Transformer abnormal state diagnosis and early warning method based on voiceprint features

Abstract

The invention discloses a transformer abnormal state diagnosis and early warning method based on voiceprint characteristics. The method comprises the following steps: acquiring a network topological graph and voiceprint information of a transformer network node; the method comprises the following steps: dividing a transformer into category voiceprint information data by using a support vector machine algorithm, generating a category voiceprint data set, establishing a transformer abnormal voiceprint data optimization model, and updating and outputting transformer abnormal state voiceprint features by using feature smoothing processing according to the transformer abnormal voiceprint data optimization model. And according to the abnormal voiceprint map feature database, through similarity analysis, generating a transformer abnormal state analysis result and outputting the transformer abnormal state analysis result. According to the method, the transformer voiceprint information can be recognized and updated in a complex sound field environment by optimizing and smoothing the voiceprint features, so that the precision and robustness of a subsequent fusion diagnosis model are improved, and the abnormal state diagnosis and early warning effects of the transformer are further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer data processing, and in particular to a transformer abnormal state diagnosis and early warning method based on voiceprint features. Background Art

[0002] Power transformers emit continuous vibration signals during operation, which contain a large number of pulses and fluctuations caused by mechanical faults. These signals are the main data source for evaluating the operating status of power transformers. When the transformer is abnormal, some of them are manifested as abnormal data of multiple monitoring / detection methods, and some are manifested as abnormal data of a single or a few monitoring / detection methods. Among them, ultrasonic and conventional pulse current methods are also considered to be effective insulation fault detection methods for electrical equipment, but if the discharge fault can also be judged based on the oil chromatography data, the diagnosis conclusion of the insulation fault is more certain, and the vibration analysis method is more effective for monitoring the condition of transformer windings and cores.

[0003] However, the existing ultrasonic, conventional pulse current and vibration analysis methods are limited in the processing of voiceprint features of transformer windings and core abnormal status diagnosis when used to diagnose the abnormal status of transformer cores. It is not easy to carry out the subsequent fusion diagnosis and processing of ultrasonic, vibration and partial discharge instrument feature information. Therefore, we propose a transformer abnormal status diagnosis and early warning method based on voiceprint features. Summary of the invention

[0004] In view of the above problems existing in the existing transformer data processing, the present invention is proposed.

[0005] Therefore, one of the objects of the present invention is to provide a method for diagnosing and warning of abnormal conditions of transformers based on voiceprint features, which optimizes and smoothes the voiceprint features of the ultrasonic and vibration voiceprint signals of the transformer in the topological structure, thereby realizing the recognition and update of transformer voiceprint information in a complex sound field environment, so as to improve the accuracy and robustness of the subsequent fusion diagnosis model, and improve the fusion diagnosis processing and analysis method, thereby further improving the effect of diagnosis and warning of abnormal conditions of transformers.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention provides a transformer abnormal state diagnosis and early warning method based on voiceprint features, comprising the following steps:

[0008] Step 1: Obtain the network topology diagram and voiceprint information of the transformer network nodes;

[0009] Step 2: Use a support vector machine algorithm to project the network topology of the transformer network nodes and the classified sample data of the voiceprint information into a high-dimensional space, divide them into category voiceprint information data, and generate a category voiceprint data set;

[0010] Step 3, establishing a transformer abnormal voiceprint data optimization model, wherein the transformer abnormal voiceprint data optimization model and objective function are determined with the maximum processing output of the transformer abnormal voiceprint in the transformer network node network topology diagram as the goal;

[0011] Step 4: Optimizing the model and the objective function according to the abnormal voiceprint data of the transformer, and using feature smoothing processing to complete the update and output of the voiceprint features of the abnormal state of the transformer, so as to form the voiceprint spectrum feature data of the abnormal state of the transformer;

[0012] Step five, based on the voiceprint feature data of the transformer abnormal state, an abnormal voiceprint feature database of the transformer abnormal state is formed, and then the transformer abnormal state analysis result is generated through similarity analysis according to the abnormal voiceprint feature database, and after matching the transformer abnormal state analysis result, the transformer abnormal state fault type detection result is obtained and output.

