A 20KV electrical brake switch multi-modal operating abnormality comprehensive monitoring system and method based on voiceprint recognition

Abstract

The present invention provides a comprehensive monitoring system and method for multi-modal abnormal working conditions of a 20KV electrical brake switch based on voiceprint recognition. The present invention uses voiceprint recognition technology to realize real-time online monitoring of all internal faults of a 20KV generator brake switch, filling the industry gap of real-time online monitoring of 20KV generator brake switches; by installing voiceprint collectors, temperature sensors, vibration sensors, SF6 gas monitoring devices, position node relays, etc. in appropriate areas, and integrating multi-modal data into the same system, a comprehensive analysis of the working conditions of the 20KV generator brake switch is realized; by deep learning of various data parameters and logical design of fault judgment processes, a comprehensive diagnosis of multi-modal abnormal working conditions of the 20KV generator brake switch is realized, greatly improving the timeliness of abnormal working conditions or fault judgment of the 20KV generator brake switch equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of safe operation of large-scale hydro-generator systems, and specifically provides a 20KV electrical brake switch multi-modal operating condition abnormality comprehensive monitoring system and method based on voiceprint recognition. Background Art

[0002] The 20KV generator brake switchgear is usually composed of two parts: an SF6 generator circuit breaker and a three-phase short-circuit (107). The 20KV generator circuit breaker is a horizontally arranged structure and is operated by a three-phase mechanical linkage (102) and a hydraulic spring mechanism (109). Three sets of voltage transformers (103, 104, 105) are installed on the side of the moving contact (generator main circuit phase-separated closed busbar) of each set of brake switchgear for use in protection, speed regulation, excitation and other systems. The electrical connection diagram is as follows: Figure 1 shown.

[0003] After a large hydroelectric generator set is shut down, its speed slows down after decreasing to 50% due to inertia. To minimize the impact of prolonged low speed on the equipment, electrical braking is required for rapid braking. When the generator set speed drops to 50%, the brake switch is engaged. Once the brake switch is engaged, the electrical brake control increases the generator excitation, rapidly increasing the current flowing through the electrical brake device to approximately the rated current of the generator. At this point, the three phases of the stator winding are symmetrically short-circuited, and the rotor is excited, causing a braking current to flow through the stator winding equal to the current at maximum capacity, generating an electric braking torque and achieving a rapid shutdown.

[0004] Common faults with 20KV generator brake switches include SF6 gas leakage and pressure drop, loose or deformed three-way connecting rods, frequent oil pump startups, partial discharges, and poor contact and burning of voltage transformers. Although 20KV generator brake switches are only put into operation during unit shutdown, the device itself is actually directly electrically connected to the generator output closed busbar and the main transformer. If faults such as loose or deformed three-way connecting rods, partial discharges, poor contact and burning of voltage transformers cause primary equipment grounding or abnormal measurement value misoperation of protective devices, this will directly lead to unit electrical accident shutdown, seriously affecting the reliability of large hydroelectric generators during grid-connected operation, and thus the stability of the power system.

[0005] The current power generation industry has no online monitoring of 20KV generator brake switches. Most large generator sets only upload the position status of the electrical brake switch to the monitoring system. Operators only conduct regular inspections to check whether the equipment is normal, and perform relevant routine inspections during unit maintenance. Other monitoring methods are zero.

[0006] Since the existing technology cannot achieve real-time online monitoring of the 20KV generator brake switch, the operating personnel can only conduct inspections based on their experience. Therefore, before the 20KV generator brake switch is put into operation during the unit shutdown process, it is impossible to accurately judge whether the 20KV generator brake switch is operating normally. The normal operating condition can only be verified by the electrical quantities of the stator and rotor after the switch is put into operation. Once an abnormality occurs before the switch is put into operation, a long-term three-phase imbalance of the stator winding may occur, affecting the reliability of the generator; secondly, during the operation of the hydro-turbine generator set, if the voltage transformer fails due to partial discharge, poor contact of the voltage transformer, etc., it may cause the unit to shut down directly. This type of fault is an internal fault and cannot be directly observed by the operating personnel. Summary of the Invention

[0007] In order to solve the current technical problems, the main purpose of the present invention is to provide a 20KV electrical brake switch 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 the 20KV generator brake switch, filling the industry gap of real-time online monitoring of the 20KV generator brake switch; by installing bone conduction pickups, cameras with thermal infrared temperature measurement functions, vibration sensors, SF6 gas monitoring devices, position node relays, etc. in appropriate areas, and integrating multi-modal data into the same system, a comprehensive analysis of the 20KV generator brake switch operating condition is realized; by deep learning of each data parameter and fault judgment process logic design, a comprehensive diagnosis of the multi-modal operating condition abnormality of the 20KV generator brake switch is realized, which greatly improves the timeliness of the abnormal operating condition or fault judgment of the 20KV generator brake switch equipment.

[0008] In order to achieve the above technical features, the purpose of the present invention is achieved as follows: a 20KV electrical brake switch multi-modal operating condition abnormality comprehensive monitoring system based on voiceprint recognition, including: a parameter acquisition module, a network transmission module, a voiceprint processing module, a multi-modal analysis module, a fault judgment module and a result output module;

[0009] The parameter acquisition module is used to collect multi-modal raw data of the electric brake switch from various sensors and devices;

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

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

[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 the system or equipment has a fault, 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] Preferably, the monitoring method is implemented using the monitoring system according to claim 1, comprising:

[0016] S1, acquisition of multi-modal raw data of the electrical brake switch:

[0017] A parameter acquisition module is formed by installing a voiceprint collector, a vibration sensor, an SF6 gas pressure gauge, a temperature sensor, and a position switch at appropriate locations on the electric brake switch. The parameter acquisition module collects multimodal raw data of different modes during the operation of the electric brake switch, including voiceprint, vibration, gas leakage, temperature, and position status data.

[0018] S2, transmission of multimodal raw data of electrical brake switch:

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

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

[0021] First, the voiceprint data of the multimodal raw data is preprocessed. Then, the features of the voiceprint data are extracted to form sound features. The voiceprint processing module compares the extracted sound features with the preset voiceprint model, and gives a similarity or anomaly score. The voiceprint data is then anomaly detected using anomaly detection technology to identify possible anomalies for subsequent device status or fault judgment.

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

[0023] The results of the voiceprint processing module in S3 are transmitted to the multimodal analysis module as part of the input data for the multimodal analysis module. The remaining multimodal raw data of vibration, gas leakage, temperature and position status data obtained in S1 are comprehensively analyzed and processed. At the same time, the multimodal raw data are fused and analyzed using a deep learning algorithm to extract characteristic information that can reflect the working condition of the electrical brake switch. The current data is then compared with the normal working condition data to determine whether there is an abnormality. If there is an abnormality, the next step is to perform fault detection and diagnosis. Otherwise, the normal working condition database is continued to be monitored and updated for subsequent comparison.

[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, determining whether the system or equipment has a fault, as well as the type and severity of the fault, so as to promptly detect and handle potential abnormalities and improve system stability and safety. At the same time, based on historical and current data, it predicts possible fault trends and provides maintenance personnel with early warning information and maintenance recommendations, so that they can understand the working condition of the electric brake switch and take appropriate maintenance measures to extend its service life and reduce maintenance costs.

[0026] S6, output of the results:

[0027] 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.

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

[0029] S1.1. Select multiple voiceprint collector locations at appropriate locations on the electrical brake switch and install voiceprint collectors at these locations; select multiple vibration sensor locations at appropriate locations on the electrical brake switch and install vibration sensors at these locations; install an SF6 gas pressure gauge at the lower portion of the three-phase short-circuit enclosed busbar housing of the electrical brake switch; install multiple temperature sensors at appropriate locations on the electrical brake switch; and install a position switch on the electrical brake switch.

[0030] S1.2, through various types of raw data during the operation of voiceprint collector, vibration sensor, SF6 gas pressure gauge, temperature sensor and position switch, different types of raw data together constitute multimodal raw data.

[0031] Preferably, there are 4 voiceprint collector points in S1.1, and there are also 4 corresponding voiceprint collectors, which correspond to the circuit breaker operating rod end, the ground knife operating rod end, the circuit breaker oil pump motor housing, and the three-phase short-circuit closed busbar housing installed on the electric brake switch, and the voiceprint data is collected through the voiceprint collectors.

[0032] Preferably, there are two vibration sensor points in S1.1, and there are two corresponding vibration sensors, which are respectively installed at the ground base of the operating mechanism bracket of the electric brake switch, and vibration data is collected by the vibration sensors.

[0033] Preferably, there is one SF6 gas pressure gauge in S1.1;

[0034] There are three temperature sensors in S1.1, which are installed on the lower shell of the three-phase voltage transformer PT respectively;

[0035] There is one position switch in S1.1, which is installed at the front end of the circuit breaker operating mechanism, and position data is collected through the position switch.

[0036] Preferably, the network transmission module in S2 specifically includes: a data output unit, a local control unit, a network transmission device, a server device and a system platform;

[0037] The data output unit is responsible for physically connecting the parameter sensor to the network after collecting the signal, and sending the data to the external device or system through the transmission interface;

[0038] The local control unit is used to collect the status and data of the field equipment, send control instructions, exchange data with the host computer or other equipment, and control and monitor the field equipment in real time;

[0039] Network transmission equipment, consisting of routers and switches, is responsible for routing and forwarding data, ensuring that data can be accurately and efficiently transmitted to the target location;

[0040] Server equipment, consisting of high-performance computer systems, large-capacity storage devices, and network connection devices, provides data storage, processing, and analysis services, and supports a variety of network services;

[0041] The system platform provides a user interaction interface and is responsible for data integration, analysis, and display, and sends processed requests or instructions to the corresponding network device or server for processing.

