) is used to isolate the part that needs to be repaired from the live part when the electrical equipment is repaired to ensure the safety of the repair personnel. The grounding switch (106) is used to connect the equipment to the ground to ensure the safety of the equipment and personnel. The transformer includes a current transformer (123) and a voltage transformer (122), which are used to measure and monitor the current and voltage in the circuit.

[0004] Common GIS circuit breaker failures include GIS internal mechanism failure, GIS external interlocking mechanism failure, internal discharge, sulfur hexafluoride leakage, excessive moisture, hydraulic mechanism oil leakage or frequent pressure failure, connecting rod shedding leading to three-phase inconsistency, poor grounding leading to continuous induced discharge, etc. Power plants using GIS circuit breakers are often the backbone power plants of the power system. Once such failures occur, power outages for maintenance are required at the least, reducing power supply reliability; in severe cases, electric closing failures and inadequate opening after tripping may occur, expanding the scope of power outage accidents in the power system.

[0005] Traditional abnormal monitoring methods often detect faults directly, such as the electrical measurement method: using the electrostatic capacitance of the GIS live body and the cylinder body for measurement, or using the electrostatic capacitance of the insulator that insulates the GIS live part from the cylinder body for partial discharge monitoring; ultrasonic detection method: when partial discharge occurs inside the electrical equipment, ultrasonic waves are generated at the discharge point and propagate in all directions until they reach the surface of the electrical equipment container. An ultrasonic sensor is installed on the outer wall of the equipment to detect this electrical signal to determine whether partial discharge has occurred inside the equipment; mechanical vibration method: partial discharge of gas will generate certain vibration waves, which will cause slight vibrations on the metal casing after transmission. It is possible to detect this signal using a highly sensitive acceleration sensor.

[0006] However, existing technologies often have various defects and have not formed an overall multi-modal system to achieve comprehensive judgment, resulting in only partial fault identification, and the fault identification rate and accuracy are very low. For example, most electrical measurement methods cannot achieve online monitoring during circuit breaker operation; ultrasonic detection methods can only identify internal faults; mechanical vibrations will be affected by the operation of the equipment operating mechanism and the noise during operation. There are also some external faults such as connecting rod detachment and abnormal shell temperature that cannot be monitored online in real time, requiring operation and maintenance personnel to actively patrol and observe. Summary of the invention

[0007] In order to solve the current technical problems, the main purpose of the present invention is to provide a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring system and method based on voiceprint recognition. This method uses voiceprint recognition technology to realize real-time online monitoring of all internal faults of GIS circuit breakers, solving the problem that the existing technology cannot guarantee the accuracy of online monitoring; by installing cameras with thermal infrared temperature measurement functions, vibration sensors, SF6 gas monitoring devices, GIS equipment location node relays, current transformers, and voltage transformers in suitable areas, and integrating multi-modal data into the same system, the current situation that the existing technology cannot be comprehensively analyzed is solved; by deep learning of each data parameter and fault judgment process logic design, 500kV GIS multi-modal operating condition abnormality comprehensive monitoring is realized, effectively improving the recognition rate and accuracy of GIS circuit breaker equipment operating condition abnormality or fault judgment.

[0008] In order to realize the above technical features, the purpose of the present invention is achieved as follows: A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring system based on voiceprint recognition, including: a data acquisition module, a network transmission module, a voiceprint scoring module, a multi-modal analysis module, a fault judgment module and a result output module;

[0009] The data acquisition module is used to collect multimodal raw data from various sensors and devices;

[0010] The network transmission module is used to transmit the data collected by the data acquisition module to the voiceprint scoring module and the multimodal analysis module through the network;

[0011] The voiceprint scoring module is used to extract voice features from the multimodal raw data, compare them with the preset voiceprint model, and give a similarity or abnormality score;

[0012] The multimodal analysis module is used to comprehensively analyze data from different modalities, fuse multiple data, and use deep learning algorithms to perform data analysis and pattern recognition to obtain analysis results;

[0013] The fault judgment module uses preset rules or models to perform fault detection and diagnosis based on the analysis results of the multimodal analysis module to determine whether there is a fault in the system or equipment, as well as the type and severity of the fault;

[0014] The result output module is used to display the final analysis and judgment results in the form of charts, reports or alarms.

[0015] A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition, the monitoring method is implemented by the monitoring system, including:

[0016] S1, Collection of multi-modal raw data of GIS circuit breaker:

[0017] The voiceprint collector, vibration sensor, SF6 gas leakage detector, thermal imager and position switch are installed at appropriate positions of the GIS circuit breaker to form a data acquisition module. The multi-modal raw data of different modes of the GIS circuit breaker during operation are collected through the data acquisition module.

[0018] S2, Transmission of multi-modal raw data of GIS circuit breaker:

[0019] The multimodal raw data collected in S1 is transmitted to the voiceprint scoring module and the multimodal analysis module in real time through the network transmission module to realize remote data transmission;

[0020] S3, extraction, analysis and scoring of voiceprint features:

[0021] Extract voice features from the voiceprint data of the multimodal raw data, compare the voice features with the preset voiceprint model through the voiceprint scoring module, give a similarity or anomaly score, and then complete the analysis and scoring of the voiceprint data for subsequent device status or fault judgment;

[0022] S4, analysis and processing of multimodal raw data:

[0023] The result of the voiceprint scoring module in S3 is transmitted to the multimodal analysis module as part of the input data for the analysis of the multimodal analysis module, and the remaining multimodal raw data obtained in S1 is comprehensively analyzed and processed. At the same time, the multimodal raw data is integrated, and the deep learning algorithm is used for data analysis and pattern recognition to obtain a comprehensive analysis result;

[0024] S5, fault judgment:

[0025] Based on the analysis results in S4, the fault judgment module uses preset rules or models to perform fault detection and diagnosis to determine whether the system or equipment has a fault, as well as the type and severity of the fault;

[0026] S6, output of results:

[0027] The analysis and judgment results finally obtained in S4 are presented to the user or system administrator in an appropriate form with the help of the result output module.

[0028] Preferably, the specific steps of S1 are:

[0029] S1.1, select multiple voiceprint collector points at appropriate locations of the GIS circuit breaker, and arrange and install voiceprint collectors at the voiceprint collector points; select multiple vibration sensor points at appropriate locations of the GIS circuit breaker, and arrange and install vibration sensors at the vibration sensor points; arrange and install SF6 gas leakage detectors directly below the GIS circuit breaker; arrange and install thermal imagers on the sides of the GIS circuit breaker; install multiple position switches on the GIS circuit breaker;

[0030] S1.2, various types of raw data during the operation of GIS circuit breaker are collected through voiceprint collector, vibration sensor, SF6 gas leakage detector and thermal imager. Different types of raw data together constitute multimodal raw data.

[0031] Preferably, there are 15 voiceprint collector points in S1.1, and there are also 15 corresponding voiceprint collectors, which respectively correspond to the A / B / C three-phase connection of the circuit breaker oil pump motor, the A / B / C three-phase connection of the circuit breaker body shell, the knife switch motor, and the ground knife motor installed in the GIS circuit breaker, and the voiceprint data is collected through the voiceprint collector.

[0032] Preferably, there are 3 vibration sensor points in S1.1, and there are 3 corresponding vibration sensors, which are respectively installed at the ground bases of three circuit breaker brackets of the GIS circuit breaker, and vibration data is collected by the vibration sensors.

[0033] Preferably, there are three position switches in S1.1, which are respectively installed at the front end of the circuit breaker and the grounding knife operating mechanism of the knife switch, and the position data is collected through the position switches.

[0034] Preferably, S2 specifically includes:

[0035] The voiceprint collector collects the voiceprint data of the corresponding position of the GIS circuit breaker and transmits it to the server through the edge computing network, the first switch, and the core switch;

[0036] The thermal imaging data is collected by the thermal imager and transmitted to the server;

[0037] The voltage transformer and current transformer collect voltage and current data respectively, and transmit them to the server through the transmitter, the local control host, the second switch and the core switch;

[0038] The vibration sensor, SF6 gas leakage detector and position switch collect corresponding data respectively, and transmit them to the server through the local control host, the second switch and the core switch;

[0039] The server is a computer system used to process data, store information and run applications;

[0040] The server is connected to the monitoring platform via network isolation;

[0041] The network isolation is used to isolate different parts of the network to reduce security risks;

[0042] The monitoring platform is used to monitor and control the entire system.

