阅读说明：本技术 基于声纹图像特征的变压器状态识别方法、系统及介质 (Transformer state identification method, system and medium based on voiceprint image characteristics ) 是由 吴晓文 卢铃 曹浩 胡胜 陈炜 彭继文 吕建红 于 2019-09-09 设计创作，主要内容包括：本发明公开了一种基于声纹图像特征的变压器状态识别方法、系统及介质,本发明识别方法为根据声音信号生成声纹特征图像,提取声纹特征图像中的声纹图像特征信息,将声纹图像特征信息和预设的变压器状态特征识别库进行比对或者将声纹图像特征信息输入预先训练好的人工智能机器学习模型,识别出目标变压器与噪声相关的状态。本发明可利用声纹图像特征识别变压器与噪声相关的异常运行状态,能够显著改善人为判断经验不足、误差大的问题以及现有声学诊断系统特征容易遗漏、准确率低的问题,可大幅提高与噪声有关的变压器运行状态诊断准确率,具有诊断准确率高、诊断速度快的优点。(The invention discloses a transformer state identification method, a system and a medium based on voiceprint image characteristics. The method can identify the abnormal running state of the transformer related to the noise by utilizing the voiceprint image characteristics, can remarkably improve the problems of insufficient manual judgment experience and large error and the problems of easy omission of the characteristics and low accuracy rate of the conventional acoustic diagnosis system, can greatly improve the diagnosis accuracy rate of the running state of the transformer related to the noise, and has the advantages of high diagnosis accuracy rate and high diagnosis speed.)

1. A transformer state identification method based on voiceprint image features is characterized by comprising the following implementation steps:

1) collecting a sound signal of a target transformer in an operating state;

2) generating a voiceprint characteristic image according to the sound signal;

3) extracting voiceprint image characteristic information in the voiceprint characteristic image;

4) comparing the voiceprint image characteristic information with a preset transformer state characteristic recognition library or inputting the voiceprint image characteristic information into a pre-trained artificial intelligence machine learning model to recognize the state of the target transformer related to the noise, wherein the transformer state characteristic recognition library comprises the voiceprint image characteristic information and the mapping relation among different states of the transformer related to the noise, and the artificial intelligence machine learning model is pre-trained to establish the mapping relation among the voiceprint image characteristic information and the different states of the transformer related to the noise.

2. The transformer state identification method based on the voiceprint image characteristics according to claim 1, wherein the voiceprint characteristic image generated in the step 2) is a two-dimensional color image, the horizontal axis and the vertical axis are respectively time and frequency, and the magnitude of the signal amplitude is represented by the shade of the color image.

3. The transformer state identification method based on the voiceprint image characteristics according to claim 1, wherein the extracting of the voiceprint image characteristic information in the voiceprint characteristic image in the step 3) specifically refers to extracting a noise signal frequency spectrum range in the voiceprint characteristic image and concentrated target frequency integral multiple frequency components in the noise signal frequency spectrum range; or extracting the voiceprint image feature information in the voiceprint feature image in the step 3), specifically, extracting the features in the voiceprint feature image by using a convolutional neural network to obtain the voiceprint image feature information.

4. The transformer state identification method based on the voiceprint image characteristics according to claim 1, wherein the detailed step of comparing the voiceprint image characteristic information with a preset transformer state characteristic identification library in the step 4) comprises:

4.1) determining the voltage grade, equipment manufacturer, structure type and cooling mode of the target transformer;

and 4.2) comparing the voiceprint image characteristic information of the target transformer with the voiceprint image characteristic information of the transformer with the same voltage level, equipment manufacturer, structure type and cooling mode in a preset transformer state characteristic identification library and mapping relations between different states related to the transformer and noise, or inputting the voiceprint image characteristic information of the target transformer into an artificial intelligent machine learning model which is trained in advance and corresponds to the transformer with the same voltage level, equipment manufacturer, structure type and cooling mode, and finally identifying the state related to the noise of the target transformer.