[0042] Preferably, the specific steps of performing voiceprint feature extraction, analysis and scoring based on the voiceprint processing module in S3 are:

[0043] S3.1, voiceprint MFCC extraction:

[0044] S3.1.1, Pre-emphasis:

[0045] Pre-emphasis is to enhance the high-frequency components in the audio signal through a high-pass filter, and balance the spectrum so that the low-frequency and high-frequency components can be processed with the same signal-to-noise ratio in subsequent spectrum analysis;

[0046] Set a pre-emphasis coefficient, which is a number less than 1;

[0047] To perform differential operation on the audio signal, use the formula:

[0048] s′(n)=s(n)-α×s(n-1);

[0049] Where: s(n) represents the audio signal sampling value at the current moment, s(n-1) represents the audio signal sampling value at the previous moment, α is a pre-emphasis coefficient, and s'(n) is the current sampling value after pre-emphasis processing;

[0050] S3.1.2, Framing:

[0051] Framing is the process of cutting a continuous audio signal into shorter segments for short-term analysis. Each frame contains N sampling points, where N is 256 or 512. There is a certain overlap between frames to ensure signal continuity. The framing steps are as follows:

[0052]

[0053] Where F represents the total number of frames obtained after the framing operation in audio processing; L represents the total length of the audio signal, which is the sampling frequency multiplied by time; N is the length of each frame, that is, the number of sampling points contained in each frame; M is the frame shift, which represents the number of sampling points in the overlapping part between two consecutive frames, and is also the number of sampling points that the new frame moves relative to the previous frame. Represents the smallest integer not less than x, that is, rounded up;

[0054] S3.1.3, Windowing:

[0055] Windowing is used to increase the continuity of the signal between frames and reduce spectrum leakage. The Hamming window is defined as follows:

[0056] The mathematical expression of the Hamming window is:

[0057]

[0058] Where n is an integer from 0 to N-1, and N is the size of the window;

[0059] According to the mathematical expression of the Hamming window, the value w(n) of the Hamming window is calculated for the sampling point n in each frame. This calculation can be repeated for each frame.

[0060] S3.1.4, Windowing Operation:

[0061] Multiply the signal S(n) of each frame by the corresponding Hamming window value w(n). Mathematically, this can be expressed as:

[0062] S w (n) = S(n) × w(n);

[0063] Among them, S w (n) is the windowed signal;

[0064] S3.1.5, Mel filtering:

[0065] The energy spectrum is passed through a set of Mel-scale triangular filter banks to define a filter bank with M filters. The filters used are triangular filters. The logarithmic energy of each filter bank output is calculated. The MFCC coefficients are obtained by discrete cosine transform. The dynamic differential parameters are extracted. The differential parameters can be calculated using the following formula:

[0066]

[0067] Where L is half the size of the differential window, that is, the total window size is 2L±1, C t is the MFCC coefficient vector of the t-th frame, ΔC t is the first-order difference coefficient vector of the t-th frame;

[0068] The second-order difference is to apply the difference operation again to the first-order difference coefficient to capture faster changes in the audio signal. The formula is as follows:

[0069]

[0070] Where, Δ 2 C t is the second-order differential coefficient vector of the t-th frame, ΔC t is the obtained first-order difference coefficient vector;

[0071] The N-dimensional MFCC parameters are composed of (N / 3 MFCC coefficients + N / 3 first-order difference parameters + N / 3 second-order difference parameters) + frame energy;

[0072] S3.2, calculate UBM:

[0073] First, audio data unrelated to the target is collected to train a UBM. Then, the target audio data is used to adjust the parameters of the UBM through an adaptive algorithm to obtain the target model parameters.

[0074] S3.3, calculate i-vector:

[0075] i-vector defines a low-dimensional vector R×1,w~N(0,I) to represent an audio segment;

[0076]

[0077] Where M is the ideal feature supervector corresponding to a device being modeled; As the UBM mean supervector, assuming that the UBM contains C Gaussian mixture components g, then is the mean vector m of all mixture components c ,c=1,...,C combination, The dimension is C*F; the dimension of the transformation matrix T is usually CF×R, where C is the number of Gaussian components in the Gaussian mixture model, F is the dimension of the acoustic feature, and R is the dimension of the i-vector; w is the required i-vector;

[0078] For each mixture component c, there is also a parameter mixture weight w c , and covariance matrix Σc; split the T matrix according to the Gaussian components, then for each component c, a submatrix V can be obtained C , whose dimension is F×R; this submatrix V C In fact, it represents the linear transformation from the feature space of the Gaussian component to the i-vector space. Mathematically, the T matrix is ​​expressed as:

[0079]

[0080] Among them, each V C Each is an F×R matrix, corresponding to a Gaussian component in GMM:

[0081] μ c =m c +V c w;

[0082] Where μ c Represents the mean vector corresponding to the cth Gaussian component after a given i-vector w; m c is the original mean vector of the c-th Gaussian component in the UBM;

[0083] Define a piece of audio feature data X, whose feature dimension is F and time sequence is T, that is, X = X1,...,XT, the subset of X belonging to the cth Gaussian component is Xc, and a certain frame in the subset but

[0084]

[0085] This gives the following formula:

[0086]

[0087] The calculation formula of i-vector is:

[0088] w=(I+T T ∑ -1 N(u)T) -1 T T ∑ -1 F(u);

[0089] Where w is the required i-vector; I is the identity matrix; T is the transformation matrix, which maps the i-vector space to the feature space; N(u) is a diagonal matrix of dimension CF×CF, whose block matrices on the diagonal are the counts of each Gaussian component; F(u) is the supervector of the first-order Baum-Welch statistics, which is a CF×1 supervector consisting of all the first-order BW statistics F~c;

[0090] S3.4, Linear Probability Discriminant Analysis:

[0091] Linear probability discriminant analysis is based on i-vector features and provides channel compensation. Assume that the training data audio consists of audio from I devices, where each device has J different audio segments. The i-vector of the j-th audio segment of the i-th device is recorded as D ij , then PLDA defines:

[0092] D ij =μ+Fh i +Gω ij +ε ij ;

[0093] In the formula, the device information part is μ+Fh i , which is only related to device i and describes the difference between devices; the noise part is Gω ij +ε ij , describing the differences between devices; μ represents the mean of all training data; F is regarded as the identity space, which contains information that can be used to represent various devices; h i It is regarded as the identity of a specific device; G is the error space, which contains information used to represent different audio changes of the same device; ω ij represents the position in G space; ε ij is the final residual noise term, which is used to represent something that has not been explained yet; this term is zero-mean Gaussian distributed with variance Σ;

[0094] Assume that each latent variable conforms to the following distribution:

[0095]

[0096] Assuming that the model M describes the relationship between the identity factor h and the input feature i-vector, the test process is to determine the ivector x p Is it related to registering ictor x? i Share the same device identity h;

[0097] hypothesis M0 represents x i and x p From different identity latent variables, hypothesis M1 represents x i and x p From the same identity latent variable, the likelihood ratio calculates the score value:

[0098]

[0099] The numerator P(x1,xp / M1) represents the assumption that M1:x i and x p From the same device, the speech sample x is observed i and x p The joint probability of; the denominator P (x1, xp / M0) means that under the assumption MO: x i and x p Speech samples x are observed from different devices i and x p By calculating the score, we can measure the similarity between the 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.

[0100] Preferably, the analysis and processing of the multimodal raw data in S4 specifically includes:

[0101] S4.1, Data Selection:

[0102] Hardware partitioning is used to eliminate the impact of equipment anomalies in other areas on fault diagnosis in this area. The partitions are as follows: circuit breaker area, motherboard sealing area, and switch area. Each input parameter: voiceprint score, vibration sensor peak-to-peak value, position switch signal, and temperature are standardized to a value range between 0 and 1.

[0103] For each parameter p i , its normalized value n i Calculated by the following formula:

[0104]

[0105] Among them, p min and p max They are parameters p i Possible minimum and maximum values;

[0106] S4.2, set the basic scoring value and weight:

[0107] Set a basic score for each parameter and assign different weights to them according to their importance. The sum of the weights should be 1.

[0108] S4.3, calculate the rate of change:

[0109] For each parameter, calculate the rate of change between its current value and the initial value or the last measured value, the rate of change r i Calculated by the following formula:

[0110]

[0111] Among them, p i,current is the current value of the parameter, p i,previous is the previous value of the parameter;

[0112] S4.4, determine the threshold and adjust the score value:

[0113] If the change rate of any parameter exceeds 20%, the score of the parameter will be reduced to 0 points; if the change rate of more than half of the parameters exceeds 10%, the score of these parameters will be halved; the score of each parameter will be adjusted according to the change rate. i :

[0114]

[0115] Among them, base_score i is the base score value of parameter i, I is an indicator function: it returns 1 if the condition is true, otherwise it returns 0, and n is the total number of parameters;

[0116] S4.5, calculate the final score:

[0117] According to the current score value of each parameter and its weight, calculate the weighted average as the final score value;

[0118] Output score:

[0119]

[0120] Among them, w i is the weight of parameter i, and satisfies Output the final score, which should be between 0 and 10.

[0121] 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 PT poor contact discharge fault judgment.