[0043] Preferably, the specific steps of performing voiceprint feature extraction, analysis and scoring based on the voiceprint scoring module in S3 are:

[0044] S3.1, based on MFCC feature extraction:

[0045] S3.1.1, Pre-emphasis:

[0046] Use a high-pass filter to filter the signal. The form of the high-pass filter is:

[0047] y[n]=x[n]-αx[n-1];

[0048] Where, in this high-pass filter, α is the pre-emphasis coefficient, whose value is between 0.9 and 1.0; x[n] is the original signal, which represents the signal value at time n; y[n] is the signal after pre-emphasis processing;

[0049] S3.1.2, Framing:

[0050] The pre-emphasized signal is divided into short time frames, the length of each frame is set to 8-40ms, there is overlap between frames, and the overlapped part is at least half of the frame length; at the same time, each frame is windowed;

[0051] S3.1.3, Fast Fourier Transform:

[0052] Perform a fast Fourier transform on each frame of signal, which converts the time domain signal x[n] into a frequency domain signal X(k);

[0053] The mathematical model of fast Fourier transform is expressed as:

[0054]

[0055] Where X(k) is the complex value of the kth frequency point in the frequency domain, representing the spectrum of the signal; x(n) is the nth sample value in the time domain signal sequence; N is the length of the signal sequence; k is the index in the frequency domain, ranging from 0 to N-1; is a complex exponential function used to perform spectrum analysis of signals; j is an imaginary unit, full j 2 = -1;

[0056] S3.1.4, Mel coefficient filtering:

[0057] A group of bandpass filters is set at the Mel frequency, and the number of bandpass filters is set to 20-40. The spectrum obtained by fast Fourier transform is passed through this group of bandpass filters to obtain the output energy of each bandpass filter;

[0058] Take the logarithm of the output of each bandpass filter and then perform discrete cosine transform to obtain the MFCC coefficients;

[0059] in,

[0060] Mel(f)=2585×log(1+f / 700);

[0061] Where Mel(f) is the Mel frequency; f is the linear frequency;

[0062] S3.1.5, Endpoint Detection:

[0063] Endpoint detection is performed based on short-time energy and short-time zero-crossing rate; energy and zero-crossing rate thresholds are set to determine the start and end points of the voiceprint signal; invalid noise and silent data segments are discarded;

[0064] S3.2, obtain i- vec to r :

[0065] S3.2.1, data preparation, collect device audio data from different devices in multiple different scenarios to ensure that GMM-UBM can capture the common features of devices in all scenarios, and perform MFCC-based feature extraction on them in S3.1;

[0066] S3.2.2, train the GMM-UBM model, use the extracted feature vectors to train the GMM-UBM, this process uses the expectation maximization algorithm, the GMM-UBM consists of multiple Gaussian components, each component has a weight, mean and covariance matrix;

[0067] S3.2.2.1, expectation step, for each feature vector in the training set, calculate the posterior probability that it belongs to each Gaussian component:

[0068]

[0069] Where Q function is the expected log-likelihood of the model parameter θ; θ is the model parameter, which is the variable to be estimated or optimized; θ (t) is the estimated value of θ at the tth iteration; X is the observed data set, each feature vector in the training set; Z is the latent variable or missing data; At a given θ (t) The expectation of the posterior distribution of the latent variable Z under the condition of and X; p(X, Z|θ) is the probability of observing data X and latent data Z given the model parameters θ;

[0070] S3.2.2.2, the maximization step, uses the posterior probabilities calculated in the expectation step to re-estimate the parameters of the GMM-UBM: weights, mean, and covariance matrix; this is done by maximizing the expectation of the log-likelihood function of the data;

[0071] S3.2.2.3, iterative step, repeat the expectation step and the maximization step until the maximum number of iterations is reached;

[0072] S3.2.3, model initialization, use the trained GMM-UBM as the initial model;

[0073] S3.2.4, MAP adaptation, for each Gaussian component in the GMM-UBM, calculate the similarity between the target device audio data and the Gaussian component, use MAP estimation to calculate new Gaussian parameters, and MAP adaptation is achieved by linearly interpolating the mean vector of the GMM-UBM;

[0074]

[0075] In the formula, is the new mean vector; α is a value between 0 and 1, which is used to control the weight between GMM-UBM and the target device audio data; is the mean vector of GMM-UBM; is the mean vector of the target device audio;

[0076] S3.2.5, extracting Supervector, using the mean vector of GMM-UBM obtained after MAP adaptation, concatenates them into a Supervector, which contains the feature information of all frames of the target device audio;

[0077] S3.2.6, calculate the global difference space matrix T, the i-vector model is expressed as:

[0078] m=M+Tw+∈;

[0079] Where m is the Supervector of the target device audio; M is the global mean Supervector; T is the global difference space matrix; w is the low-dimensional i-vector, representing the target device; ∈: residual term;

[0080] The estimation of the T matrix is ​​done using the expectation-maximization EM algorithm;

[0081] S3.2.6, extract i-vector, use the T matrix obtained in the training phase, and project the Supervector into a low-dimensional subspace by multiplying the Supervector with the T matrix;

[0082] S3.3, PLDA discriminant analysis:

[0083] In PLDA, we calculate whether two audio tracks are represented by the feature h in the device space. i Generated by h i The generated likelihood is used to measure the similarity of the two audios, and the likelihood ratio is used to calculate the score:

[0084]

[0085] In the formula, η1 and η2 are the i-vector vectors of two audios respectively. The assumption that these two audios come from the same space is The assumptions from different spaces are in is the likelihood function that the two audios come from the same space; where η1 and η2 are likelihood functions of different sub-spaces respectively, and score is the score of voiceprint data. By calculating the score, we can measure the similarity of two audios. The higher the value, the higher the score, and the greater the possibility that the two audios belong to the same device; otherwise, the smaller the possibility.

[0086] Preferably, the analysis and processing of the multimodal raw data in S4 specifically includes:

[0087] Vibration data collection and processing:

[0088] The vibration sensor converts the vibration signal into an electrical signal, which is then transmitted to the signal amplifier for amplification. The amplified analog signal is sent to the A / D converter, which converts the analog signal into a digital signal. The digital signal processor receives the digital signal from the A / D converter, and processes the data through filtering, Fourier transform, and wavelet transform algorithms to extract the vibration frequency, amplitude, and phase.

[0089] SF6 gas leakage data collection and processing:

[0090] When the SF6 gas leak detector is exposed to an environment containing SF6 gas, SF6 gas will enter the sensor through the gas permeable membrane of the sensor and react chemically with the specific substance on the sensing electrode. On the sensing electrode, SF6 gas will undergo an oxidation-reduction reaction with the material on the electrode. During the process, electron transfer will occur, thus forming a current. The magnitude of the current is proportional to the concentration of SF6 gas. The current is captured by the circuit inside the sensor and converted into a voltage signal. The magnitude of the voltage signal represents the concentration of SF6 gas. The voltage signal is amplified by an amplifier for subsequent processing and recording.

[0091] Position switch signal acquisition and processing:

[0092] Use the travel switch to detect the position of the circuit breaker or the knife-isolating operating mechanism and convert it into an electrical signal. The electrical signal is input into the closing relay and the tripping relay. When the circuit breaker is in the closing position, the closing relay will be energized, and when the circuit breaker is in the opening position, the tripping relay will be energized.

[0093] Current / voltage transformer data acquisition and processing:

[0094] The voltage and current signals are collected through the voltage transformer and the current transformer respectively, and the digital signals after amplification, filtering and analog-to-digital conversion are subjected to various digital signal processing operations;

[0095] Multimodal Scoring:

[0096] Collect data of multiple modes related to system or equipment performance, select a quadratic polynomial regression model to fit the degradation process, and use historical data to train the model to capture the trend and pattern of performance degradation by setting a scoring range of 0 to 100, where 0 represents complete failure and 100 represents a brand new state;

[0097] in:

[0098] Voiceprint sensor data: X1, X2, ..., X 15 ;

[0099] Vibration sensor data: X 16 , X 17 , X 18 ;

[0100] SF6 gas density data: X 19 ;

[0101] Switch position data: X 20 , X 21 , X 22 ;

[0102] Voltage data: X 23 ;

[0103] Current data: X 24 ;

[0104] The quadratic polynomial regression model has the following form:

[0105]

[0106] In the formula, Y is the multimodal score, β0 is the intercept term, and β i is the coefficient of the first-order term, β ij is the coefficient of the quadratic interaction term, and ∈ is the error term.

[0107] Preferably, the fault judgment in S5 specifically includes: internal fault judgment of the oil pump motor, SF6 gas leakage fault judgment, operating mechanism connecting rod fault judgment and casing poor grounding discharge fault judgment.

[0108] Preferably, the specific process of judging the internal fault of the oil pump motor is as follows:

[0109] S5.1.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process;

[0110] S5.1.2, vibration sensor data judgment, if the data is normal, other faults are output. If the vibration sensor installed at the ground base of the GIS circuit breaker bracket, that is, close to the oil pump motor, exceeds the limit abnormality, enter S5.1.3;

[0111] S5.1.3, judge the voiceprint score value. If the score value is normal, other faults are output. If the score value of the A / B / C voiceprint sensor of the A / B / C three-phase phase of the circuit breaker oil pump motor installed on the GIS circuit breaker exceeds the limit, enter S5.1.4;

[0112] S5.1.4, according to whether the soundprint sensor scores of the A / B / C three-phase parts are abnormal, select the A / B / C phase, select the soundprint data of the fault phase, and enter S5.1.5;

[0113] S5.1.5, compare the current fault voiceprint data with the standard in the fault library. If they do not match, mark the fault library according to the specific fault. If the match is an internal fault of the oil pump motor in the fault library, enter S5.1.6;

[0114] S5.1.6, alarm output oil pump motor internal fault;

[0115] S5.1.7, if an internal fault of the oil pump motor is detected, perform an open cover inspection and handle it according to the specific fault location;

[0116] S5.1.8: To prevent the switch from closing due to an internal fault, which would reduce the stability of the system, necessary switch closing locks are performed to prohibit the switch from closing.