[0013] As a preferred solution of the present invention, in step 2, the network topology diagram of the transformer network node and the classified sample data of the voiceprint information are classified into linear data in a high-dimensional space by using a support vector machine algorithm, and the network topology diagram of the transformer network node and the voiceprint information samples are divided by finding the optimal classification hyperplane;

[0014] Project the divided sample data back to the source space to realize the division of nonlinear samples, and according to the Mercer theorem of kernel function, use the kernel function that meets the preset conditions as the inner product of the high-dimensional space to replace the inner product of the source space;

[0015] The nonlinear transformation of SVM is transformed into the objective function of quadratic programming to obtain the voiceprint information classification function, and a category voiceprint data set is generated according to the voiceprint information classification function.

[0016] As a preferred solution of the present invention, the maximum processing output of the abnormal voiceprint of the transformer in the transformer network node network topology diagram is taken as the goal, and the optimization model and objective function of the abnormal voiceprint data of the transformer are determined, which are specifically as follows:

[0017]

[0018] Where maxf(bit) is the maximum amount of processed output data of transformer abnormal voiceprint in the network topology diagram within the preset time, B jis the rated processing capacity data of transformer j, m represents the total number of transformer groups in the network topology, T represents the preset time data, t j represents the processing time of transformer j, is the increased processing capacity data of transformer j.

[0019] As a preferred solution of the present invention, wherein: the transformer abnormal state fault types include transformer winding fault, transformer core fault, transformer internal discharge fault and transformer tap switch fault;

[0020] The transformer winding faults include turn-to-turn short circuit faults, phase-to-phase short circuit faults, winding strand short circuit faults and winding deformation faults;

[0021] The transformer core fault includes multi-point grounding fault and core overheating fault;

[0022] The transformer internal discharge fault includes a partial discharge fault, a spark discharge fault, and an arc discharge fault.

[0023] As a preferred solution of the present invention, an abnormal voiceprint spectrum feature database of the transformer abnormal state is formed based on the voiceprint spectrum feature data of the transformer abnormal state, specifically, according to several voiceprint spectrum feature data of the transformer abnormal state, combined with a preset voiceprint vibration sensor signal weighting matrix, a weighted MUSIC algorithm is used to calculate the spatial spectrum of several voiceprint spectrum feature data of the transformer abnormal state, and the spatial spectrum is converted into a plane image, and then the fault type of the transformer abnormal state is selected and determined.

[0024] As a preferred solution of the present invention, in step 4, the feature smoothing process is as follows:

[0025]

[0026] Among them, p i Represented as the predicted value of the i-th category, λ i represents the weight value of the i-th category, n is the total number of items in the ith category, r i represents the characteristic parameter value of the i-th category, r n Expressed as the characteristic parameter value of the total category, p n It is expressed as the predicted value of the total category, and cos(·) represents the cosine function.

[0027] As a preferred solution of the present invention, wherein: according to the abnormal voiceprint feature database, similarity is calculated specifically between the voiceprint feature data to be identified and each abnormal voiceprint feature sample in the abnormal voiceprint feature database through similarity analysis; the maximum similarity calculation result of the similarity calculation of the voiceprint feature data to be identified is compared with the preset threshold of each corresponding abnormal voiceprint feature sample; and the corresponding abnormal voiceprint feature and the abnormal state of the transformer are matched and output after comparative analysis.

[0028] As a preferred solution of the present invention, wherein: step 1, when obtaining the network topology diagram and voiceprint information of the transformer network nodes, also includes denoising the obtained transformer voiceprint information.

[0029] As a preferred solution of the present invention, after matching the transformer abnormal state analysis result in step 5 to obtain the transformer abnormal state fault type detection result and output it, it also includes a fusion pulse current method detection method to fuse the transformer abnormal state fault type detection result, and the fusion model is as follows:

[0030]

[0031] Among them, G(U,E) represents the fusion data of the transformer abnormal state fault type detection results, U represents the set of unified data of the current transformer abnormal voiceprint fusion processing; E represents the current transformer abnormal voiceprint fusion processing event; U j The unified data representing the current transformer abnormal state fault type, U a Represents the unified data of the abnormal transformer soundprint detected by the current transformer abnormal soundprint pulse current method; w(U j ,E) represents the support of the unified data of the current transformer abnormal state fault type for the current transformer abnormal voiceprint fusion processing event; corr(U j ,U a ) represents the unified data U of the current transformer abnormal state fault type j The unified data U of the transformer abnormal soundprint detected by the current transformer abnormal soundprint pulse current method a of relevance.