[0122] Preferably, the oil pump motor internal fault judgment process is:

[0123] S5.1.1: The circuit breaker zone score value drops to 6, triggering the fault judgment process;

[0124] S5.1.2: The vibration sensor parameters and voiceprint score values ​​are judged simultaneously. If the 1# / 2# vibration sensor and 3# voiceprint sensor are out of limit, proceed to the next step.

[0125] S5.1.3: Compare the current fault voiceprint data with the standards in the fault library. If a fault and event match, directly output the corresponding fault event alarm. If it cannot match the oil pump pressure model, calculate the change rate of the last five pressure times. If the change rate is greater than 0, it indicates that the oil pump motor is slowly deteriorating. It is necessary to open the cover and inspect to determine the fault. The specific fault will be marked and added to the model library.

[0126] S5.1.4: If the matching model is the oil pump pressure model, count for 2 minutes, record the number of pressures plus 1 and the duration, and proceed to the next step;

[0127] S5.1.5: When the oil pump pressurization does not exceed 2 minutes, the output is the oil pump motor pressurization event;

[0128] S5.1.6: If the oil pump pressurization exceeds 2 minutes, the oil pump motor pressurization timeout alarm will be output and the next step will be entered;

[0129] S5.1.7: To prevent the stator winding from being damaged due to the occurrence of disconnection lockout or three-phase inconsistency during the operation of the 20KV electrical brake switch with a defect, it is necessary to lock the switch operation and notify the equipment maintenance personnel by phone to handle the situation on site.

[0130] Preferably, the SF6 gas leakage fault judgment process is:

[0131] S5.2.1: The score of the mother area drops to 6, triggering the fault judgment process;

[0132] S5.2.2: Simultaneously assess the vibration sensor parameters, voiceprint score, and SF6 pressure. If the 1# / 2# vibration sensors are normal but the 4# voiceprint sensor and SF6 gas pressure gauge are out of range, proceed to the next step.

[0133] S5.2.3: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library.

[0134] S5.2.4: If the match is SF6 gas leakage, an SF6 gas leakage alarm will be issued and the next step will be entered;

[0135] S5.2.5: In the event of confirmed SF6 gas leakage, the system will activate the corresponding emergency plan and lock the switch operation, which includes shutting down related equipment, ventilation, and evacuating personnel to ensure safety.

[0136] Preferably, the operating mechanism connecting rod fault judgment process is:

[0137] S5.3.1: The circuit breaker zone or switch zone score value drops to 6, triggering the fault judgment process;

[0138] S5.3.2: Determine the position of the circuit breaker switch and check whether the 1# / 2# vibration sensors are abnormal. Check whether the voiceprint scores of the 1# voiceprint sensor and the 2# voiceprint sensor are abnormal. If abnormal, proceed to the next step.

[0139] S5.3.3: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library.

[0140] S5.3.4: Handle circuit breaker linkage failure. If the fault is matched to a circuit breaker linkage failure, the system will immediately lock out the circuit breaker to ensure safety. At the same time, the system will determine whether the unit is in the process of shutting down. If so, the system will initiate the emergency shutdown process and skip the 20KV electrical brake switch activation step to prevent the fault from escalating or causing more serious consequences.

[0141] S5.3.5: Handle knife switch linkage failure. If the fault is determined to be a knife switch linkage failure, the system will immediately lock the knife switch operation to prevent the fault from worsening. At the same time, the system will check whether the unit is in the electrical brake engaged state. If so, the system will issue an electrical brake exit command and disconnect the electrical brake circuit breaker to ensure the stability and safety of the system.

[0142] Preferably, the PT poor contact discharge fault judgment process is:

[0143] S5.4.1: The circuit breaker zone or switch zone score is normal but the main lock zone score drops to 6, triggering the fault judgment process;

[0144] S5.4.2: Determine the status of the sensor parameters in the sealing mother area. If the 4# soundprint sensor is abnormal, the SF6 gas pressure gauge is normal, the 1# / 2# vibration sensors are normal, and the temperature sensor is abnormally high, proceed to the next step.

[0145] S5.4.3: If abnormal, locate the fault point and identify the phase based on the abnormal condition of the temperature sensor, that is, determine which phase has the problem;

[0146] S5.4.4: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library.

[0147] S5.4.5: After completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the PT poor contact discharge fault. If so, the system will output a PT poor contact discharge fault alarm and proceed to the next step.

[0148] S5.4.6: Determine the unit's operating status and address the issue. Confirm the unit's current status. If the unit is in shutdown mode, the system will set the unit to a disabled startup mode to prevent it from being accidentally started before the fault is resolved. Immediately arrange for inspection and repair.

[0149] S5.4.7: Adjust the unit shutdown priority; if the unit is currently in the on state, the system will increase the shutdown priority of the unit to the highest; once the branch plant needs to shut down, this unit will be given priority for shutdown, and during this period, attention and monitoring of the unit must be strengthened to ensure safety.

[0150] The present invention has the following beneficial effects:

[0151] 1. The present invention realizes multi-modal online real-time monitoring of 20KV electrical brake switchgear area by installing various types of acquisition sensors, filling the gap in the industry.

[0152] 2. The bone conduction microphone of the present invention collects audio data of 50Hz-4KHz and can detect almost all abnormal conditions in the 20KV electrical brake switch equipment area. At the same time, it is assisted by data such as temperature, vibration, and gas monitoring. Therefore, this solution can identify all faults in the 20KV electrical brake switch equipment area of ​​the equipment, greatly improving the accuracy of abnormal condition monitoring of the 20KV electrical brake switch equipment.

[0153] 3. The present invention uses voiceprint recognition technology and comprehensive judgment and analysis of multimodal data such as temperature, vibration, image, and gas monitoring to greatly improve the accuracy of abnormal operating condition monitoring of 20KV electrical brake switchgear, save human resources, and reduce the time required for fault identification, ensuring the safe and stable operation of the unit and avoiding accidents during startup and shutdown. BRIEF DESCRIPTION OF THE DRAWINGS

[0154] The present invention will be further described below with reference to the accompanying drawings and examples.

[0155] Figure 1This is the electrical connection diagram of the generator brake switch device of the present invention.

[0156] Figure 2 This is a system structure diagram of the present invention.

[0157] Figure 3 This is the hardware diagram of the parameter acquisition module of the present invention.

[0158] Figure 4 This is a structural diagram of the network transmission module of the present invention.

[0159] Figure 5 This is the MFCC extraction flow chart of the present invention.

[0160] Figure 6 The present invention intercepts 100ms of original sound data as continuous sound samples.

[0161] Figure 7 This is a waveform diagram of 100ms audio data after pre-emphasis processing according to the present invention.

[0162] Figure 8 The waveform after framing of the present invention takes the fifth frame as an example.

[0163] Figure 9 This is a comparison diagram of the waveforms before and after windowing of the present invention.

[0164] Figure 10 It is the 100ms logarithmic energy of the present invention.

[0165] Figure 11 It is the 13th-order MFCC value of the present invention.

[0166] Figure 12 FIG. 4 is a flow chart of calculating UBM according to the present invention.

[0167] Figure 13 The audio energy diagram of the 15s speech sample of the present invention compared with the model.

[0168] Figure 14 This is the analysis and processing flow of the multimodal raw data of the present invention.

[0169] Figure 15 is the peak-to-peak signal of the vibration sensor of the present invention.

[0170] Figure 16 It is the switch position signal of the present invention.

[0171] Figure 17 is the temperature signal of the present invention.

[0172] Figure 18 This is the internal fault judgment process of the oil pump motor of the present invention.

[0173] Figure 19 This is the SF6 gas leakage fault judgment process of the present invention.

[0174] Figure 20 This is the operating mechanism connecting rod fault judgment process of the present invention.

[0175] Figure 21 This is the PT poor contact discharge fault judgment process of the present invention. DETAILED DESCRIPTION

[0176] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0177] Example 1:

[0178] See also Figure 2 A 20KV electrical brake switch multi-modal operating condition abnormality comprehensive monitoring system based on voiceprint recognition includes: a parameter acquisition module 1, a network transmission module 2, a voiceprint processing module 3, a multi-modal analysis module 4, a fault judgment module 5 and a result output module; the parameter acquisition module 1 is used to collect multi-modal raw data of the electrical brake switch from various sensors and devices; the network transmission module 2 is used to transmit the multi-modal raw data collected by the parameter acquisition module 1 to the voiceprint processing module 3 and the multi-modal analysis module 4 through the network; the voiceprint processing module 3 is used to extract voiceprints from the multi-modal raw data The multimodal analysis module 4 is used to comprehensively analyze data from different modes, integrate multiple data, and use deep learning algorithms to perform data analysis and pattern recognition to obtain analysis results; the fault judgment module 5 is based on the analysis results of the multimodal analysis module 4 and 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; the result output module is used to display the final analysis and judgment results in the form of charts, reports or alarms.

[0179] Example 2:

[0180] See also Figure 2 The monitoring method is implemented using the monitoring system according to claim 1, comprising:

[0181] S1, acquisition of multi-modal raw data of the electrical brake switch:

[0182] A voiceprint collector, a vibration sensor, an SF6 gas pressure gauge, a temperature sensor, and a position switch are arranged and installed at appropriate locations on the electric brake switch to form a parameter acquisition module 1. The parameter acquisition module 1 collects multimodal raw data of different modes during the operation of the electric brake switch, including voiceprint, vibration, gas leakage, temperature, and position status data. These sensors can collect the operating condition data of the electric brake switch in real time and transmit it to a data processing center.