[0117] Preferably, the specific process of judging the SF6 gas leakage fault is as follows:

[0118] S5.2.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process;

[0119] S5.2.2, when the SF6 gas leakage detector detects an abnormality, the process enters S5.2.3;

[0120] S5.2.3, next, the system will judge the voiceprint score. If the voiceprint score is normal, the process will continue to monitor; if the voiceprint score is abnormal, it will enter S5.2.4;

[0121] S5.2.4, in the case of abnormal voiceprint scoring value, the system will compare the current status with the records in the fault database to determine the specific fault type;

[0122] S5.2.5, after completing the comparison between the fault library and the model library, the system will determine whether the current fault is related to SF6 gas leakage; if so, it will proceed to the next step; if not, it is necessary to take measures to deal with the current fault and mark it as a new fault;

[0123] S5.2.6, in the event of confirmed SF6 gas leakage, the system will initiate the corresponding emergency plan, including shutting down relevant equipment, ventilation, and evacuating personnel to ensure safety.

[0124] Preferably, the specific process of judging the operating mechanism connecting rod fault is as follows:

[0125] S5.3.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process;

[0126] S5.3.2, switch position node judgment, check whether the position switch installed at the end of the 500kV isolation switch or the position switch installed at the end of the 500kV grounding switch has changed position. If it has changed position, go to S5.3.3;

[0127] S5.3.3, vibration sensor parameter judgment, if the parameters are normal, other faults are output. If the vibration sensor installed at the ground base of the GIS circuit breaker bracket, that is, close to the oil pump motor, exceeds the limit abnormality, go to the next step;

[0128] S5.3.4, judge the voiceprint score value. If the score value is normal, output other faults. If the score value of the voiceprint sensor installed at the end of the 500kV isolation switch or the voiceprint sensor installed at the end of the 500kV grounding switch exceeds the limit, go to the next step;

[0129] S5.3.5, in the case of abnormal voiceprint scoring value, the system will compare the current status with the records in the fault database to determine the specific fault type;

[0130] S5.3.6, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the operating mechanism connecting rod fault; if so, it will enter S5.3.7; if not, it is necessary to take measures to deal with the current fault and mark it as a new fault;

[0131] S5.3.7, if an operating mechanism linkage failure is detected, conduct an on-site inspection and treatment;

[0132] S5.3.8: To prevent the switch from closing due to an internal fault, which would reduce the stability of the system, the switch closing lock is implemented to prohibit the switch from closing.

[0133] Preferably, the specific process of judging the discharge fault due to poor grounding of the shell is as follows:

[0134] S5.4.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process;

[0135] S5.4.2, the process checks the position of the knife switch and confirms that the switch position is not in motion; next, the process checks the status of the SF6 leakage meter to ensure that SF6 is not leaking; the process continues to check the status of the voltage transformer, current transformer and vibration sensor to ensure that the system is operating normally; if everything is normal, the process enters S5.4.3;

[0136] S5.4.3, the process will evaluate the voiceprint score. Based on the result of the voiceprint score, the process will determine whether there is an abnormality in the voiceprint sensors installed at the front, middle and tail A / B / C phases of the GIS circuit breaker body. If an abnormality is detected, the process will proceed to the next step;

[0137] S5.5.4, further inspection using a handheld thermal imager, which measures the temperature distribution on the surface of the equipment to identify overheating or other abnormal conditions and records the information into the system;

[0138] S5.5.5, if abnormal, locate the fault point; if normal, determine the phase difference based on the abnormality of the soundprint sensors installed at the front, middle and tail A / B / C phases of the GIS circuit breaker body, that is, determine which phase A / B / C has a problem;

[0139] S5.5.6, the process will compare the current status with the records in the fault library to determine the specific fault type;

[0140] S5.5.7, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the shell grounding fault;

[0141] S5.5.8, if an operating mechanism linkage failure is detected, conduct an on-site inspection and treatment;

[0142] S5.5.9, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the partial discharge fault;

[0143] S5.5.10, If a partial discharge fault is detected, adjust the switch room load and observe whether the operating condition improves.

[0144] The present invention has the following beneficial effects:

[0145] 1. The present invention realizes real-time online monitoring of all internal faults of GIS circuit breakers through voiceprint recognition technology, ensures the accuracy of fault judgment, and realizes online monitoring.

[0146] 2. The present invention solves the problem that the existing technology cannot conduct comprehensive analysis by installing cameras with thermal infrared temperature measurement functions, vibration sensors, SF6 gas monitoring devices, GIS equipment location node relays, current transformers, and voltage transformers in appropriate areas, and integrating multimodal data into the same system.

[0147] 3. The present invention realizes comprehensive monitoring of 500kV GIS multi-modal operating abnormalities through deep learning of various data parameters and logical design of fault judgment process, greatly improving the recognition rate and accuracy of GIS circuit breaker equipment operating abnormalities or fault judgments.

[0148] 4. The present invention can be used to extract features from the acquired voiceprint data by adopting a voiceprint scoring module, and use the MFCC algorithm for weighted dimensionality reduction optimization, and finally achieve the scoring of the voiceprint data, and use the scoring result as one of the input signals of the subsequent multimodal data to improve the accuracy of fault judgment.

[0149] 5. The present invention uses a bone conduction microphone to collect 50Hz-4KHz audio data, which can collect almost all abnormal conditions in the 500kV GIS equipment area. At the same time, it is assisted by temperature, vibration, electrical quantity, gas monitoring and other data. Therefore, this solution can identify all faults in the 500kV GIS equipment area and greatly improve the accuracy of abnormal monitoring of the 500kV GIS equipment operating conditions.

[0150] 6. The present invention uses voiceprint recognition technology and comprehensive judgment and analysis of multimodal data such as temperature, vibration, voltage / current, image, and gas monitoring to greatly improve the accuracy of abnormal operating condition monitoring of 500kV GIS equipment, save human resources, and reduce the time required for fault identification, providing an opportunity for emergency response to power system accidents and greatly improving the stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0151] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0152] Figure 1 It is a flow chart of the system structure and method of the present invention.

[0153] Figure 2 This is an installation layout diagram of different types of sensors in the data acquisition module of the present invention.

[0154] Figure 3 This is a system structure diagram of the network transmission module of the present invention.

[0155] Figure 4 The figure is a specific flow chart of the extraction, analysis and scoring of voiceprint features of the present invention.

[0156] Figure 5 This is the original waveform of a 50ms voiceprint data selected in the embodiment of the present invention.

[0157] Figure 6 The pre-emphasized waveform is obtained after the 50ms audio is pre-emphasized in the embodiment of the present invention.

[0158] Figure 7 It is a waveform diagram of the first frame after framing in an embodiment of the present invention.

[0159] Figure 8 This is a windowed waveform diagram in an embodiment of the present invention.

[0160] Fig. 9 This is a spectrum diagram after FFT transformation in an embodiment of the present invention.

[0161] Fig.10 This is a diagram of the discrete cosine transform output of the Mel filter after discrete cosine transform in an embodiment of the present invention.

[0162] Fig.11 Schematic diagram of the energy distribution after passing through the Mel filter in the embodiment of the present invention.

[0163] Fig.12 This is the score value of a 3-minute audio in the embodiment of the present invention.

[0164] Fig.13 It is a waveform diagram collected by the vibration sensor 1 within 30 minutes in the embodiment of the present invention.

[0165] Fig.14 This is an alarm signal diagram of the SF6 gas leakage detector in an embodiment of the present invention.

[0166] Fig.15 This is a position switch change signal diagram in an embodiment of the present invention.

[0167] Fig.16 2 is a voltage signal diagram in an embodiment of the present invention.

[0168] Fig.17 This is a flow chart of internal fault judgment of the oil pump motor in an embodiment of the present invention.

[0169] Fig.18 This is a flow chart of SF6 gas leakage fault judgment in an embodiment of the present invention.

[0170] Fig.19 The figure is a flow chart of the fault judgment of the operating mechanism connecting rod in the embodiment of the present invention.

[0171] Fig. 20 The figure is a flow chart of the discharge fault judgment caused by poor shell grounding in an embodiment of the present invention.

[0172] Fig.21 This is a rendering of a thermal imager in an embodiment of the present invention. DETAILED DESCRIPTION

[0173] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0174] Embodiment 1:

[0175] like Figure 1 As shown, a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring system based on voiceprint recognition includes: a data acquisition module 1, a network transmission module 2, a voiceprint scoring module 3, a multi-modal analysis module 4, a fault judgment module 5 and a result output module 6;

[0176] Furthermore, the data acquisition module 1 is used to collect multimodal raw data from various sensors and devices; the data acquisition module sends the collected raw data to subsequent processing links through the network transmission module.

[0177] Furthermore, the network transmission module 2 is used to transmit the data collected by the data acquisition module 1 to the voiceprint scoring module 3 and the multimodal analysis module 4 securely and efficiently through the network; the network can be a signal cable, optical fiber or network cable. The network transmission module transmits the processed data to the voiceprint scoring module and the multimodal analysis module at the same time for them to perform further analysis.

[0178] Furthermore, the voiceprint scoring module 3 is used to extract sound features from the multimodal raw data, compare them with the preset voiceprint model, and give a similarity or abnormality score; the result of the voiceprint scoring module 3 will be transmitted to the multimodal analysis module as part of the input data for the multimodal analysis.

[0179] Furthermore, the multimodal analysis module 4 is used to comprehensively analyze data from different modes: sound, vibration, gas, current, and voltage, integrate multiple data, and use deep learning algorithms to perform data analysis and pattern recognition to obtain analysis results; the comprehensive analysis results of the multimodal analysis module will be passed to the fault judgment module for fault detection and diagnosis.