5. The transformer state identification method based on the voiceprint image characteristics according to claim 1, wherein the different states of the transformer related to the noise in the step 4) comprise: normal, over-excitation, three-phase unbalanced load, harmonic load, cooling system defect, direct current bias magnet, winding and iron core looseness, accessory looseness and overload.

6. The transformer state identification method based on the voiceprint image characteristics according to any one of claims 1 to 5, wherein a transformer state characteristic identification library is established before the step 3), and the detailed steps comprise:

A1) respectively collecting sound signal samples of transformers with different voltage grades, equipment manufacturers, structure types and cooling modes under different state conditions;

A2) generating a voiceprint characteristic image according to the sound signal sample;

A3) extracting the voice print image characteristic information in the voice print characteristic image,

A4) and according to the voiceprint image characteristic information of the sound signal sample of the transformer with the same voltage grade, equipment manufacturer, structure type and cooling mode, establishing a mapping relation among the voltage grade, the equipment manufacturer, the structure type, the voiceprint image characteristic information of the transformer with the cooling mode and different states related to the noise of the transformer, and finally obtaining a transformer state characteristic identification library formed by the voiceprint image characteristic information of the transformer with all the voltage grades, the equipment manufacturer, the structure type and the cooling mode and the mapping relation among the different states related to the noise of the transformer.

7. The transformer state recognition method based on the voiceprint image features according to any one of claims 1 to 5, wherein before the step 3), a step of training an artificial intelligence machine learning model is performed, and the detailed steps comprise:

B1) respectively collecting sound signal samples of transformers with different voltage grades, equipment manufacturers, structure types and cooling modes under different state conditions;

B2) generating a voiceprint characteristic image according to the sound signal sample;

B3) extracting voiceprint image characteristic information in the voiceprint characteristic image, and adding a state label of a corresponding transformer to the voiceprint image characteristic information to construct a training set and a test set;

B4) respectively establishing an artificial intelligence machine learning model aiming at each transformer with the same voltage grade, equipment manufacturer, structure type and cooling mode, training the artificial intelligence machine learning model through a corresponding training set, and finishing the training of the artificial intelligence machine learning model when the testing accuracy reaches the preset requirement through the testing set.

8. A transformer state identification system based on voiceprint image characteristics is characterized by comprising:

the sound signal acquisition program unit is used for acquiring sound signals of the target transformer in the running state;

an image generation program unit for generating a voiceprint feature image from the sound signal;

a voiceprint image feature extraction program unit for extracting the voiceprint image feature information in the voiceprint feature image,

and the transformer state recognition program unit is used for comparing the voiceprint image characteristic information with a preset transformer state characteristic recognition library or inputting the voiceprint image characteristic information into a pre-trained artificial intelligence machine learning model to recognize the state of the target transformer related to the noise, the transformer state characteristic recognition library comprises the voiceprint image characteristic information and the mapping relation among different states of the transformer related to the noise, and the artificial intelligence machine learning model is pre-trained to establish the mapping relation among the different states of the voiceprint image characteristic information and the transformer related to the noise.

9. A transformer state identification system based on voiceprint image characteristics, comprising a computer device, characterized in that the computer device is programmed or configured to execute the steps of the transformer state identification method based on voiceprint image characteristics as claimed in any one of claims 1 to 7, or a storage medium of the computer device has stored thereon a computer program programmed or configured to execute the transformer state identification method based on voiceprint image characteristics as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, wherein the computer-readable storage medium stores thereon a computer program programmed or configured to execute the transformer state identification method based on voiceprint image features according to any one of claims 1 to 7.

Technical Field

The invention relates to the technical field of power transformer running state evaluation, in particular to a transformer state identification method, a transformer state identification system and a transformer state identification medium based on voiceprint image characteristics.