[0183] S2, transmission of multimodal raw data of electrical brake switch:

[0184] The multimodal raw data collected in S1 is transmitted in real time to the voiceprint processing module 3 and the multimodal analysis module 4 through the network transmission module 2 to achieve remote data transmission; specifically, signal cables, optical fibers or wireless network transmission can be used;

[0185] S3, extraction, analysis and scoring of voiceprint features:

[0186] First, the voiceprint data of the multimodal raw data is preprocessed, such as noise reduction and filtering, to improve the quality of the voice data. Then, the features of the voiceprint data are extracted to form sound features. The voiceprint processing module 3 compares the extracted sound features with the preset voiceprint model, and gives a similarity or anomaly score. The voiceprint data is then anomaly detected using anomaly detection technology to identify possible anomalies for subsequent device status or fault judgment.

[0187] S4, Analysis and processing of multimodal raw data:

[0188] The result of the voiceprint processing 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. The remaining multimodal raw data of the vibration, gas leakage, temperature and position status data obtained in S1 are comprehensively analyzed and processed. At the same time, the multimodal raw data are fused and the multimodal data are fused and analyzed using a deep learning algorithm to extract characteristic information that can reflect the working condition of the electric brake switch. The difference between the current data and the normal working condition data is then compared to determine whether there is an abnormality. If there is an abnormality, the next step is to perform fault detection and diagnosis. Otherwise, the normal working condition database is continued to be monitored and updated for subsequent comparison.

[0189] S5, fault judgment:

[0190] 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 the system or equipment has a fault, as well as the type and severity of the fault, so as to promptly discover and handle potential abnormal situations and thus improve the stability and safety of the system. At the same time, based on historical data and current data, possible fault trends are predicted to provide maintenance personnel with early warning information and maintenance suggestion services, so that they can understand the working condition of the electric brake switch and take appropriate maintenance measures to extend the service life and reduce maintenance costs.

[0191] S6, output of the results:

[0192] 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.

[0193] Example 3:

[0194] The specific steps of S1 are:

[0195] S1.1. Select multiple voiceprint collector locations at appropriate locations on the electrical brake switch and install voiceprint collectors at these locations; select multiple vibration sensor locations at appropriate locations on the electrical brake switch and install vibration sensors at these locations; install an SF6 gas pressure gauge at the lower portion of the three-phase short-circuit enclosed busbar housing of the electrical brake switch; install multiple temperature sensors at appropriate locations on the electrical brake switch; and install a position switch on the electrical brake switch.

[0196] S1.2, through various types of raw data during the operation of voiceprint collector, vibration sensor, SF6 gas pressure gauge, temperature sensor and position switch, different types of raw data together constitute multimodal raw data.

[0197] Preferably, there are 4 voiceprint collector points in S1.1, and there are also 4 corresponding voiceprint collectors 113-116, which correspond to the circuit breaker operating rod end, the ground knife operating rod end, the circuit breaker oil pump motor housing, and the three-phase short-circuit closed busbar housing installed on the electrical brake switch, and the voiceprint data is collected through the voiceprint collectors.

[0198] Preferably, there are two vibration sensor points in S1.1, and there are also two corresponding vibration sensors 118-119, which are respectively installed at the ground base of the operating mechanism bracket of the electric brake switch, and vibration data is collected through the vibration sensors.

[0199] Preferably, there is one SF6 gas pressure gauge 117 in S1.1;

[0200] There are three temperature sensors 120-122 in S1.1, which are respectively installed at the lower shell of the three-phase voltage transformer PT;

[0201] There is one position switch 123 in S1.1, which is installed at the front end of the circuit breaker operating mechanism, and position data is collected through the position switch.

[0202] Furthermore, in this embodiment:

[0203] Voiceprint collector:

[0204] Model: ZKHY-IVOD-001 Industrial Diagnostic Pickup;

[0205] Points: 113-116, a total of 4 points;

[0206] Installation location: circuit breaker operating rod end, ground switch operating rod end, circuit breaker oil pump motor housing, three-phase short-circuit enclosed busbar housing;

[0207] Vibration sensor:

[0208] Type: DSP vibration sensor;

[0209] Points: 118-119, 2 points in total;

[0210] Installation location: operating mechanism bracket ground base;

[0211] SF6 gas pressure gauge:

[0212] Model: GDM-100;

[0213] Point: 117, a total of 1 point;

[0214] Installation location: lower part of the three-phase short-circuit enclosed busbar housing;

[0215] Temperature sensor:

[0216] Model: PT100;

[0217] Points: 120-122, 3 points in total;

[0218] Installation location: below the casing of the three-phase voltage transformer PT;

[0219] Position switch:

[0220] Type: LX series limit switch;

[0221] Point: 123, a total of 1 point;

[0222] Installation location: front end of circuit breaker operating mechanism;

[0223] This comprehensive monitoring system uses multimodal monitoring methods, including sound patterns, vibration, gas leaks, temperature, and location status, to comprehensively monitor the operating status of 20kV electrical brake switches, promptly detecting and addressing abnormalities to ensure the stable and safe operation of the power system. Each sensor and device has been carefully selected and configured to meet diverse monitoring needs and provide comprehensive data support.

[0224] Example 4:

[0225] See also Figure 4 , the network transmission module 2 in S2 specifically includes: data output units 206-210, local control unit 205, network transmission device 204, server device 201 and system platforms 202-203;

[0226] The data output units 206-210 are responsible for collecting signals from parameter sensors and physically connecting to the network, and sending data to external devices or systems through transmission interfaces;

[0227] The local control unit 205 is used to collect the status and data of the field equipment, send control instructions, exchange data with the host computer or other equipment, and control and monitor the field equipment in real time;

[0228] Network transmission equipment 204, consisting of routers and switches, is responsible for routing and forwarding data, ensuring that data can be accurately and efficiently transmitted to the target location;

[0229] Server device 201, consisting of a high-performance computer system, a large-capacity storage device, and a network connection device, provides data storage, processing, and analysis services and supports a variety of network services;

[0230] The system platform 202-203 provides a user interaction interface, is responsible for data integration, analysis and display, and sends the processed requests or instructions to the corresponding network device or server for processing.

[0231] Example 4:

[0232] The voiceprint recognition algorithm engine performs weighted dimensionality reduction optimization based on MFCC feature vectors and applies a vector quantization algorithm to identify mechanical noise signals outside the cable. This includes voiceprint MFCC extraction, UBM calculation, i-vector calculation, and linear probability discriminant analysis.

[0233] The specific steps of voiceprint feature extraction, analysis and scoring based on the voiceprint processing module 3 in S3 are:

[0234] S3.1, voiceprint MFCC extraction:

[0235] The MFCC extraction process includes preprocessing, fast Fourier transform, Mei filter bank, logarithmic operation, discrete cosine transform, dynamic feature extraction and other steps. Figure 5 shown.

[0236] Get continuous audio: intercept 100ms of original sound data as continuous sound samples, such as Figure 6 shown.

[0237] S3.1.1, Pre-emphasis:

[0238] Pre-emphasis is the process of boosting the high-frequency components of an audio signal through a high-pass filter. The purpose of this step is to balance the spectrum so that in subsequent spectrum analysis, the low-frequency and high-frequency components can be processed with the same signal-to-noise ratio.

[0239] Set a pre-emphasis coefficient, which is a number less than 1;

[0240] To perform differential operation on the audio signal, use the formula:

[0241] s′(n)=s(n)-α×s(n-1);

[0242] Where: s(n) represents the audio signal sampling value at the current moment, s(n-1) represents the audio signal sampling value at the previous moment, α is a pre-emphasis coefficient, which is 0.95, and s'(n) is the current sampling value after pre-emphasis processing;

[0243] The waveform of 100ms audio data after pre-emphasis processing is as follows Figure 7 shown.

[0244] S3.1.2, Framing:

[0245] Framing is the process of cutting a continuous audio signal into shorter segments for short-term analysis. Each frame contains N sampling points, where N is 256 or 512. There is a certain overlap between frames to ensure signal continuity. The framing steps are as follows:

[0246]

[0247] Where F represents the total number of frames obtained after the framing operation in audio processing; L represents the total length of the audio signal, usually expressed as the number of sampling points, which is the sampling frequency multiplied by time: 16000*0.1=1600; N is the length of each frame, that is, the number of sampling points contained in each frame, which is 256; M is the frame shift, which represents the number of sampling points in the overlapping part between two consecutive frames, and is also the number of sampling points that the new frame moves relative to the previous frame, which is 128; x represents the smallest integer not less than x, that is, rounded up;

[0248] Take the 5th frame as an example of the waveform after framing. Figure 8 shown.

[0249] S3.1.3, Windowing:

[0250] Windowing is used to increase the continuity of the signal between frames and reduce spectrum leakage. The Hamming window is defined as follows:

[0251] The mathematical expression of the Hamming window is:

[0252]

[0253] Where n is an integer from 0 to N-1, and N is the size of the window;

[0254] According to the mathematical expression of the Hamming window, the value w(n) of the Hamming window is calculated for the sampling point n in each frame. This calculation can be repeated for each frame.

[0255] S3.1.4, Windowing Operation:

[0256] Multiply the signal S(n) of each frame by the corresponding Hamming window value w(n). Mathematically, this can be expressed as:

[0257] S w (n) = S(n) × w(n);

[0258] Among them, S w (n) is the windowed signal;

[0259] The waveform comparison before and after windowing is as follows Figure 9 shown.