[0180] Furthermore, the fault judgment module 5 uses preset rules or models to perform fault detection and diagnosis based on the analysis results of the multimodal analysis module 4, and judges whether there is a fault in the system or equipment, as well as the type and severity of the fault. The judgment result of the fault judgment module will finally be passed to the result output module and displayed to the user or system administrator in an appropriate form.

[0181] Furthermore, the result output module 6 is used to display the final analysis and judgment results in the form of charts, reports or alarms.

[0182] Embodiment 2:

[0183] This embodiment provides a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition. The monitoring method is implemented by the monitoring system, including:

[0184] S1, Collection of multi-modal raw data of GIS circuit breaker:

[0185] S1.1, select multiple voiceprint collector points at appropriate locations of the GIS circuit breaker, and arrange and install the voiceprint collectors at the voiceprint collector points; select multiple vibration sensor points at appropriate locations of the GIS circuit breaker, and arrange and install vibration sensors at the vibration sensor points; arrange and install an SF6 gas leakage detector 117 directly below the GIS circuit breaker; arrange and install a thermal imager 118 on the side of the GIS circuit breaker; install multiple position switches on the GIS circuit breaker;

[0186] S1.2, various types of raw data during the operation of the GIS circuit breaker are collected through the voiceprint collector, vibration sensor, SF6 gas leakage detector 117, and thermal imager 118. Different types of raw data together constitute multi-modal raw data.

[0187] For further information, see Figure 2 There are 15 voiceprint collector points in S1.1, and there are 15 corresponding voiceprint collectors 107-113, which correspond to the A / B / C three-phase phase of the circuit breaker oil pump motor, the A / B / C three-phase phase of the circuit breaker body shell, the knife switch motor, and the ground knife motor installed in the GIS circuit breaker, and the voiceprint data is collected through the voiceprint collector.

[0188] For further information, see Figure 2There are three vibration sensor points in S1.1, and there are three corresponding vibration sensors 114-116, which are respectively installed at the ground bases of three circuit breaker brackets of the GIS circuit breaker, and vibration data are collected through the vibration sensors.

[0189] For further information, see Figure 2 There are three position switches 119-121 in S1.1, which are respectively installed at the front end of the circuit breaker and the grounding knife operating mechanism of the knife switch, and the position data is collected through the position switches.

[0190] In this embodiment, the voiceprint collector adopts ZKHY-IVOD-001 industrial diagnostic microphone with a frequency response of 20-20kHz (enhanced version supports 5-150kHz); the vibration sensor adopts DSP vibration sensor with a measurement range of 0-50g; the SF6 gas leakage detector model is TIF XP-1A with a sensitivity of up to 0.01ppm, which can detect extremely low concentration SF6 gas leakage; the handheld thermal imager adopts (UNI-T) UTi260A; the position switch adopts LX series limit switch.

[0191] S2, Transmission of multi-modal raw data of GIS circuit breaker:

[0192] The multimodal raw data collected in S1 is transmitted in real time to the voiceprint scoring module 3 and the multimodal analysis module 4 through the network transmission module 2 to realize remote data transmission;

[0193] S2.1, the voiceprint collector collects the voiceprint data of the corresponding position of the GIS circuit breaker and transmits it to the server 201 through the edge computing network 206, the first switch 205, and the core switch 202;

[0194] S2.2, the thermal imaging camera 118 collects thermal imaging data and transmits it to the server 201;

[0195] S2.3, the voltage transformer 210 and the current transformer 211 collect voltage and current data respectively, and transmit them to the server 201 through the transmitter 209, the local control host 208, the second switch 207 and the core switch 202;

[0196] S2.4, the vibration sensor, SF6 gas leakage detector 117 and position switch respectively collect corresponding data, and transmit them to the server 201 through the local control host 208, the second switch 207 and the core switch 202;

[0197] S2.5, the server 201 is a computer system for processing data, storing information and running applications;

[0198] S2.6, the server 201 is connected to the monitoring platform 204 via the network isolation 203;

[0199] S2.7, the network isolation 203 is used to isolate different parts of the network to reduce security risks;

[0200] S2.8, the monitoring platform 204 is used to monitor and control the entire system.

[0201] In this embodiment, see Figure 3 In the network transmission module 2, the server 201 is a computer system used to process data, store information or run applications; the core switch and the switch constitute part of the network, responsible for data forwarding and communication; network isolation is a network protection measure set up to ensure security, used to isolate different parts of the network and reduce security risks; the monitoring platform is used to monitor and control the entire system or a specific part of the software or system; the edge computing gateway and the local control host assume specific roles in the system, such as data processing, control execution, etc.; voiceprint pickups, current transformers, voltage transformers, etc. represent different types of sensors or measuring devices, which are used to collect various types of data.

[0202] S3, extraction, analysis and scoring of voiceprint features:

[0203] Extracting voice features from the voiceprint data of the multimodal raw data, comparing the voice features with the preset voiceprint model through the voiceprint scoring module 3, giving a similarity or abnormality score, and then completing the analysis and scoring of the voiceprint data for subsequent device status or fault judgment;

[0204] In this embodiment, Figure 4 As shown in the figure, the adopted voiceprint recognition algorithm engine performs weighted dimensionality reduction optimization based on MFCC feature vectors, and applies vector quantization algorithm to identify the mechanical noise signal outside the cable line. Take a 50ms original waveform, as shown in Figure 5 shown.

[0205] S3.1, based on MFCC feature extraction:

[0206] S3.1.1, Pre-emphasis:

[0207] Use a high-pass filter to filter the signal. The form of the high-pass filter is:

[0208] y[n]=x[n]-αx[n-1];

[0209] Where, in this high-pass filter, α is the pre-emphasis coefficient, whose value is between 0.9 and 1.0; x[n] is the original signal, which represents the signal value at time n; y[n] is the signal after pre-emphasis processing;

[0210] The purpose of this step is to enhance the high frequency part, because the high frequency part of the speech signal is more easily attenuated during transmission. The attached figure shows the comparison of the spectrum before and after pre-emphasis, showing that the high frequency part has been enhanced.

[0211] After the 50ms audio is pre-emphasized, a pre-emphasized waveform is obtained, such as Figure 6 shown.

[0212] S3.1.2, Framing:

[0213] The pre-emphasized signal is divided into short time frames, and the length of each frame is set to 8-40ms. There is overlap between frames, and the overlapping part is at least half of the frame length. In this example, the pre-emphasized audio data is divided into multiple frames, each frame is 16 milliseconds, and the frame shift is 8 milliseconds. The waveform of the first frame after framing is obtained, as shown in Figure 7 shown.

[0214] In order to make the transition between frames smooth, each frame is usually windowed. Commonly used window functions include Hamming Window, etc. The waveform after windowing is as follows: Figure 8 shown.

[0215] S3.1.3, Fast Fourier Transform:

[0216] Perform a fast Fourier transform (FFT) on each frame of signal, which converts the time domain signal x[n] into a frequency domain signal X(k);

[0217] The mathematical model of Fast Fourier Transform (FFT) is expressed as:

[0218]

[0219] Where X(k) is the complex value of the kth frequency point in the frequency domain, representing the spectrum of the signal; x(n) is the nth sample value in the time domain signal sequence; N is the length of the signal sequence; k is the index in the frequency domain, ranging from 0 to N-1; is a complex exponential function used to perform spectrum analysis of signals; j is an imaginary unit, full j 2 = -1;

[0220] The spectrum after FFT transformation is as follows Fig. 9 shown.

[0221] S3.1.4, Mel coefficient filtering:

[0222] A set of bandpass filters is set at the Mel frequency, and the number of bandpass filters is set to 20-40, and 24 is selected in this example. The spectrum obtained by fast Fourier transform is passed through this set of bandpass filters to obtain the output energy of each bandpass filter;

[0223] The output of each filter is logarithmically transformed and then subjected to discrete cosine transform (DCT) to obtain the MFCC coefficients.

[0224] The Mel filter DCT output after discrete cosine transform (DCT) is as follows: Fig.10 shown.

[0225] Its relationship with frequency can be approximately expressed by the following formula:

[0226] Mel(f)=2585×log(1+f / 700);

[0227] Where Mel(f) is the Mel frequency; f is the linear frequency;

[0228] After the Mel filter, the energy is divided into steps as follows: Fig.11 shown.

[0229] S3.1.5, Endpoint Detection:

[0230] Endpoint detection is usually performed based on short-time energy and short-time zero-crossing rate; energy and zero-crossing rate thresholds are set to determine the start and end points of the speech signal; for invalid data segments (such as noise, silence, etc.), you can choose to discard or perform other processing.

[0231] The accompanying drawings may show the results of endpoint detection, including valid voice segments and invalid data segments.