[0032] Compared with the prior art, the present invention has the following beneficial effects: the present invention optimizes and smoothes the voiceprint features of the ultrasonic and vibration voiceprint signals of the transformer in the topological structure, thereby realizing the recognition and update of the transformer voiceprint information in a complex sound field environment, so as to improve the accuracy and robustness of the subsequent fusion diagnosis model, and improves the fusion diagnosis processing and analysis method, further improving the effect of transformer abnormal state diagnosis and early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0034] Figure 1 It is a flow chart of the transformer abnormal state diagnosis and early warning method based on voiceprint features of the present invention;

[0035] Figure 2 It is a schematic diagram of the influence of 100Hz, 200Hz and 300Hz no-load voltage on the vibration characteristics of the iron core of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0037] First, this embodiment proposes a transformer abnormal state diagnosis and early warning method based on voiceprint features, which optimizes and smoothes the ultrasonic and vibration voiceprint signals of the transformer in the topological structure, and can realize the recognition and update of transformer voiceprint information in a complex sound field environment, so as to improve the accuracy and robustness of the subsequent fusion diagnosis model. Secondly, the fusion diagnosis processing and analysis method is improved, further improving the effect of transformer abnormal state diagnosis and early warning.

[0038] This embodiment is a multi-dimensional performance optimization method to deal with problems caused by environmental noise, mixed faults and unknown faults. Experimental results show that the proposed condition monitoring and fault diagnosis system can quickly distinguish fault operating conditions in different noise environments and complete transformer abnormal state diagnosis and early warning.

[0039] specifically refer to Figure 1 and Figure 2 , is an embodiment of the present invention, which provides a transformer abnormal state diagnosis and early warning method based on voiceprint features, comprising the following steps:

[0040] Step 1: Obtain the network topology diagram and voiceprint information of the transformer network nodes;

[0041] Step 2: Use the support vector machine algorithm to project the network topology of the transformer network nodes and the classified sample data of the voiceprint information into a high-dimensional space, divide them into category voiceprint information data, and generate a category voiceprint data set;

[0042] Step 3, establishing a transformer abnormal voiceprint data optimization model, wherein the transformer abnormal voiceprint data optimization model and objective function are determined with the maximum processing output of the transformer abnormal voiceprint in the transformer network node network topology diagram as the goal;

[0043] Step 4: Optimize the model and objective function according to the abnormal voiceprint data of the transformer, and use feature smoothing processing to complete the update and output of the voiceprint features of the abnormal state of the transformer, so as to form the voiceprint spectrum feature data of the abnormal state of the transformer;

[0044] Step five, based on the voiceprint feature data of the transformer abnormal state, an abnormal voiceprint feature database of the transformer abnormal state is formed, and then the transformer abnormal state analysis result is generated through similarity analysis based on the abnormal voiceprint feature database. After matching the transformer abnormal state analysis result, the transformer abnormal state fault type detection result is obtained and output.

[0045] Step 6: Fusion pulse current detection method, fusion processing of transformer abnormal state fault type detection results; after matching transformer abnormal state analysis results to obtain transformer abnormal state fault type detection results and output, fusion processing of transformer abnormal state fault type detection results is performed, and the fusion model is as follows:

[0046]

[0047] Among them, G(U,E) represents the fusion data of the transformer abnormal state fault type detection results, U represents the set of unified data of the current transformer abnormal voiceprint fusion processing; E represents the current transformer abnormal voiceprint fusion processing event; U j The unified data representing the current transformer abnormal state fault type, U a Represents the unified data of the abnormal transformer soundprint detected by the current transformer abnormal soundprint pulse current method; w(U j ,E) represents the support of the unified data of the current transformer abnormal state fault type for the current transformer abnormal voiceprint fusion processing event; corr(U j ,U a ) represents the unified data U of the current transformer abnormal state fault type j The unified data U of the transformer abnormal soundprint detected by the current transformer abnormal soundprint pulse current method a of relevance.

[0048] Based on the above, it can be seen that the optimized voiceprint information is fused with the data of other pulse current methods to construct a high-precision and high-robustness diagnostic model. This model can comprehensively analyze various data and provide a more comprehensive transformer status assessment. In addition, the optimized voiceprint features and fusion diagnostic model of this embodiment significantly improve the accuracy and robustness of transformer abnormal status diagnosis, reduce the possibility of false alarms and missed alarms, and at the same time, by improving data processing and analysis methods, make the diagnostic process more efficient and can quickly respond to abnormal conditions of the transformer.