[0260] S3.1.5, Mel filtering:

[0261] The energy spectrum is passed through a set of Mel-scale triangular filter banks to define a filter bank with M filters, where the filters used are triangular filters; the logarithmic energy of each filter bank output is calculated; the MFCC coefficients are obtained by discrete cosine transform; the logarithmic energy; and the dynamic classification parameters are extracted;

[0262] The energy spectrum is passed through a set of Mel-scale triangular filter banks to define a filter bank with M filters (the number of filters is close to the number of critical bands). The filters used are triangular filters, and M is 24.

[0263] Calculate the logarithmic energy of each filter group output: Logarithmic operation includes taking absolute value and log operation. Taking absolute value is to use only amplitude value and ignore the influence of phase, because phase information is not very useful in voiceprint recognition. After FFT transformation, convolution becomes multiplication, and after taking logarithm, multiplication becomes addition, converting the convolution signal into additive signal. 100ms logarithmic energy is as follows Figure 10 shown.

[0264] Obtaining MFCC coefficients through a discrete cosine transform (DCT): In the previous step, the fundamental frequency information and the vocal tract information were made additive. Another DCT can be performed to separate them, which is called the "cepstrum domain." Therefore, the low-frequency portion of the cepstrum domain depicts the vocal tract information, while the high-frequency portion depicts the fundamental frequency information. Substituting the aforementioned logarithmic energy into the discrete cosine transform, we determine the L-order Mel-scale Cepstrum parameter. L-order refers to the order of the MFCC coefficients, which is 13. Here, M is the number of triangular filters.

[0265] It turns out that the filter bank coefficients calculated in the previous step are highly correlated, which can be problematic for some machine learning algorithms. Therefore, a discrete cosine transform (DCT) is applied to decorrelate the filter bank coefficients and produce a compressed representation of the filter bank. The resulting cepstral coefficients 2-13 are retained, and the rest are discarded; num_ceps = 12.

[0266] Logarithmic energy: The volume (i.e., energy) of a frame is also an important audio feature and is very easy to calculate. Therefore, adding the logarithmic energy of a frame (defined as the sum of the squares of the signals within a frame, then taking the base 10 logarithm value, and multiplying it by 10) adds an additional dimension to the basic audio features of each frame, including the logarithmic energy and the remaining cepstrum parameters.

[0267] Extraction of dynamic classification parameters (including first-order and second-order differences): Standard MFCC parameters only reflect the static characteristics of audio parameters. The dynamic characteristics of audio can be described by the differential spectrum of these static characteristics. Experiments have shown that combining dynamic and static characteristics can effectively improve the recognition performance of the system. Extraction of dynamic differential parameters; The calculation of differential parameters can be done using the following formula:

[0268]

[0269] Where L is half the size of the differential window (i.e. the total window size is 2L±1), C t is the MFCC coefficient vector of the t-th frame, ΔC t is the first-order difference coefficient vector of the t-th frame;

[0270] The second-order difference is to apply the difference operation again to the first-order difference coefficient to capture faster changes in the audio signal. The formula is as follows:

[0271]

[0272] Where, Δ 2 C t is the second-order differential coefficient vector of the t-th frame, ΔC t is the obtained first-order difference coefficient vector.

[0273] From the above steps, we can see that MFCC is actually composed of: N-dimensional MFCC parameters (N / 3 MFCC coefficients + N / 3 first-order difference parameters + N / 3 second-order difference parameters) + frame energy (this item can be replaced according to needs).

[0274] The 13th-order MFCC value is as follows Figure 11 shown.

[0275] S3.2, calculate UBM:

[0276] First, audio data unrelated to the target is collected to train a UBM. Then, the target audio data is used to adjust the parameters of the UBM through an adaptive algorithm to obtain the target model parameters.

[0277] S3.3, calculate i-vector:

[0278] i-vector defines a low-dimensional vector R×1,w~N(0,I) to represent an audio segment;

[0279]

[0280] Where M is the ideal feature supervector corresponding to a device being modeled; As the UBM mean supervector, assuming that the UBM contains C Gaussian mixture components g, then is the mean vector m of all mixture components c ,c=1,...,C combination, The dimension is C*F; the dimension of the transformation matrix T is usually CF×R, where C is the number of Gaussian components in the Gaussian mixture model, F is the dimension of the acoustic feature, and R is the dimension of the i-vector; w is the required i-vector;

[0281] For each mixture component c, there is also a parameter mixture weight w c , and covariance matrix Σc; split the T matrix according to the Gaussian components, then for each component c, a submatrix V can be obtained C , whose dimension is F×R; this submatrix V C In fact, it represents the linear transformation from the feature space of the Gaussian component to the i-vector space. Mathematically, the T matrix is ​​expressed as:

[0282]

[0283] Among them, each V C Each is an F×R matrix, corresponding to a Gaussian component in GMM:

[0284] μ c =mc +V c w;

[0285] Where μ c Represents the mean vector corresponding to the cth Gaussian component after a given i-vector w; m c is the original mean vector of the c-th Gaussian component in the UBM;

[0286] Define a piece of audio feature data X, whose feature dimension is F and time sequence is T, that is, X = X1,...,XT, the subset of X belonging to the cth Gaussian component is Xc, and a certain frame in the subset but

[0287]

[0288] This gives the following formula:

[0289]

[0290] The calculation formula of i-vector is:

[0291] w=(I+T T ∑ -1 N(u)T) -1 T T ∑ -1 F(u);

[0292] Where w is the required i-vector; I is the identity matrix; T is the transformation matrix, which maps the i-vector space to the feature space; N(u) is a diagonal matrix of dimension CF×CF, whose block matrices on the diagonal are the counts of each Gaussian component; F(u) is the supervector of the first-order Baum-Welch statistics, which is a CF×1 supervector consisting of all the first-order BW statistics F~c;

[0293] S3.4, Linear Probability Discriminant Analysis:

[0294] Linear probability discriminant analysis is based on i-vector features and provides channel compensation. Assume that the training data audio consists of audio from I devices, where each device has J different audio segments. The i-vector of the j-th audio segment of the i-th device is recorded as D ij , then PLDA defines:

[0295] D ij =μ+F i +Gω ij +ε ij ;

[0296] In the formula, the device information part is μ+Fi , which is only related to device i and describes the difference between devices; the noise part is Gω ij +ε ij , describing the differences between devices; μ represents the mean of all training data; F is regarded as the identity space, which contains information that can be used to represent various devices; i It is regarded as the identity of a specific device; G is the error space, which contains information used to represent different audio changes of the same device; ω ij Represents the position in G space; ε ij is the final residual noise term, which is used to represent something that has not been explained yet; this term is zero-mean Gaussian distributed with variance Σ;

[0297] Assume that each latent variable conforms to the following distribution:

[0298]

[0299] Training set D = {D ij}, the model parameters are θ = {μ, F, G, Σ}, and the maximum likelihood estimation MLE criterion is used for optimization, that is:

[0300] arg max{P(θ / D)}∝argmax{P(D / θ)};

[0301] ① Initialize parameter θ;

[0302] ②E-step: estimate latent variables;

[0303] ③M-step: Update the parameter θ based on the estimated latent variable;

[0304] ④Repeat ② and ③ N times.

[0305] The likelihood ratio strategy is used for device identification.

[0306] Assuming that the model M describes the relationship between the identity factor h and the input feature i-vector, the test process is to determine the ivector x p Is it related to registering ictor x? i Share the same device identity h;

[0307] hypothesis M0 represents x i and x p From different identity latent variables, hypothesis M1 represents x i and x p From the same identity latent variable, the likelihood ratio calculates the score value:

[0308]

[0309] The numerator P(x1,xp / M1) represents the assumption that M1:x i and x p From the same device, the speech sample x is observed i and x p The joint probability of; the denominator P (x1, xp / M0) means that under the assumption MO: x i and x p Speech samples x are observed from different devices i and x p By calculating the score, we can measure the similarity between the 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.

[0310] like Figure 13 As shown in the figure, it is an audio energy graph comparing a 15s speech sample with the model. The final score of this audio segment is 90.

[0311] Example 5:

[0312] See also Figure 14 Preferably, the analysis and processing of the multimodal raw data in S4 specifically includes:

[0313] S4.1, Data Selection:

[0314] The hardware partitioning used in the 20KV electrical brake switch multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is to eliminate the impact of equipment abnormalities in other areas on fault judgment in this area. The partitioning is as follows:

[0315] Circuit breaker area: voiceprint collectors 113 and 115, position switch 123, vibration sensors 118 and 119;

[0316] Sealing area: Voiceprint collector 116, SF6 gas pressure gauge 117, temperature measurement RTD 120, 121, 122

[0317] Knife gate area: Voiceprint collector 114, vibration sensors 118, 119

[0318] Normalize each input parameter (voiceprint score, vibration sensor peak-to-peak value, position switch signal, temperature) so that its value range is between 0 and 1. Linear transformation or other normalization methods can be used.

[0319] The peak-to-peak signal of the vibration sensor is as follows: Figure 15 As shown, the switch position signal is Figure 16 As shown, the temperature signal is Figure 17 shown.

[0320] For each parameter p i , its normalized value ni Calculated by the following formula:

[0321]

[0322] Among them, p min and p max They are parameters p i Possible minimum and maximum values;

[0323] S4.2, set the basic scoring value and weight:

[0324] Set a basic score for each parameter. For example, set it to 10 points to indicate the healthiest state. According to the importance of each parameter, assign different weights to them. The sum of the weights should be 1.