[0232] S3.2, obtain i- vec to r :

[0233] S3.2.1, Data preparation, collect a large amount of device audio data, which should come from different devices in multiple different scenarios to ensure that GMM-UBM can capture the common features of devices in all scenarios, and perform MFCC-based feature extraction on them in S3.1;

[0234] S3.2.2, train the GMM-UBM model, use the extracted feature vectors to train the GMM-UBM, this process usually uses the expectation maximization (EM) algorithm. The GMM-UBM consists of multiple Gaussian components, each of which has a weight, mean and covariance matrix;

[0235] S3.2.2.1, E step (expectation step), for each feature vector in the training set, calculate the posterior probability that it belongs to each Gaussian component;

[0236]

[0237] Where Q function is the expected log-likelihood about the model parameter θ; θ is the model parameter, which is the variable we want to estimate or optimize; θ(t) is the estimated value of θ at the tth iteration; X is the observed data set, each feature vector in the training set; Z is the latent variable or missing data; is the expectation of the posterior distribution of the latent variable Z given θ(t) and X; p(X, Z|θ) is the probability of observing data X and latent data Z given the model parameters θ;

[0238] S3.2.2.2, M step (maximization step), use the posterior probability calculated in step E to re-estimate the parameters of GMM-UBM (weights, mean and covariance matrix). This is usually done by maximizing the expectation of the log-likelihood function of the data;

[0239] S3.2.2.3, iterative step, repeating step E and step M until the maximum number of iterations is reached;

[0240] S3.2.3, model initialization, use the trained GMM-UBM as the initial model;

[0241] S3.2.4, MAP adaptation, for each Gaussian component in the GMM-UBM, calculate the similarity between the target device audio data and the Gaussian component, and use the MAP estimation to calculate new Gaussian parameters. MAP adaptation is usually implemented by linearly interpolating the mean vector of the GMM-UBM;

[0242]

[0243] In the formula, is the new mean vector; α is a value between 0 and 1, which is used to control the weight between GMM-UBM and the target device audio data; is the mean vector of GMM-UBM; is the mean vector of the target device audio;

[0244] S3.2.5, extracting Supervector, using the mean vector of GMM-UBM obtained after MAP adaptation, concatenates them into a Supervector, which contains the feature information of all frames of the target device audio;

[0245] S3.2.6, calculate the global difference space matrix T, the i-vector model can be expressed as:

[0246] m=M+Tw+∈;

[0247] Where m is the Supervector of the target device audio; M is the global mean Supervector; T is the global difference space matrix; w is the low-dimensional i-vector, representing the target device; ∈: residual term;

[0248] The estimation of the T matrix is ​​usually done using the expectation-maximization EM algorithm;

[0249] S3.2.6, extract i-vector, use the T matrix obtained in the training phase, and project the Supervector into a low-dimensional subspace by multiplying the Supervector with the T matrix;

[0250] S3.3, PLDA discriminant analysis:

[0251] Probabilistic Linear Discriminant Analysis (PLDA) is a channel compensation algorithm, also known as the probabilistic form of the LDA algorithm;

[0252] PLDA usually works based on i-Vector features to reduce the interference of channel information on speaker information and enhance the robustness of the system;

[0253] Model definition:

[0254] In PLDA, it is assumed that the training data consists of audio from multiple devices, and each device has multiple different audio segments.

[0255] x ij =μ+F i +Gw ij +∈ ij ;

[0256] Define the jth audio of the ith device as x ij , whose generative model includes the mean μ of all data, the identity space F, and the position h of the device in the identity space i , error space G, and position w in G space ij and the residual noise term ∈ ij .

[0257] Training objectives:

[0258] The goal of PLDA model training is to input a bunch of data x ij , find the parameters θ = [μ, F, G, ∑] that can best represent the data set.

[0259] Similarity calculation:

[0260] In PLDA, we calculate whether two audio tracks are represented by the feature h in the device space. i Generated by h iThe generated likelihood is used to measure the similarity of the two audios, and the likelihood ratio is used to calculate the score:

[0261]

[0262] In the formula, η1 and η2 are the i-vector vectors of two audios respectively. The assumption that these two audios come from the same space is s , the assumptions from different spaces are d , where p(η1,η2| s ) is the likelihood function that the two audio frequencies come from the same space; p(η1| d ), p(η2| d ) are the likelihood functions of different sub-spaces of η1 and η2 respectively, and score is the score of the voiceprint data. By calculating the score, the similarity of the two audios can be measured. The higher the value, the higher the score, and the greater the possibility that the two audios belong to the same device; otherwise, the smaller the possibility.

[0263] Compare and analyze each 50ms audio data separately, and calculate the score value. Take the score value of a 3min audio as follows Fig.12 shown.

[0264] S4, analysis and processing of multimodal raw data:

[0265] The result of the voiceprint scoring module (3) in S3 is transmitted to the multimodal analysis module (4) as part of the input data for the analysis of the multimodal analysis module (4), and the remaining multimodal raw data obtained in S1 is comprehensively analyzed and processed, and the multimodal raw data is integrated, and the deep learning algorithm is used to perform data analysis and pattern recognition to obtain a comprehensive analysis result;

[0266] Vibration data collection and processing:

[0267] The vibration sensor converts the vibration signal into an electrical signal, which is then transmitted to the signal amplifier for amplification. The amplified analog signal is sent to the A / D converter, which converts the analog signal into a digital signal. The digital signal processor receives the digital signal from the A / D converter, and processes the data through filtering, Fourier transform, and wavelet transform algorithms to extract the vibration frequency, amplitude, and phase.

[0268] Filtering is usually used to remove noise and smooth data. Taking a first-order low-pass filter as an example, its algorithm can be expressed as:

[0269] Y n =a·X n +(1-a)·Y n-1 ;

[0270] Where: Yn is the output value of this filtering; X n is the sampling value of this time; Y n-1 is the last filter output value; a is the filter coefficient, which is usually much smaller than 1. n , its discrete Fourier transform (DFT) can be expressed as:

[0271]

[0272] Where: X[k] is the complex value of the kth frequency component in the frequency domain; x[n] is the signal sequence in the time domain; N is the length of the signal sequence; k is the index in the frequency domain, ranging from 0 to N-1; is a complex exponential function used to perform spectrum analysis of the signal; j is an imaginary unit that satisfies j 2 =-1.

[0273] The formula of wavelet transform can be expressed as:

[0274]

[0275] In discrete form, it can be approximated as:

[0276] W f (j, k) = ∑ n f[n]ψ j，k [n];

[0277] in:

[0278] W(a, b) is the wavelet coefficient, which is a function of the scale parameter a and the translation parameter b.

[0279] f(t) is the signal to be analyzed.

[0280] ψ(t) is the basic wavelet function, or called the mother wavelet.

[0281] a is the scale parameter that controls the expansion and contraction of the wavelet function.

[0282] b is the translation parameter that controls the position of the wavelet function on the time axis.

[0283] W f (j, k) represents the wavelet coefficients at scale j and position k.

[0284] f[n] is the discrete signal sequence to be analyzed.

[0285] ψ j，k [n] is the discrete wavelet function, which depends on the scale parameter j and the translation parameter k.

[0286] ∑n means summing all signal samples n.

[0287] Phase and amplitude extraction:

[0288] The phase and amplitude can usually be extracted from the results of Fourier transform or wavelet transform. For the Fourier transform result X[k], its amplitude and phase can be expressed as:

[0289] Amplitude: X[k];

[0290] phase: arg(X[k]);

[0291] Where arg represents the argument of the complex number.

[0292] Since the vibration signal is a continuous signal, considering factors such as data acquisition, processing and storage, the sampling interval is selected as 500ms. Fig.13 As shown, this is the waveform collected by the vibration sensor within 30 minutes.

[0293] SF6 gas leakage data collection and processing:

[0294] The electrochemical sensor generally includes a sensing electrode and a reference electrode. When the SF6 gas leak detector 117 is exposed to an environment containing SF6 gas, the SF6 gas will enter the sensor through the gas permeable membrane of the sensor and react chemically with the specific substance on the sensing electrode.

[0295] Generate current: On the sensing electrode, SF6 gas will undergo an oxidation-reduction reaction with the material on the electrode, and this process will produce electron transfer, thus forming a current. The magnitude of this current is proportional to the concentration of SF6 gas.

[0296] Signal conversion: The generated current is captured by the circuit inside the sensor and converted into a voltage signal. The magnitude of this voltage signal represents the concentration of SF6 gas.

[0297] Signal amplification and processing: Voltage signals are usually very weak, so they need to be amplified by an amplifier for subsequent processing and recording. The amplified signal will pass through an analog-to-digital converter (ADC) to convert the analog signal into a digital signal for processing by digital circuits or microprocessors.

[0298] SF6 gas leakage detector alarm signal Fig.14 shown.

[0299] Position switch signal acquisition and processing:

[0300] Use the travel switch to detect the position of the circuit breaker or the knife-type operating mechanism and convert it into an electrical signal. The signal is input to HWJ (closed position relay) and TWJ (trip position relay). When the circuit breaker is in the closed position, HWJ will be energized, and when the circuit breaker is in the open position, TWJ will be energized.

[0301] Position switch change signal such as Fig.15 shown.

[0302] Current / voltage transformer data acquisition and processing:

[0303] The voltage and current signals are collected by the voltage transformer 210 and the current transformer 211 respectively, and various digital signal processing operations are performed on the digital signals after amplification, filtering and analog-to-digital conversion;

[0304] Current Transformer (CT): The main function of a current transformer is to convert high current into low current while keeping the shape and phase of the current waveform unchanged. In practical applications, the secondary side of the CT is connected to a measurement or protection device to collect current signals.

[0305] Voltage Transformer (VT): The main function of the voltage transformer is to convert high voltage into low voltage for use in measurement and protection devices. The secondary side of the VT is also connected to the corresponding equipment to collect voltage signals.