[0049] Specifically, in step 1, when obtaining the network topology diagram and voiceprint information of the transformer network node, the obtained transformer voiceprint information is also subjected to denoising, and the voiceprint information includes the ultrasonic and vibration voiceprint signals of the transformer. The basic principle of ultrasonic detection of transformer winding deformation is as follows: an ultrasonic probe is attached to a certain position on the wall of the transformer housing, and the center of the ultrasonic probe is aligned with the winding to be tested, and the probe is in close contact with the body of the transformer. The ultrasonic transmitting circuit sends a signal synchronously, so that the ultrasonic probe simultaneously transmits ultrasonic waves. Ultrasonic waves propagate in the transformer in a longitudinal wave mode, and reach the transformer winding after passing through the steel wall and transformer oil, and are reflected at the junction of the insulation of the transformer winding and the copper surface. Similarly, the reflected echo passes through the transformer oil and the transformer box wall shell, and reaches the ultrasonic receiving probe through a certain path and generates a receiving electrical pulse signal. The vibration analysis method starts from the mechanical structure characteristics of the transformer winding, and regards it as a mechanical structure composed of mass, stiffness, damping, etc. When any changes occur in the winding structure or force, it can be reflected in its mechanical dynamic characteristics, that is, vibration characteristics.

[0050] Also refer to Figure 2 As shown, where U is the load voltage, U N is the rated voltage, (U / U N ) 2 Indicates the fundamental frequency amplitude. Under no-load conditions, the vibration is measured from the box surface using a piezoelectric acceleration sensor. From the figure, it can be seen that as the no-load voltage increases, the components of each frequency of the vibration signal increase. When the no-load voltage is less than half of the rated voltage, the fundamental frequency 100Hz component of the vibration signal is approximately linearly related to the square of the voltage. When the no-load voltage is higher than half of the rated voltage, a nonlinear trend gradually occurs. The explanation is as follows: the nonlinearity of the magnetostrictive effect of the core causes the vibration to not increase linearly with the square of the magnetic flux when the excitation flux increases; for slightly higher frequency harmonics, it may come from the vibration generated by the electromagnetic attraction between the silicon steel sheets.

[0051] In addition, in step 2 of the present embodiment, the network topology diagram of the transformer network nodes and the classified sample data of the voiceprint information are classified into linear data in a high-dimensional space by using a support vector machine algorithm, and the network topology diagram of the transformer network nodes and the classified sample data of the voiceprint information are divided by finding the optimal classification hyperplane;

[0052] Project the divided sample data back to the source space to realize the division of nonlinear samples, and according to the Mercer theorem of kernel function, where Mercer theorem is an important concept in machine learning and kernel method, it provides a function that can be used as the necessary and sufficient conditions of the kernel function, so that the kernel function that meets the preset conditions can be used as the inner product of the high-dimensional space to replace the inner product of the source space;

[0053] The nonlinear transformation of SVM is transformed into the objective function of quadratic programming to obtain the voiceprint information classification function, and a category voiceprint data set is generated according to the voiceprint information classification function.

[0054] In step 3 of this embodiment, preferably, the maximum processing output of the abnormal voiceprint of the transformer in the transformer network node network topology diagram is taken as the goal to determine the optimization model and objective function of the abnormal voiceprint data of the transformer, which are specifically as follows:

[0055]

[0056] Where, max f(bit) is the maximum amount of processed output data of transformer abnormal voiceprint in the network topology diagram within the preset time, B j is the rated processing capacity data of transformer j, m represents the total number of transformer groups in the network topology, T represents the preset time data, t j represents the processing time of transformer j, is the increased processing capacity data of transformer j.

[0057] This embodiment is particularly optimized for the recognition of incremental transformer voiceprint information in the topology of network nodes in a complex sound field environment, so that it can accurately and efficiently recognize the voiceprint information of the transformer in incremental situations under various environmental conditions.

[0058] Specifically, in this embodiment, the transformer abnormal state fault types include transformer winding fault, transformer core fault, transformer internal discharge fault and transformer tap changer fault;

[0059] Transformer winding faults include turn-to-turn short circuit faults, phase-to-phase short circuit faults, winding strand short circuit faults and winding deformation faults;

[0060] Transformer core faults include multi-point grounding faults and core overheating faults;

[0061] Transformer internal discharge faults include partial discharge faults, spark discharge faults, and arc discharge faults.