[0325] S4.3, calculate the rate of change:

[0326] For each parameter, calculate the rate of change between its current value and the initial value or the last measured value, the rate of change r i Calculated by the following formula:

[0327]

[0328] Among them, p i,current is the current value of the parameter, p i,previous is the previous value of the parameter;

[0329] S4.4, determine the threshold and adjust the score value:

[0330] If the change rate of any parameter exceeds 20%, the score of the parameter will be reduced to 0 points; if the change rate of more than half of the parameters exceeds 10%, the score of these parameters will be halved; the score of each parameter will be adjusted according to the change rate. i :

[0331]

[0332] Among them, base_score i is the base score value of parameter i, I is an indicator function: it returns 1 if the condition is true, otherwise it returns 0, and n is the total number of parameters;

[0333] S4.5, calculate the final score:

[0334] According to the current score value of each parameter and its weight, calculate the weighted average as the final score value;

[0335] Output score:

[0336]

[0337] Among them, w i is the weight of parameter i, and satisfies Output the final score, which should be between 0 and 10.

[0338] Example 6:

[0339] 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 PT poor contact discharge fault judgment.

[0340] See also Figure 18 Preferably, the oil pump motor internal fault judgment process is:

[0341] S5.1.1: The circuit breaker zone score value drops to 6, triggering the fault judgment process;

[0342] S5.1.2: The vibration sensor parameters and voiceprint score values ​​are judged simultaneously. If the 1# / 2# vibration sensor 118 / 119 and the 3# voiceprint sensor 115 are out of limit, proceed to the next step;

[0343] S5.1.3: Compare the current fault voiceprint data with the standards in the fault library. If a fault and event match, directly output the corresponding fault event alarm. If it cannot match the oil pump pressure model, calculate the change rate of the last five pressure times. If the change rate is greater than 0, it indicates that the oil pump motor is slowly deteriorating. It is necessary to open the cover and inspect to determine the fault. The specific fault will be marked and added to the model library.

[0344] S5.1.4: If the matching model is the oil pump pressure model, count for 2 minutes, record the number of pressures plus 1 and the duration, and proceed to the next step;

[0345] S5.1.5: When the oil pump pressurization does not exceed 2 minutes, the output is the oil pump motor pressurization event;

[0346] S5.1.6: If the oil pump pressurization exceeds 2 minutes, the oil pump motor pressurization timeout alarm will be output and the next step will be entered;

[0347] S5.1.7: To prevent the stator winding from being damaged due to the occurrence of disconnection lockout or three-phase inconsistency during the operation of the 20KV electrical brake switch with a defect, it is necessary to lock the switch operation and notify the equipment maintenance personnel by phone to handle the situation on site.

[0348] See also Figure 19 Preferably, the SF6 gas leakage fault judgment process is:

[0349] S5.2.1: The score of the mother area drops to 6, triggering the fault judgment process;

[0350] S5.2.2: The vibration sensor parameters, voiceprint score, and SF6 pressure are determined simultaneously. If the 1# / 2# vibration sensors 118 / 119 are normal but the 4# voiceprint sensor 116 and the SF6 gas pressure gauge 117 are out of limit, proceed to the next step.

[0351] S5.2.3: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library.

[0352] S5.2.4: If the match is SF6 gas leakage, an SF6 gas leakage alarm will be issued and the next step will be entered;

[0353] S5.2.5: In the event of confirmed SF6 gas leakage, the system will activate the corresponding emergency plan and lock the switch operation, which includes shutting down related equipment, ventilation, and evacuating personnel to ensure safety.

[0354] See also Figure 20 Preferably, the operating mechanism connecting rod fault judgment process is:

[0355] S5.3.1: The circuit breaker zone or switch zone score value drops to 6, triggering the fault judgment process;

[0356] S5.3.2: Determine the position of the circuit breaker position switch 123, check whether the 1# / 2# vibration sensors 118 / 119 are abnormal, check whether the voiceprint score values ​​of the 1# voiceprint sensor 113 and the 2# voiceprint sensor 11 are abnormal. If abnormal, proceed to the next step;

[0357] S5.3.3: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library.

[0358] S5.3.4: Handle circuit breaker linkage failure. If the fault is matched to a circuit breaker linkage failure, the system will immediately lock out the circuit breaker to ensure safety. At the same time, the system will determine whether the unit is in the process of shutting down. If so, the system will initiate the emergency shutdown process and skip the 20KV electrical brake switch activation step to prevent the fault from escalating or causing more serious consequences.

[0359] S5.3.5: Handle knife switch linkage failure. If the fault is determined to be a knife switch linkage failure, the system will immediately lock the knife switch operation to prevent the fault from worsening. At the same time, the system will check whether the unit is in the electrical brake engaged state. If so, the system will issue an electrical brake exit command and disconnect the electrical brake circuit breaker to ensure the stability and safety of the system.

[0360] See also Figure 21Preferably, the PT poor contact discharge fault judgment process is:

[0361] S5.4.1: The circuit breaker zone or switch zone score is normal but the main lock zone score drops to 6, triggering the fault judgment process;

[0362] S5.4.2: Determine the status of the sensor parameters in the sealing mother area. If the 4# soundprint sensor 116 is abnormal, the SF6 gas pressure gauge 117 is normal, the 1# / 2# vibration sensors 118 / 119 are normal, and the temperature sensors 120-122 are abnormally high, proceed to the next step.

[0363] S5.4.3: If abnormal, locate the fault point and determine the phase based on the abnormal conditions of temperature sensors 120-122, that is, determine which phase has a problem;

[0364] S5.4.4: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library.

[0365] S5.4.5: After completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the PT poor contact discharge fault. If so, the system will output a PT poor contact discharge fault alarm and proceed to the next step.

[0366] S5.4.6: Determine the unit's operating status and address the issue. Confirm the unit's current status. If the unit is in shutdown mode, the system will set the unit to a disabled startup mode to prevent it from being accidentally started before the fault is resolved. Immediately arrange for inspection and repair.

[0367] S5.4.7: Adjust the unit shutdown priority; if the unit is currently in the on state, the system will increase the shutdown priority of the unit to the highest; once the branch plant needs to shut down, this unit will be given priority for shutdown, and during this period, attention and monitoring of the unit must be strengthened to ensure safety.

[0368] In summary, the main innovative solutions of the present invention are:

[0369] 1. A comprehensive monitoring process for multi-modal abnormal working conditions of a 20KV electrical brake switch based on voiceprint recognition. The process covers real-time monitoring and abnormality identification of multiple working condition parameters in the 20KV electrical brake switch.

[0370] 2. A method for arranging monitoring equipment in a 20KV electrical brake switch based on voiceprint recognition, including rationally arranging specific locations of monitoring equipment such as voiceprint sensors and vibration sensors within the equipment area to achieve comprehensive and effective equipment status monitoring.

[0371] 3. A 20KV electrical brake switch signal parameter network transmission system based on voiceprint recognition. The system is used to transmit signal parameters collected by various sensors in the equipment area in real time to achieve remote monitoring and data analysis.

[0372] 4. A method for extracting voiceprint feature values ​​of a 20KV electrical brake switch based on voiceprint recognition, including a specific process and algorithm for extracting key voiceprint feature values ​​from the signal collected by the voiceprint sensor.

[0373] 5. A 20KV electrical brake switch voiceprint probabilistic linear discriminant analysis model algorithm based on voiceprint recognition uses probabilistic linear discriminant analysis to perform discriminant analysis on the extracted voiceprint feature values ​​and give corresponding scoring values.

[0374] 6. A multimodal scoring model algorithm for a 20KV electrical brake switch based on voiceprint recognition. This algorithm uses hardware partitioning to avoid interference and is used to identify the operating conditions of equipment in different areas of a 20KV electrical brake switch.

[0375] 7. A 20KV electrical brake switch oil pump motor internal fault judgment logic system based on voiceprint recognition. The logic of judging the internal fault of the oil pump motor is based on multimodal data, combined with the signals of voiceprint and vibration sensor;

[0376] 8. A 20KV electrical brake switch SF6 gas leakage fault judgment logic system based on voiceprint recognition. The system uses multimodal data, combined with voiceprints and SF6 gas leakage detector signals, to judge the logic of SF6 gas leakage faults.

[0377] 9. A 20KV electrical brake switch operating mechanism connecting rod abnormality fault judgment logic system based on voiceprint recognition. Based on multimodal data, combined with voiceprint, vibration sensor, and position node signals, the system judges the logic of operating mechanism connecting rod abnormality fault;

[0378] 10. A 20KV electrical brake switch PT poor contact discharge fault judgment logic system based on voiceprint recognition, which uses multimodal data and combines the signals of voiceprints, vibration sensors, SF6 gas leak detectors, and temperature sensors to judge the logic of PT poor contact discharge faults.