[0306] After amplification, filtering, and analog-to-digital conversion (ADC), the digital signal can be subjected to various digital signal processing operations, such as filtering, spectrum analysis, waveform recognition, etc. Fig.16 shown.

[0307] Multimodal Scoring:

[0308] Collect data of multiple modes related to system or equipment performance, select a quadratic polynomial regression model to fit the degradation process, and use historical data to train the model to capture the trend and pattern of performance degradation by setting a scoring range of 0 to 100, where 0 represents complete failure and 100 represents a brand new state;

[0309] in:

[0310] Voiceprint sensor data: X1, X2, ..., X 15 ;

[0311] Vibration sensor data: X 16 , X 17 , X 18 ;

[0312] SF6 gas density data: X 19 ;

[0313] Switch position data: X 20 , X 21 , X 22 ;

[0314] Voltage data: X 23 ;

[0315] Current data: X 24 ;

[0316] The quadratic polynomial regression model has the following form:

[0317]

[0318] In the formula, Y is the multimodal score, β0 is the intercept term, and β i is the coefficient of the first-order term, β ij is the coefficient of the quadratic interaction term, and ∈ is the error term.

[0319] In practical applications, historical data (training set) is needed to train the model and find the optimal β value, which is usually achieved through the least squares method or other optimization algorithms.

[0320] S5, fault judgment:

[0321] Based on the analysis results in S4, the fault judgment module 5 uses preset rules or models to perform fault detection and diagnosis to determine whether there is a fault in the system or equipment, as well as the type and severity of the fault;

[0322] The fault judgment in S5 specifically includes: internal fault judgment of the oil pump motor, SF6 gas leakage fault judgment, operating mechanism connecting rod fault judgment and casing grounding fault judgment.

[0323] See also Fig.17 The specific process of judging the internal fault of the oil pump motor is as follows:

[0324] S5.1.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process;

[0325] S5.1.2, vibration sensor data judgment, if the data is normal, other faults are output; if the 114# vibration sensor installed at the ground base of the GIS circuit breaker bracket (near the oil pump motor) is out of limit, enter S5.1.3;

[0326] S5.1.3, judge the voiceprint score value. If the score value is normal, other faults are output. If the score value of the 107A / B / C# voiceprint sensor of the A / B / C phase of the oil pump motor of the GIS circuit breaker is abnormal, enter S5.1.4;

[0327] S5.1.4, according to whether the soundprint sensor scores of the A / B / C three-phase parts are abnormal, select the A / B / C phase, select the soundprint data of the fault phase, and enter S5.1.5;

[0328] S5.1.5, compare the current fault voiceprint data with the standard in the fault library. If they do not match, mark the fault library according to the specific fault. If the match is an internal fault of the oil pump motor in the fault library, enter S5.1.6;

[0329] S5.1.6, alarm output oil pump motor internal fault;

[0330] S5.1.7, if an internal fault of the oil pump motor is detected, perform an open cover inspection and handle it according to the specific fault location;

[0331] S5.1.8: To prevent the switch from closing due to an internal fault, which would reduce the stability of the system, necessary switch closing locks are performed to prohibit the switch from closing.

[0332] See also Fig.18 The specific process of SF6 gas leakage fault judgment is as follows:

[0333] S5.2.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process;

[0334] S5.2.2, when the SF6 gas leakage detector detects an abnormality, the process enters S5.2.3;

[0335] S5.2.3, next, the system will judge the voiceprint score. If the voiceprint score is normal, the process will continue to monitor; if the voiceprint score is abnormal, it will enter S5.2.4;

[0336] S5.2.4, in the case of abnormal voiceprint scoring value, the system will compare the current status with the records in the fault database to determine the specific fault type;

[0337] S5.2.5, after completing the comparison between the fault library and the model library, the system will determine whether the current fault is related to SF6 gas leakage; if so, it will proceed to the next step; if not, it is necessary to take measures to deal with the current fault and mark it as a new fault;

[0338] S5.2.6, in the event of confirmed SF6 gas leakage, the system will initiate the corresponding emergency plan, including shutting down relevant equipment, ventilation, and evacuating personnel to ensure safety.

[0339] See also Fig.19 The specific process of judging the operating mechanism connecting rod fault is as follows:

[0340] S5.3.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process;

[0341] S5.3.2, switch position node judgment, check the 120# position switch installed at the end of the 500kV isolation switch or the 121# position switch installed at the end of the 500kV grounding switch (here directly write whether the position switch installed at that position has changed. If it has changed, enter S5.3.3;

[0342] S5.3.3, vibration sensor parameter judgment, if the parameters are normal, other faults will be output. If the 114# vibration sensor installed on the ground base of the GIS circuit breaker bracket (near the oil pump motor) is out of limit, go to the next step;

[0343] S5.3.4, judge the voiceprint score value. If the score value is normal, other faults will be output. If the score value of the 111# voiceprint sensor installed at the end of the 500kV isolation switch or the 112# voiceprint sensor installed at the end of the 500kV grounding switch exceeds the limit, go to the next step;

[0344] S5.3.5, in the case of abnormal voiceprint scoring value, the system will compare the current status with the records in the fault database to determine the specific fault type;

[0345] S5.3.6, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the operating mechanism connecting rod fault; if so, it will enter S5.3.7; if not, it is necessary to take measures to deal with the current fault and mark it as a new fault;

[0346] S5.3.7, if an operating mechanism linkage failure is detected, conduct an on-site inspection and treatment;

[0347] S5.3.8: To prevent the switch from closing due to an internal fault, which would reduce the stability of the system, the switch closing lock is implemented to prohibit the switch from closing.

[0348] See also Fig. 20 The specific process of judging the discharge fault caused by poor grounding of the shell is as follows:

[0349] S5.4.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process;

[0350] S5.4.2, the process checks the position of the knife switch and confirms that the switch position is not actuated; next, the process checks the status of the SF6 leakage meter to ensure that SF6 is not leaking; the process continues to check the status of the voltage transformer, current transformer and vibration sensor to ensure that the system is operating normally; if everything is normal, the process enters S5.4.3;

[0351] S5.4.3, the process will evaluate the voiceprint score. According to the result of the voiceprint score, the process will determine whether there is an abnormality in the 108A / B / C#, 109A / B / C#, and 110A / B / C# voiceprint sensors installed at the front, middle, and tail A / B / C phases of the GIS circuit breaker body. If an abnormality is detected, the process will proceed to the next step;

[0352] S5.5.4, use a handheld thermal imager for further inspection. The thermal imager measures the temperature distribution on the surface of the equipment to identify overheating or other abnormal conditions and enters the data into the system; the thermal imager effect is as follows Fig.21 The data is entered into the following table as shown.

[0353] Table 1 Temperature data table

[0354]

[0355] S5.5.5, if abnormal, locate the fault point; if normal, determine the phase difference based on the abnormal conditions of the 108A / B / C#, 109A / B / C#, 110A / B / C# voiceprint sensors installed at the front, middle and tail A / B / C phases of the GIS circuit breaker body, that is, determine which phase of A / B / C has a problem;

[0356] S5.5.6, the process will compare the current status with the records in the fault library to determine the specific fault type;

[0357] S5.5.7, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the shell grounding fault;

[0358] S5.5.8, if an operating mechanism linkage failure is detected, conduct an on-site inspection and treatment;

[0359] S5.5.9, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the partial discharge fault;

[0360] S5.5.10, If a partial discharge fault is detected, adjust the switch room load and observe whether the operating condition improves.

[0361] S6, output of results:

[0362] The analysis and judgment results finally obtained in S4 are presented to the user or system administrator in an appropriate form by means of the result output module 6 .

[0363] In summary, the main innovative solutions of the present invention are:

[0364] 1. The process of the present invention covers the real-time monitoring and abnormal identification of various operating parameters in the 500kV GIS equipment area.

[0365] 2. The present invention includes reasonable arrangement of specific points of monitoring equipment such as voiceprint sensors and vibration sensors in the equipment area to achieve comprehensive and effective equipment status monitoring.

[0366] 3. The system of the present invention is used to transmit in real time the signal parameters collected by various sensors in the equipment area to achieve remote monitoring and data analysis.

[0367] 4. The present invention includes specific processes and algorithms for extracting key voiceprint feature values ​​from the signals collected by the voiceprint sensor.

[0368] 5. The present invention uses PLDA technology to perform discriminant analysis on the extracted voiceprint feature values ​​and gives corresponding scoring values.

[0369] 6. The algorithm of the present invention combines voiceprint recognition technology and polynomial regression model to identify the degradation trend of GIS equipment working conditions.

[0370] 7. The present invention uses multimodal data, combined with signals from voiceprint and vibration sensors, to determine the logic of internal faults in the oil pump motor;

[0371] 8. The present invention is based on multimodal data, combined with voiceprint and SF6 gas leakage instrument signals, to determine the logic of SF6 gas leakage fault;

[0372] 9. The present invention is based on multimodal data, combined with the signals of voiceprint, vibration sensor and position node, to judge the logic of abnormal fault of the operating mechanism connecting rod;

[0373] 10. The present invention determines the logic of discharge fault based on multimodal data, combined with the signals of voiceprint, vibration sensor, SF6 gas leakage meter, voltage transformer and current transformer.