[0062] This embodiment further forms an abnormal voiceprint spectrum feature database of the transformer abnormal state based on the voiceprint spectrum feature data of the transformer abnormal state. Specifically, according to the voiceprint spectrum feature data of several transformer abnormal states, combined with a preset voiceprint vibration sensor signal weighting matrix, a weighted MUSIC algorithm is used to calculate the spatial spectrogram of the voiceprint spectrum feature data of several transformer abnormal states, and the spatial spectrogram is converted into a plane image, and then the fault type of the transformer abnormal state is selected and determined.

[0063] In order to further reduce the impact of noise, this embodiment uses a specific algorithm to smooth the voiceprint signal to ensure the stability and reliability of the signal. Specifically, in step 4, the feature smoothing process is shown in the following formula:

[0064]

[0065] Among them, p i Represented as the predicted value of the i-th category, λ i represents the weight value of the i-th category, n is the total number of items in the ith category, r i represents the characteristic parameter value of the i-th category, r n Expressed as the characteristic parameter value of the total category, p n It is expressed as the predicted value of the total category, and cos(·) represents the cosine function.

[0066] Specifically, this embodiment performs similarity analysis based on the abnormal voiceprint spectrum feature database, specifically performs similarity calculation between the voiceprint spectrum feature data to be identified and each abnormal voiceprint spectrum feature sample in the abnormal voiceprint spectrum feature database, and compares the maximum similarity calculation result of the similarity calculation of the voiceprint spectrum feature data to be identified with the preset threshold of each corresponding abnormal voiceprint spectrum feature sample, and matches and outputs the corresponding abnormal voiceprint spectrum feature and the abnormal state of the transformer after comparative analysis. Through the above processing, the present invention can realize real-time identification and updating of transformer voiceprint information, which is crucial for monitoring the operating state of the transformer and predicting potential faults.

[0067] This embodiment optimizes and smoothes the voiceprint features of the ultrasonic and vibration voiceprint signals of the transformer in the topological structure, thereby realizing the recognition and update of transformer voiceprint information in a complex sound field environment, so as to improve the accuracy and robustness of the subsequent fusion diagnosis model, and improves the fusion diagnosis processing and analysis method, further improving the effect of transformer abnormal state diagnosis and early warning.

[0068] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0069] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0070] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0071] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A transformer abnormal state diagnosis and early warning method based on voiceprint features, characterized by: The following steps are involved: Step 1: Obtain the network topology diagram and voiceprint information of the transformer network nodes; Step 2: Use a support vector machine algorithm to project the network topology of the transformer network nodes and the classified sample data of the voiceprint information into a high-dimensional space, divide them into category voiceprint information data, and generate a category voiceprint data set; Step 3, establishing a transformer abnormal voiceprint data optimization model, wherein the transformer abnormal voiceprint data optimization model and objective function are determined with the maximum processing output of the transformer abnormal voiceprint in the transformer network node network topology diagram as the goal; Step 4: Optimizing the model and the objective function according to the abnormal voiceprint data of the transformer, and using feature smoothing processing to complete the update and output of the voiceprint features of the abnormal state of the transformer, so as to form the voiceprint spectrum feature data of the abnormal state of the transformer; Step five, based on the voiceprint feature data of the transformer abnormal state, an abnormal voiceprint feature database of the transformer abnormal state is formed, and then the transformer abnormal state analysis result is generated through similarity analysis according to the abnormal voiceprint feature database, and after matching the transformer abnormal state analysis result, the transformer abnormal state fault type detection result is obtained and output.

2. The transformer abnormal state diagnosis and early warning method based on voiceprint features according to claim 1 is characterized in that: In step 2, the network topology diagram of the transformer network node and the classified sample data of the voiceprint information are classified into linear data in a high-dimensional space by using a support vector machine algorithm, and the network topology diagram of the transformer network node and the classified sample data of the voiceprint information are divided by finding the optimal classification hyperplane; Project the divided sample data back to the source space to realize the division of nonlinear samples, and according to the Mercer theorem of kernel function, use the kernel function that meets the preset conditions as the inner product of the high-dimensional space to replace the inner product of the source space; The nonlinear transformation of SVM is transformed into the objective function of quadratic programming to obtain the voiceprint information classification function, and a category voiceprint data set is generated according to the voiceprint information classification function.