Claims

1. A comprehensive monitoring method for multi-modal abnormal working conditions of a 20KV electrical brake switch based on voiceprint recognition, characterized in that: The monitoring method comprises: S1, acquisition of multi-modal raw data of the electrical brake switch: A voiceprint collector, a vibration sensor, an SF6 gas pressure gauge, a temperature sensor, and a position switch are respectively arranged and installed at appropriate positions of an electric brake switch to form a parameter acquisition module (1). The parameter acquisition module (1) collects multi-modal raw data of different modes of the electric brake switch during operation, specifically including: voiceprint, vibration, gas leakage, temperature, and position status data. S2, transmission of multimodal raw data of electrical brake switch: The multimodal raw data collected in S1 is transmitted in real time to the voiceprint processing 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: First, the voiceprint data of the multimodal raw data is preprocessed, and then the features of the voiceprint data are extracted to form sound features. The extracted sound features are compared with the preset voiceprint model through the voiceprint processing module (3), and a similarity or anomaly score is given. The voiceprint data is then anomaly detected using anomaly detection technology to identify possible anomalies for subsequent device status or fault judgment; S4, Analysis and processing of multimodal raw data: The result of the voiceprint processing module (3) in S3 is transmitted to the multimodal analysis module (4) as a part of the input data for the analysis of the multimodal analysis module (4), and the remaining multimodal raw data of the vibration, gas leakage, temperature and position status data obtained in S1 are comprehensively analyzed and processed, and the multimodal raw data are fused at the same time. The multimodal data are fused and analyzed using a deep learning algorithm to extract characteristic information that can reflect the working condition of the electric brake switch; then the difference between the current data and the normal working condition data is compared to determine whether there is an abnormality; if there is an abnormality, the next step is to perform fault detection and diagnosis; otherwise, the normal working condition database is continued to be monitored and updated for subsequent comparison; 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 the system or equipment has a fault, as well as the type and severity of the fault, so as to timely discover and handle potential abnormal situations and thus improve the stability and safety of the system; at the same time, based on historical data and current data, possible fault trends are predicted to provide maintenance personnel with early warning information and maintenance suggestion services, so as to understand the working condition of the electric brake switch and take corresponding maintenance measures to extend the service life and reduce maintenance costs; S6, output of the results: The analysis and judgment results finally obtained in S4 are presented to the user or system administrator in the form of charts, reports or alarms with the help of the result output module.

2. According to claim 1, a 20KV electrical brake switch multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: The specific steps of S1 are: S1.

1. Select multiple voiceprint collector locations at appropriate locations on the electrical brake switch and install voiceprint collectors at these locations; select multiple vibration sensor locations at appropriate locations on the electrical brake switch and install vibration sensors at these locations; install an SF6 gas pressure gauge at the lower portion of the three-phase short-circuit enclosed busbar housing of the electrical brake switch; install multiple temperature sensors at appropriate locations on the electrical brake switch; and install a position switch on the electrical brake switch. S1.2, through various types of raw data during the operation of voiceprint collector, vibration sensor, SF6 gas pressure gauge, temperature sensor and position switch, different types of raw data together constitute multimodal raw data.

3. According to claim 2, a 20KV electrical brake switch multi-modal operating abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: There are a total of 4 voiceprint collector points in S1.1, and there are also 4 corresponding voiceprint collectors (113-116), which correspond to the circuit breaker operating link end installed on the electric brake switch, the ground knife operating link end, the circuit breaker oil pump motor housing, and the three-phase short-circuit closed busbar housing, and the voiceprint data is collected through the voiceprint collector.

4. According to claim 2, a 20KV electrical brake switch multi-modal operating condition abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: There are two vibration sensor points in S1.1, and there are also two corresponding vibration sensors (118-119), which are respectively installed at the ground base of the operating mechanism bracket of the electric brake switch, and vibration data is collected by the vibration sensors.

5. According to claim 2, a 20KV electrical brake switch multi-modal operating abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: There is one SF6 gas pressure gauge (117) in S1.1; There are three temperature sensors (120-122) in S1.1, which are respectively installed at the lower shell of the three-phase voltage transformer PT; There is one position switch (123) in S1.1, which is installed at the front end of the circuit breaker operating mechanism, and position data is collected through the position switch.

6. According to claim 2, a 20KV electrical brake switch multi-modal operating abnormality comprehensive monitoring method based on voiceprint recognition is characterized in that: The network transmission module (2) in S2 specifically includes: a data output unit (206-210), a local control unit (205), a network transmission device (204), a server device (201) and a system platform (202-203); The data output unit (206-210) is responsible for physically connecting the parameter sensor to the network after collecting the signal, and sending the data to the external device or system through the transmission interface; The local control unit (205) is used to collect the status and data of the field equipment, send control instructions, exchange data with the host computer or other equipment, and control and monitor the field equipment in real time; Network transmission equipment (204), composed of routers and switch network equipment, is responsible for routing and forwarding data to ensure that data can be accurately and efficiently transmitted to the target location; The server device (201) is composed of a high-performance computer system, a large-capacity storage device, and a network connection device, and provides data storage, processing, and analysis services, and supports a variety of network services; The system platform (202-203) provides a user interaction interface, is responsible for data integration, analysis and display, and sends the processed requests or instructions to the corresponding network device or server for processing.

7. A 20KV electrical brake switch multi-modal operating abnormality comprehensive monitoring method based on voiceprint recognition according to claim 6, characterized in that: The specific steps of voiceprint feature extraction, analysis and scoring based on the voiceprint processing module (3) in S3 are as follows: S3.1, voiceprint MFCC extraction: S3.1.1, Pre-emphasis: Pre-emphasis is to enhance the high-frequency components in the audio signal through a high-pass filter, and balance the spectrum so that the low-frequency and high-frequency components can be processed with the same signal-to-noise ratio in subsequent spectrum analysis; Set a pre-emphasis coefficient, which is a number less than 1; To perform differential operation on the audio signal, use the formula: s′(n)=s(n)-α×s(n-1); Where: s(n) represents the audio signal sampling value at the current moment, s(n-1) represents the audio signal sampling value at the previous moment, α is a pre-emphasis coefficient, and s'(n) is the current sampling value after pre-emphasis processing; S3.1.2, Framing: Framing is the process of cutting a continuous audio signal into shorter segments for short-term analysis. Each frame contains N sampling points, where N is 256 or 512. There is a certain overlap between frames to ensure signal continuity. The framing steps are as follows: Where F represents the total number of frames obtained after the framing operation in audio processing; L represents the total length of the audio signal, which is the sampling frequency multiplied by time; N is the length of each frame, that is, the number of sampling points contained in each frame; M is the frame shift, which represents the number of sampling points in the overlapping part between two consecutive frames, and is also the number of sampling points that the new frame moves relative to the previous frame. Represents the smallest integer not less than x, that is, rounded up; S3.1.3, Windowing: Windowing is used to increase the continuity of the signal between frames and reduce spectrum leakage. The Hamming window is defined as follows: The mathematical expression of the Hamming window is: Where n is an integer from 0 to N-1, and N is the size of the window; According to the mathematical expression of the Hamming window, the value w(n) of the Hamming window is calculated for the sampling point n in each frame. This calculation can be repeated for each frame. S3.1.4, Windowing Operation: Multiply the signal S(n) of each frame by the corresponding Hamming window value w(n). Mathematically, this can be expressed as: S w (n)=S(n)×w(n); Among them, S w (n) is the windowed signal; S3.1.5, Mel filtering: The energy spectrum is passed through a set of Mel-scale triangular filter banks to define a filter bank with M filters. The filters used are triangular filters. The logarithmic energy of each filter bank output is calculated. The MFCC coefficients are obtained by discrete cosine transform. The dynamic differential parameters are extracted. The differential parameters can be calculated using the following formula: Where L is half the size of the differential window, that is, the total window size is 2L±1, C t is the MFCC coefficient vector of the t-th frame, ΔC t is the first-order difference coefficient vector of the t-th frame; The second-order difference is to apply the difference operation again to the first-order difference coefficient to capture faster changes in the audio signal. The formula is as follows: Where, Δ 2 C t is the second-order differential coefficient vector of the t-th frame, ΔC t is the obtained first-order difference coefficient vector; N-dimensional MFCC parameters are composed of (N / 3 MFCC coefficients + N / 3 first-order difference parameters + N / 3 second-order difference parameters) + frame energy; S3.2, calculate UBM: First, audio data unrelated to the target is collected to train a UBM. Then, the target audio data is used to adjust the parameters of the UBM through an adaptive algorithm to obtain the target model parameters. S3.3, calculate i-vector: i-vector defines a low-dimensional vector R×1,w~N(0,I) to represent an audio segment; M=m+T×w; Where M is the ideal feature supervector corresponding to a device being modeled; m is the UBM mean supervector. Assuming that the UBM contains C Gaussian mixture components g, then m is the mean vector m of all mixture components. c ,c=1,...,C combination, the dimension of m is C*F; the dimension of the transformation matrix T is usually CF×R, where C is the number of Gaussian components in the Gaussian mixture model, F is the dimension of the acoustic feature, and R is the dimension of the i-vector; w is the required i-vector; For each mixture component c, there is also a parameter mixture weight w c, and covariance matrix Σc; split the T matrix according to the Gaussian components, then for each component c, a submatrix V can be obtained C , whose dimension is F×R; this submatrix V C In fact, it represents the linear transformation from the feature space of the Gaussian component to the i-vector space. Mathematically, the T matrix is ​​expressed as: Among them, each V C Each is an F×R matrix, corresponding to a Gaussian component in GMM: μ c =m c +V c w; Where μ c Represents the mean vector corresponding to the cth Gaussian component after a given i-vector w; m c is the original mean vector of the c-th Gaussian component in the UBM; Define a piece of audio feature data X, whose feature dimension is F and time sequence is T, that is, X = X1,...,XT, the subset of X belonging to the cth Gaussian component is Xc, and a certain frame in the subset but This gives the following formula: The calculation formula of i-vector is: w=(I+T T ∑ -1 N(u)T) -1 T T ∑ -1 F(u); Where w is the required i-vector; I is the identity matrix; T is the transformation matrix, which maps the i-vector space to the feature space; N(u) is a diagonal matrix of dimension CF×CF, whose block matrices on the diagonal are the counts of each Gaussian component; F(u) is the supervector of the first-order Baum-Welch statistics, which is a CF×1 supervector consisting of all the first-order BW statistics F~c; S3.4, Linear Probability Discriminant Analysis: Linear probability discriminant analysis is based on i-vector features and provides channel compensation. Assume that the training data audio consists of audio from I devices, where each device has J different audio segments. The i-vector of the j-th audio segment of the i-th device is recorded as D ij , then PLDA defines: D ij =μ+Fh i +Gω ij +e ij ; In the formula, the device information part is μ+Fh i , which is only related to device i and describes the difference between devices; the noise part is Gω ij +ε ij , describing the differences between devices; μ represents the mean of all training data; F is regarded as the identity space, which contains information that can be used to represent various devices; h i It is regarded as the identity of a specific device; G is the error space, which contains information used to represent different audio changes of the same device; ω ij Represents the position in G space; ε ij is the final residual noise term, which is used to represent something that has not been explained yet; this term is zero-mean Gaussian distributed with variance Σ; Assume that each latent variable conforms to the following distribution: Assuming that the model M describes the relationship between the identity factor h and the input feature i-vector, the test process is to determine ivectorx p Is it related to registering ictor x? i Share the same device identity h; Assume that M0 represents x i and x p From different identity latent variables, assume that M1 represents x i and x p From the same identity latent variable, the likelihood ratio calculates the score value: The numerator P(x1,xp / M1) represents the assumption that M1:x i and x p From the same device, the speech sample x is observed i and x p The joint probability of; the denominator P (x1, xp / M0) means that under the assumption MO: x i and x p Speech samples x are observed from different devices i and x p By calculating the score, we can measure the similarity between the 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.