Claims

1. A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring system based on voiceprint recognition, characterized in that: include: Data collection module (1), network transmission module (2), voiceprint scoring module (3), multimodal analysis module (4), fault judgment module (5) and result output module (6); The data acquisition module (1) is used to collect multi-modal raw data from various sensors and devices; The network transmission module (2) is used to transmit the data collected by the data acquisition module (1) to the voiceprint scoring module (3) and the multimodal analysis module (4) through the network; The voiceprint scoring module (3) is used to extract voice features from the multimodal raw data, compare them with the preset voiceprint model, and give a similarity or abnormality score; The multimodal analysis module (4) is used to comprehensively analyze data from different modalities, fuse multiple data, and use deep learning algorithms to perform data analysis and pattern recognition to obtain analysis results; The fault judgment module (5) performs fault detection and diagnosis based on the analysis result of the multimodal analysis module (4) using preset rules or models to judge whether there is a fault in the system or equipment, as well as the type and severity of the fault; The result output module (6) is used to display the final analysis and judgment results in the form of charts, reports or alarms.

2. A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition, characterized in that: The monitoring method is implemented by using the monitoring system according to claim 1, comprising: S1, Collection of multi-modal raw data of GIS circuit breaker: A voiceprint collector, a vibration sensor, an SF6 gas leakage detector (117), a thermal imager (118) and a position switch are respectively arranged and installed at appropriate positions of the GIS circuit breaker, together forming a data acquisition module (1); the data acquisition module (1) is used to collect multi-modal raw data of different modes of the GIS circuit breaker during operation; S2, Transmission of multi-modal raw data of GIS circuit breaker: The multimodal raw data collected in S1 is transmitted in real time to the voiceprint scoring module (3) and the multimodal analysis module (4) through the network transmission module (2) to realize remote data transmission; S3, extraction, analysis and scoring of voiceprint features: Extracting voice features from the voiceprint data of the multimodal raw data, comparing the voice features with the preset voiceprint model through the voiceprint scoring module (3), giving a similarity or anomaly score, and then completing the analysis and scoring of the voiceprint data for subsequent device status or fault judgment; S4, analysis and processing of multimodal raw data: The result of the voiceprint scoring module (3) in S3 is transmitted to the multimodal analysis module (4) as part of the input data for the analysis of the multimodal analysis module (4), and the remaining multimodal raw data obtained in S1 is comprehensively analyzed and processed, and the multimodal raw data is integrated, and the deep learning algorithm is used to perform data analysis and pattern recognition to obtain a comprehensive analysis result; S5, fault judgment: Based on the analysis results in S4, a fault judgment module (5) is used to perform fault detection and diagnosis using preset rules or models to determine whether there is a fault in the system or equipment, as well as the type and severity of the fault; S6, output of results: The analysis and judgment results finally obtained in S4 are displayed to the user or system administrator in an appropriate form with the help of the result output module (6).

3. According to claim 2, a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: The specific steps of S1 are: S1.1, selecting a plurality of voiceprint collector points at appropriate locations of the GIS circuit breaker, and arranging and installing the voiceprint collectors at the voiceprint collector points; selecting a plurality of vibration sensor points at appropriate locations of the GIS circuit breaker, and arranging and installing vibration sensors at the vibration sensor points; arranging and installing an SF6 gas leakage detector (117) directly below the GIS circuit breaker; arranging and installing a thermal imaging camera (118) on the side of the GIS circuit breaker; and installing a plurality of position switches on the GIS circuit breaker; S1.2, various types of raw data during the operation of the GIS circuit breaker are collected through a voiceprint collector, a vibration sensor, an SF6 gas leakage detector (117), and a thermal imager (118), and different types of raw data together constitute multi-modal raw data.

4. According to claim 3, a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: There are 15 voiceprint collector points in S1.1, and there are 15 corresponding voiceprint collectors, which respectively correspond to the A / B / C three-phase phase of the oil pump motor, the A / B / C three-phase phase of the circuit breaker body shell, the knife switch motor, and the ground knife motor installed in the GIS circuit breaker, and the voiceprint data is collected through the voiceprint collectors.

5. According to claim 3, a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: There are three vibration sensor points in S1.1, and there are also three corresponding vibration sensors (114-116), which are respectively installed at the ground bases of the three circuit breaker brackets of the GIS circuit breaker, and vibration data is collected through the vibration sensors.

6. According to claim 3, a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: There are three position switches in S1.1, which are respectively installed at the front end of the circuit breaker and the grounding knife operating mechanism of the knife switch, and the position data is collected through the position switches.

7. According to claim 3, a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: The S2 specifically includes: The voiceprint collector collects the voiceprint data of the corresponding position of the GIS circuit breaker and transmits it to the server (201) through the edge computing network (206), the first switch (205), and the core switch (202); The thermal imaging data is collected by a thermal imager (118) and transmitted to a server (201); The voltage transformer (210) and the current transformer (211) respectively collect voltage and current data, and transmit the data to the server (201) through the transmitter (209), the local control host (208), the second switch (207) and the core switch (202); The vibration sensor, the SF6 gas leakage detector (117) and the position switch respectively collect corresponding data, and transmit the data to the server (201) through the local control host (208), the second switch (207) and the core switch (202); The server (201) is a computer system for processing data, storing information and running applications; The server (201) is connected to the monitoring platform (204) via a network isolation (203); The network isolation (203) is used to isolate different parts of the network to reduce security risks; The monitoring platform (204) is used to monitor and control the entire system.

8. According to claim 7, a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: The specific steps of performing voiceprint feature extraction, analysis and scoring based on the voiceprint scoring module (3) in S3 are as follows: S3.1, based on MFCC feature extraction: S3.1.1, Pre-emphasis: Use a high-pass filter to filter the signal. The form of the high-pass filter is: y[n]=x[n]-αx[n-1]; Where, in this high-pass filter, α is the pre-emphasis coefficient, whose value is between 0.9 and 1.0; x[n] is the original signal, which represents the signal value at time n; y[n] is the signal after pre-emphasis processing; S3.1.2, Framing: The pre-emphasized signal is divided into short time frames, the length of each frame is set to 8-40ms, there is overlap between frames, and the overlapped part is at least half of the frame length; at the same time, each frame is windowed; S3.1.3, Fast Fourier Transform: Perform a fast Fourier transform on each frame of signal, which converts the time domain signal x[n] into a frequency domain signal X(k); The mathematical model of fast Fourier transform is expressed as: Where X(k) is the complex value of the kth frequency point in the frequency domain, representing the spectrum of the signal; x(n) is the nth sample value in the time domain signal sequence; N is the length of the signal sequence; k is the index in the frequency domain, ranging from 0 to N-1; is a complex exponential function used to perform spectrum analysis of signals; j is an imaginary unit, full j 2 = -1; S3.1.4, Mel coefficient filtering: A group of bandpass filters is set at the Mel frequency, and the number of bandpass filters is set to 20-40. The spectrum obtained by fast Fourier transform is passed through this group of bandpass filters to obtain the output energy of each bandpass filter; Take the logarithm of the output of each bandpass filter and then perform discrete cosine transform to obtain the MFCC coefficients; in, Mel(f)=2585×log(1+f / 700); Where Mel(f) is the Mel frequency; f is the linear frequency; S3.1.5, Endpoint Detection: Endpoint detection is performed based on short-time energy and short-time zero-crossing rate; energy and zero-crossing rate thresholds are set to determine the start and end points of the voiceprint signal; invalid noise and silent data segments are discarded; S3.2, get i-vector: S3.2.1, data preparation, collect device audio data from different devices in multiple different scenarios to ensure that GMM-UBM can capture the common features of devices in all scenarios, and perform MFCC-based feature extraction on them in S3.1; S3.2.2, train the GMM-UBM model, use the extracted feature vectors to train the GMM-UBM, this process uses the expectation maximization (EM) algorithm, the GMM-UBM consists of multiple Gaussian components, each component has a weight, mean and covariance matrix; S3.2.2.1, expectation step, for each feature vector in the training set, calculate the posterior probability that it belongs to each Gaussian component: Where Q function is the expected log-likelihood of the model parameter θ; θ is the model parameter, which is the variable to be estimated or optimized; θ (t) is the estimated value of θ at the tth iteration; X is the observed data set, each feature vector in the training set; Z is the latent variable or missing data; At a given θ (t) p(X,Z|θ) is the expectation of the posterior distribution of the latent variable Z under the condition of and X; p(X,Z|θ) is the probability of observing data X and latent data Z given the model parameters θ; S3.2.2.2, the maximization step, uses the posterior probabilities calculated in the expectation step to re-estimate the parameters of the GMM-UBM: weights, mean, and covariance matrix; this is done by maximizing the expectation of the log-likelihood function of the data; S3.2.2.3, iterative step, repeat the expectation step and the maximization step until the maximum number of iterations is reached; S3.2.3, model initialization, use the trained GMM-UBM as the initial model; S3.2.4, MAP adaptation, for each Gaussian component in the GMM-UBM, calculate the similarity between the target device audio data and the Gaussian component, use MAP estimation to calculate new Gaussian parameters, and MAP adaptation is achieved by linearly interpolating the mean vector of the GMM-UBM; In the formula, is the new mean vector; α is a value between 0 and 1, which is used to control the weight between GMM-UBM and the target device audio data; is the mean vector of GMM-UBM; is the mean vector of the target device audio; S3.2.5, extracting Supervector, using the mean vector of GMM-UBM obtained after MAP adaptation, concatenates them into a Supervector, which contains the feature information of all frames of the target device audio; S3.2.6, calculate the global difference space matrix T, the i-vector model is expressed as: m=M+Tw+∈; Where m is the Supervector of the target device audio; M is the global mean Supervector; T is the global difference space matrix; w is the low-dimensional i-vector, representing the target device; ∈: residual term; The estimation of the T matrix is ​​done using the expectation-maximization EM algorithm; S3.2.6, extract i-vector, use the T matrix obtained in the training phase, and project the Supervector into a low-dimensional subspace by multiplying the Supervector with the T matrix; S3.3, PLDA discriminant analysis: In PLDA, we calculate whether two audio tracks are represented by the feature h in the device space. i Generated by h i The generated likelihood is used to measure the similarity of two audios, and the likelihood ratio is used to calculate the score: In the formula, η1 and η2 are the i-vector vectors of two audios respectively. The assumption that these two audios come from the same space is The assumptions from different spaces are in is the likelihood function that the two audios come from the same space; are the likelihood functions of different sub-spaces of η1 and η2 respectively, and score is the score of the voiceprint data. By calculating the score, the similarity of the two audios can be measured. The higher the value, the higher the score, and the greater the possibility that the two audios belong to the same device; otherwise, the smaller the possibility.