3. The transformer abnormal state diagnosis and early warning method based on voiceprint features according to claim 1 is characterized in that: Taking the maximum processing output of abnormal voiceprint of the transformer in the transformer network node network topology as the goal, the optimization model and objective function of the abnormal voiceprint data of the transformer are determined, as follows: Where, max f(bit) is the maximum amount of processed output data of transformer abnormal voiceprint in the network topology diagram within the preset time, B j is the rated processing capacity data of transformer j, m represents the total number of transformer groups in the network topology, T represents the preset time data, t j represents the processing time of transformer j, is the increased processing capacity data of transformer j.

4. The transformer abnormal state diagnosis and early warning method based on voiceprint features according to claim 1 is characterized in that: The transformer abnormal state fault types include transformer winding fault, transformer core fault, transformer internal discharge fault and transformer tap changer fault; The transformer winding faults include turn-to-turn short circuit faults, phase-to-phase short circuit faults, winding strand short circuit faults and winding deformation faults; The transformer core fault includes multi-point grounding fault and core overheating fault; The transformer internal discharge fault includes a partial discharge fault, a spark discharge fault, and an arc discharge fault.

5. The transformer abnormal state diagnosis and early warning method based on voiceprint features as claimed in claim 4 is characterized in that: Based on the voiceprint spectrum feature data of the abnormal state of the transformer, an abnormal voiceprint spectrum feature database of the abnormal state of the transformer is formed. Specifically, according to the voiceprint spectrum feature data of several abnormal states of the transformer, combined with a preset voiceprint vibration sensor signal weighting matrix, a weighted MUSIC algorithm is used to calculate the spatial spectrograms of the voiceprint spectrum feature data of several abnormal states of the transformer, and the spatial spectrogram is converted into a plane image, and then the fault type of the abnormal state of the transformer is selected and determined.

6. The transformer abnormal state diagnosis and early warning method based on voiceprint features according to claim 1 is characterized in that: In step 4, the feature smoothing process is as follows: Among them, p i Represented as the predicted value of the i-th category, λ i represents the weight value of the i-th category, n is the total number of items in the i-th category, r i represents the characteristic parameter value of the i-th category, r n Expressed as the characteristic parameter value of the total category, p n It is expressed as the predicted value of the total category, and cos(·) represents the cosine function.

7. The transformer abnormal state diagnosis and early warning method based on voiceprint features according to claim 1 is characterized in that: Through similarity analysis based on the abnormal voiceprint feature database, specifically, similarity calculation is performed between the voiceprint feature data to be identified and each abnormal voiceprint feature sample in the abnormal voiceprint feature database, and the maximum similarity calculation result of the similarity calculation of the voiceprint feature data to be identified is compared with the preset threshold of each corresponding abnormal voiceprint feature sample, and the corresponding abnormal voiceprint feature and the abnormal state of the transformer are matched and output after comparative analysis.

8. The transformer abnormal state diagnosis and early warning method based on voiceprint features according to claim 1 is characterized in that: Step 1, when obtaining the network topology diagram and voiceprint information of the transformer network nodes, also includes denoising the obtained transformer voiceprint information.

9. The transformer abnormal state diagnosis and early warning method based on voiceprint features according to claim 1 is characterized in that: After matching the transformer abnormal state analysis results in step 5 to obtain the transformer abnormal state fault type detection results and outputting them, it also includes a fusion pulse current method detection method to fuse the transformer abnormal state fault type detection results. The fusion model is as follows: Among them, G(U,E) represents the fusion data of the transformer abnormal state fault type detection results, U represents the set of unified data of the current transformer abnormal voiceprint fusion processing; E represents the current transformer abnormal voiceprint fusion processing event; U j The unified data representing the current transformer abnormal state fault type, U a Represents the unified data of the abnormal transformer soundprint detected by the current transformer abnormal soundprint pulse current method; w(U j ,E) represents the support of the unified data of the current transformer abnormal state fault type for the current transformer abnormal voiceprint fusion processing event; corr(U j ,U a ) represents the unified data U of the current transformer abnormal state fault type j The unified data U of the transformer abnormal soundprint detected by the current transformer abnormal soundprint pulse current method a of relevance.

Images