8. The method for comprehensive monitoring of multi-modal abnormal working conditions of a 20KV electrical brake switch based on voiceprint recognition according to claim 7 is characterized in that: The analysis and processing of the multimodal raw data in S4 specifically includes: S4.1, Data Selection: Hardware partitioning is used to eliminate the impact of equipment anomalies in other areas on fault diagnosis in this area. The partitions are as follows: circuit breaker area, motherboard sealing area, and switch area. Each input parameter: voiceprint score, vibration sensor peak-to-peak value, position switch signal, and temperature are standardized to a value range between 0 and 1. For each parameter p i , its normalized value n i Calculated by the following formula: Among them, p min and p max They are parameters p i Possible minimum and maximum values; S4.2, set the basic scoring value and weight: Set a basic score for each parameter and assign different weights to them according to their importance. The sum of the weights should be 1. S4.3, calculate the rate of change: For each parameter, calculate the rate of change between its current value and the initial value or the last measured value, the rate of change r i Calculated by the following formula: Among them, p i,current is the current value of the parameter, p i,previous is the previous value of the parameter; S4.4, determine the threshold and adjust the score value: If the change rate of any parameter exceeds 20%, the score of the parameter will be reduced to 0 points; if the change rate of more than half of the parameters exceeds 10%, the score of these parameters will be halved; the score of each parameter will be adjusted according to the change rate. i : Among them, base_score i is the base score value of parameter i, I is an indicator function: it returns 1 if the condition is true, otherwise it returns 0, and n is the total number of parameters; S4.5, calculate the final score: According to the current score value of each parameter and its weight, calculate the weighted average as the final score value; Output score: Among them, w i is the weight of parameter i, and satisfies Output the final score, which should be between 0 and 10.

9. The method for comprehensive monitoring of multi-modal abnormal working conditions of a 20KV electrical brake switch based on voiceprint recognition according to claim 8 is 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 PT poor contact discharge fault judgment.

10. A 20KV electrical brake switch multi-modal operating abnormality comprehensive monitoring method based on voiceprint recognition according to claim 9, characterized in that: The oil pump motor internal fault judgment process is as follows: S5.1.1: The circuit breaker zone score value drops to 6, triggering the fault judgment process; S5.1.2: The vibration sensor parameters and voiceprint score values ​​are judged simultaneously. If the 1# / 2# vibration sensor (118 / 119) and the 3# voiceprint sensor (115) are out of limit, proceed to the next step; S5.1.3: Compare the current fault voiceprint data with the standards in the fault library. If a fault and event match, directly output the corresponding fault event alarm. If it cannot match the oil pump pressure model, calculate the change rate of the last five pressure times. If the change rate is greater than 0, it indicates that the oil pump motor is slowly deteriorating. It is necessary to open the cover and inspect to determine the fault. The specific fault will be marked and added to the model library. S5.1.4: If the matching model is the oil pump pressure model, count for 2 minutes, record the number of pressures plus 1 and the duration, and proceed to the next step; S5.1.5: When the oil pump pressurization does not exceed 2 minutes, the output is the oil pump motor pressurization event; S5.1.6: If the oil pump pressurization exceeds 2 minutes, the oil pump motor pressurization timeout alarm will be output and the next step will be entered; S5.1.7: To prevent the stator winding from being damaged due to the occurrence of disconnection lockout or three-phase inconsistency during the operation of the 20KV electrical brake switch with a defect, it is necessary to lock the switch operation and notify the equipment maintenance personnel by phone to handle the situation on site.

11. The method for comprehensive monitoring of multi-modal abnormal working conditions of a 20KV electrical brake switch based on voiceprint recognition according to claim 9, characterized in that: The SF6 gas leakage fault judgment process is as follows: S5.2.1: The score of the mother area drops to 6, triggering the fault judgment process; S5.2.2: The vibration sensor parameters, voiceprint score and SF6 pressure are judged simultaneously. If the 1# / 2# vibration sensors (118 / 119) are normal but the 4# voiceprint sensor (116) and the SF6 gas pressure gauge (117) are out of limit, proceed to the next step. S5.2.3: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library. S5.2.4: If the match is SF6 gas leakage, an SF6 gas leakage alarm will be issued and the next step will be entered; S5.2.5: In the event of confirmed SF6 gas leakage, the system will activate the corresponding emergency plan and lock the switch operation, which includes shutting down related equipment, ventilation, and evacuating personnel to ensure safety.

12. The method for comprehensive monitoring of multi-modal abnormal working conditions of a 20KV electrical brake switch based on voiceprint recognition according to claim 9, characterized in that: The operating mechanism connecting rod fault judgment process is as follows: S5.3.1: The circuit breaker zone or switch zone score value drops to 6, triggering the fault judgment process; S5.3.2: The circuit breaker position switch (123) determines the position change, checks whether the 1# / 2# vibration sensor (118 / 119) is abnormal, checks whether the voiceprint score value of the 1# voiceprint sensor (113) and the voiceprint score value of the 2# voiceprint sensor (11) are abnormal, and if abnormal, proceeds to the next step; S5.3.3: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library. S5.3.4: Handle a circuit breaker linkage fault. If the fault is matched to a circuit breaker linkage fault, the system will immediately lock out the circuit breaker to ensure safety. At the same time, the system will determine whether the unit is in the process of shutting down. If so, the system will initiate the emergency shutdown process and skip the 20KV electrical brake switch activation step to prevent the fault from expanding or causing more serious consequences. S5.3.5: Handle knife switch linkage failure. If the fault is determined to be a knife switch linkage failure, the system will immediately lock the knife switch operation to prevent the fault from worsening. At the same time, the system will check whether the unit is in the electrical brake engaged state. If so, the system will issue an electrical brake exit command and disconnect the electrical brake circuit breaker to ensure the stability and safety of the system.

13. The method for comprehensive monitoring of multi-modal abnormal working conditions of a 20KV electrical brake switch based on voiceprint recognition according to claim 9, characterized in that: The PT poor contact discharge fault judgment process is as follows: S5.4.1: The circuit breaker zone or switch zone score is normal but the main lock zone score drops to 6, triggering the fault judgment process; S5.4.2: Determine the status of the sensor parameters in the sealing mother area. If the 4# sound print sensor (116) is abnormal, the SF6 gas pressure gauge (117) is normal, the 1# / 2# vibration sensors (118 / 119) are normal, and the temperature sensors (120-122) are abnormally high, proceed to the next step if the above conditions are met. S5.4.3: If abnormal, locate the fault point and determine the phase based on the abnormal condition of the temperature sensor (120-122), that is, determine which phase has the problem; S5.4.4: Compare the current fault voiceprint data with the standards in the fault library. If they match, proceed to the next step. If not, an on-site inspection is required to confirm the fault and enter it into the model library. S5.4.5: After completing the comparison between the fault library and the model library, the system will determine whether the current fault matches the PT poor contact discharge fault. If so, the system will output a PT poor contact discharge fault alarm and proceed to the next step. S5.4.6: Determine the unit's operating status and address the issue. Confirm the unit's current status. If the unit is in shutdown mode, the system will set the unit to a disabled startup mode to prevent it from being accidentally started before the fault is resolved. Immediately arrange for inspection and repair. S5.4.7: Adjust the unit shutdown priority; if the unit is currently in the on state, the system will increase the shutdown priority of the unit to the highest; Once the branch plant needs to be shut down, this unit will be given priority for shutdown, and during this period, attention and monitoring of the unit will be strengthened to ensure safety.

Images