Background

The operation failure of the power transformer is a key cause of a large-scale power failure accident of a power system. More than 30% of typical fault defects of the transformer (such as overexcitation, three-phase unbalanced load, harmonic load, cooling system defect, direct current magnetic bias, winding and iron core looseness, accessory looseness, overload and the like) can be detected by using noise. The existing fault detection method related to the noise of the transformer mainly depends on that operators judge the fault type by using human ears and experience, or the waveform and the frequency spectrum of a sound signal are used, and a certain number of sound characteristics are extracted from the waveform and the frequency spectrum to be used as a basis for distinguishing different running states. The former mainly depends on the experience of personnel, and has the problems of difficult universal application, high misjudgment probability, low identification accuracy and the like; in the latter, because of more transformer fault types, the extraction of sound features which can obviously distinguish different fault types is very difficult, and the problems of insufficient feature information and low fault diagnosis accuracy rate are easily caused.

Disclosure of Invention

The technical problems to be solved by the invention are as follows: aiming at the problems in the prior art, the invention provides a transformer state identification method, a transformer state identification system and a transformer state identification medium based on voiceprint image characteristics.

In order to solve the technical problems, the invention adopts the technical scheme that:

a transformer state identification method based on voiceprint image characteristics comprises the following implementation steps:

1) collecting a sound signal of a target transformer in an operating state;

2) generating a voiceprint characteristic image according to the sound signal;

3) extracting voiceprint image characteristic information in the voiceprint characteristic image;

4) comparing the voiceprint image characteristic information with a preset transformer state characteristic recognition library or inputting the voiceprint image characteristic information into a pre-trained artificial intelligence machine learning model to recognize the state of the target transformer related to the noise, wherein the transformer state characteristic recognition library comprises the voiceprint image characteristic information and the mapping relation among different states of the transformer related to the noise, and the artificial intelligence machine learning model is pre-trained to establish the mapping relation among the voiceprint image characteristic information and the different states of the transformer related to the noise.

Optionally, the voiceprint feature image generated in step 2) is a two-dimensional color image, and the horizontal axis and the vertical axis are time and frequency, respectively, and the magnitude of the signal amplitude is represented by the shade of the color image.

Optionally, the extracting of the voiceprint image feature information in the voiceprint feature image in step 3) specifically includes extracting a noise signal spectrum range in the voiceprint feature image and concentrated integral multiple frequency components of the target frequency in the noise signal spectrum range; or extracting the voiceprint image feature information in the voiceprint feature image in the step 3), specifically, extracting the features in the voiceprint feature image by using a convolutional neural network to obtain the voiceprint image feature information.

Optionally, the detailed step of comparing the voiceprint image feature information with the preset transformer state feature recognition library in step 4) includes:

4.1) determining the voltage grade, equipment manufacturer, structure type and cooling mode of the target transformer;

and 4.2) comparing the voiceprint image characteristic information of the target transformer with the voiceprint image characteristic information of the transformer with the same voltage level, equipment manufacturer, structure type and cooling mode in a preset transformer state characteristic identification library and mapping relations between different states related to the transformer and noise, or inputting the voiceprint image characteristic information of the target transformer into an artificial intelligent machine learning model which is trained in advance and corresponds to the transformer with the same voltage level, equipment manufacturer, structure type and cooling mode, and finally identifying the state related to the noise of the target transformer.

Optionally, the different states of the transformer related to noise in step 4) include: normal, over-excitation, three-phase unbalanced load, harmonic load, cooling system defect, direct current bias magnet, winding and iron core looseness, accessory looseness and overload.

Optionally, before step 3), a transformer state feature identification library is established, and the detailed steps include:

A1) respectively collecting sound signal samples of transformers with different voltage grades, equipment manufacturers, structure types and cooling modes under different state conditions;

A2) generating a voiceprint characteristic image according to the sound signal sample;

A3) extracting the voice print image characteristic information in the voice print characteristic image,

A4) and according to the voiceprint image characteristic information of the sound signal sample of the transformer with the same voltage grade, equipment manufacturer, structure type and cooling mode, establishing a mapping relation among the voltage grade, the equipment manufacturer, the structure type, the voiceprint image characteristic information of the transformer with the cooling mode and different states related to the noise of the transformer, and finally obtaining a transformer state characteristic identification library formed by the voiceprint image characteristic information of the transformer with all the voltage grades, the equipment manufacturer, the structure type and the cooling mode and the mapping relation among the different states related to the noise of the transformer.