9. According to claim 8, a 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: The analysis and processing of the multimodal raw data in S4 specifically includes: Vibration data collection and processing: The vibration sensor converts the vibration signal into an electrical signal, which is then transmitted to the signal amplifier for amplification. The amplified analog signal is sent to the A / D converter, which converts the analog signal into a digital signal. The digital signal processor receives the digital signal from the A / D converter, and processes the data through filtering, Fourier transform, and wavelet transform algorithms to extract the vibration frequency, amplitude, and phase. SF6 gas leakage data collection and processing: When the SF6 gas leakage detector (117) is exposed to an environment containing SF6 gas, the SF6 gas will enter the sensor through the gas permeable membrane of the sensor and react chemically with the specific substance on the sensing electrode. On the sensing electrode, the SF6 gas will undergo an oxidation-reduction reaction with the material on the electrode. During the process, electron transfer will occur, thereby forming an electric current. The magnitude of the current is proportional to the concentration of the SF6 gas. The current is captured by the circuit inside the sensor and converted into a voltage signal. The magnitude of the voltage signal represents the concentration of the SF6 gas. The voltage signal is amplified by an amplifier for subsequent processing and recording. Position switch signal acquisition and processing: Use the travel switch to detect the position of the circuit breaker or the knife-isolating operating mechanism and convert it into an electrical signal. The electrical signal is input into the closing relay and the tripping relay. When the circuit breaker is in the closing position, the closing relay will be energized, and when the circuit breaker is in the opening position, the tripping relay will be energized. Current / voltage transformer data acquisition and processing: The voltage and current signals are collected respectively by a voltage transformer (210) and a current transformer (211), and various digital signal processing operations are performed on the digital signals after amplification, filtering and analog-to-digital conversion; Multimodal Scoring: Collect data of multiple modes related to system or equipment performance, select a quadratic polynomial regression model to fit the degradation process, and use historical data to train the model to capture the trend and pattern of performance degradation by setting a scoring range of 0 to 100, where 0 represents complete failure and 100 represents a brand new state; in: Voiceprint sensor data: X1, X2, ..., X 15 ; Vibration sensor data: X 16 , X 17 , X 18 ; SF6 gas density data: X 19 ; Switch position data: X 20 , X 21 , X 22 ; Voltage data: X 23 ; Current data: X 24 ; The quadratic polynomial regression model has the following form: In the formula, Y is the multimodal score, β0 is the intercept term, and β i is the coefficient of the first-order term, β ij is the coefficient of the quadratic interaction term, and ∈ is the error term.

10. A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition according to claim 9, characterized in that: The fault judgment in S5 specifically includes: internal fault judgment of the oil pump motor, SF6 gas leakage fault judgment, operating mechanism connecting rod fault judgment and casing grounding fault judgment.

11. A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition according to claim 10, characterized in that: The specific process of judging the internal fault of the oil pump motor is as follows: S5.1.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process; S5.1.2, vibration sensor data judgment, if the data is normal, other faults are output. If the vibration sensor installed at the ground base of the GIS circuit breaker bracket, that is, close to the oil pump motor, exceeds the limit abnormality, enter S5.1.3; S5.1.3, judge the voiceprint score value. If the score value is normal, other faults are output. If the score value of the A / B / C voiceprint sensor of the A / B / C three-phase phase of the circuit breaker oil pump motor installed on the GIS circuit breaker exceeds the limit, enter S5.1.4; S5.1.4, according to whether the soundprint sensor scores of the A / B / C three-phase parts are abnormal, select the A / B / C phase, select the soundprint data of the fault phase, and enter S5.1.5; S5.1.5, compare the current fault voiceprint data with the standard in the fault library. If they do not match, mark the fault library according to the specific fault. If the match is an internal fault of the oil pump motor in the fault library, enter S5.1.6; S5.1.6, alarm output oil pump motor internal fault; S5.1.7, if an internal fault of the oil pump motor is detected, perform an open cover inspection and handle it according to the specific fault location; S5.1.8: To prevent the switch from closing due to an internal fault, which would reduce the stability of the system, necessary switch closing locks are performed to prohibit the switch from closing.

12. A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition according to claim 10, characterized in that: The specific process of SF6 gas leakage fault judgment is as follows: S5.2.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process; S5.2.2, when the SF6 gas leakage detector detects an abnormality, the process enters S5.2.3; S5.2.3, next, the system will judge the voiceprint score. If the voiceprint score is normal, the process will continue to monitor; if the voiceprint score is abnormal, it will enter S5.2.4; S5.2.4, in the case of abnormal voiceprint scoring value, the system will compare the current status with the records in the fault database to determine the specific fault type; S5.2.5, after completing the comparison between the fault library and the model library, the system will determine whether the current fault is related to SF6 gas leakage; if so, it will proceed to the next step; if not, it is necessary to take measures to deal with the current fault and mark it as a new fault; S5.2.6, in the event of confirmed SF6 gas leakage, the system will initiate the corresponding emergency plan, including shutting down relevant equipment, ventilation, and evacuating personnel to ensure safety.

13. A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition according to claim 10, characterized in that: The specific process of judging the operating mechanism connecting rod fault is as follows: S5.3.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process; S5.3.2, switch position node judgment, check whether the position switch installed at the end of the 500kV isolation switch or the position switch installed at the end of the 500kV grounding switch has changed position. If it has changed position, go to S5.3.3; S5.3.3, vibration sensor parameter judgment, if the parameters are normal, other faults are output. If the vibration sensor installed at the ground base of the GIS circuit breaker bracket, that is, close to the oil pump motor, exceeds the limit abnormality, go to the next step; S5.3.4, judge the voiceprint score value. If the score value is normal, output other faults. If the score value of the voiceprint sensor installed at the end of the 500kV isolation switch or the voiceprint sensor installed at the end of the 500kV grounding switch exceeds the limit, go to the next step; S5.3.5, in the case of abnormal voiceprint scoring value, the system will compare the current status with the records in the fault library to determine the specific fault type; S5.3.6, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the operating mechanism connecting rod fault; if so, it will enter S5.3.7; if not, it is necessary to take measures to deal with the current fault and mark it as a new fault; S5.3.7, if an operating mechanism linkage failure is detected, conduct an on-site inspection and treatment; S5.3.8: To prevent the switch from closing due to an internal fault, which would reduce the stability of the system, the switch closing lock is implemented to prohibit the switch from closing.

14. A 500kV GIS multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition according to claim 10, characterized in that: The specific process of judging the discharge fault of the poor grounding of the shell is as follows: S5.4.1, the multi-modal scoring value exceeds the limit to trigger the fault judgment process; S5.4.2, the process checks the position of the knife switch and confirms that the switch position is not in motion; next, the process checks the status of the SF6 leakage meter to ensure that SF6 is not leaking; the process continues to check the status of the voltage transformer, current transformer and vibration sensor to ensure that the system is operating normally; if everything is normal, the process enters S5.4.3; S5.4.3, the process will evaluate the voiceprint score. Based on the result of the voiceprint score, the process will determine whether there is an abnormality in the voiceprint sensors installed at the front, middle and tail A / B / C phases of the GIS circuit breaker body. If an abnormality is detected, the process will proceed to the next step; S5.5.4, further inspection using a handheld thermal imager, which measures the temperature distribution on the surface of the equipment to identify overheating or other abnormal conditions and records the information into the system; S5.5.5, if abnormal, locate the fault point; if normal, determine the phase difference based on the abnormality of the soundprint sensors installed at the front, middle and tail A / B / C phases of the GIS circuit breaker body, that is, determine which phase A / B / C has a problem; S5.5.6, the process will compare the current status with the records in the fault library to determine the specific fault type; S5.5.7, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the shell grounding fault; S5.5.8, if an operating mechanism linkage failure is detected, conduct an on-site inspection and treatment; S5.5.9, after completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the partial discharge fault; S5.5.10, If a partial discharge fault is detected, adjust the switch room load and observe whether the operating condition improves.

Images