Optionally, the step 3) is preceded by a step of training an artificial intelligence machine learning model, and the detailed steps include:

B1) respectively collecting sound signal samples of transformers with different voltage grades, equipment manufacturers, structure types and cooling modes under different state conditions;

B2) generating a voiceprint characteristic image according to the sound signal sample;

B3) extracting voiceprint image characteristic information in the voiceprint characteristic image, and adding a state label of a corresponding transformer to the voiceprint image characteristic information to construct a training set and a test set;

B4) respectively establishing an artificial intelligence machine learning model aiming at each transformer with the same voltage grade, equipment manufacturer, structure type and cooling mode, training the artificial intelligence machine learning model through a corresponding training set, and finishing the training of the artificial intelligence machine learning model when the testing accuracy reaches the preset requirement through the testing set.

The invention also provides a transformer state identification system based on voiceprint image characteristics, which comprises:

the sound signal acquisition program unit is used for acquiring sound signals of the target transformer in the running state;

an image generation program unit for generating a voiceprint feature image from the sound signal;

a voiceprint image feature extraction program unit for extracting the voiceprint image feature information in the voiceprint feature image,

and the transformer state recognition program unit is used for comparing the voiceprint image characteristic information with a preset transformer state characteristic recognition library or inputting the voiceprint image characteristic information into a pre-trained artificial intelligence machine learning model to recognize the state of the target transformer related to the noise, the transformer state characteristic recognition library comprises the voiceprint image characteristic information and the mapping relation among different states of the transformer related to the noise, and the artificial intelligence machine learning model is pre-trained to establish the mapping relation among the different states of the voiceprint image characteristic information and the transformer related to the noise.

The invention also provides a transformer state identification system based on the voiceprint image characteristics, which comprises a computer device programmed or configured to execute the steps of the transformer state identification method based on the voiceprint image characteristics.

The invention also provides a transformer state identification system based on the voiceprint image characteristics, which comprises a computer device, wherein a storage medium of the computer device is stored with a computer program which is programmed or configured to execute the transformer state identification method based on the voiceprint image characteristics.

The present invention also provides a computer readable storage medium having stored thereon a computer program programmed or configured to execute the voiceprint image feature based transformer state identification method.

Compared with the prior art, the invention has the following advantages: the invention generates a voiceprint characteristic image according to a sound signal, extracts voiceprint image characteristic information in the voiceprint characteristic image, and identifies the state of a target transformer related to noise based on the voiceprint image characteristic information, the invention can identify the abnormal operation state of the transformer related to the noise by utilizing the voiceprint image characteristic, and because the voiceprint characteristic image contains all time domain and frequency domain information of the detected sound signal and the amplitude value of the detected sound signal is differentially displayed by utilizing the color depth, the voiceprint characteristic image can contain all fault information which can be reacted in the signal, and the state of the target transformer related to the noise is identified based on the voiceprint image characteristic information, the problems of insufficient artificial judgment experience and large error and the problems of easy omission of the characteristics and low accuracy rate of the existing acoustic diagnosis system can be obviously improved, and the abnormal operation state of the transformer related to the noise can be diagnosed and classified and identified without depending on the experience of personnel, the method can greatly improve the diagnosis accuracy rate of the transformer running state related to the noise, and has the advantages of high diagnosis accuracy rate and high diagnosis speed.

Drawings

FIG. 1 is a schematic diagram of a basic process of an embodiment of the present invention.

Fig. 2 is a normal transformer voiceprint image in an embodiment of the invention.

Fig. 3 is a transformer voiceprint image in a dc magnetic bias abnormal operation state according to an embodiment of the present invention.

FIG. 4 is a schematic diagram of a basic flow of a second method according to an embodiment of the present invention.

Detailed Description