Machine learning-based rotating machine equipment abnormity identification system and identification method

Abstract

The invention belongs to the technical field of rotating machine equipment abnormity identification, and discloses a rotating machine equipment abnormity identification system and method based on machine learning, and the method comprises the steps: collecting first noise data of rotating machine equipment operation; second noise data of the area where the rotating machine equipment is located are collected; respectively extracting noise frequencies in the first noise data and the second noise data, acquiring the acquisition time of the corresponding noise frequencies, determining environmental noise data according to the acquisition time, and removing the noise data consistent with the noise frequency corresponding to the environmental noise data in the first noise data to obtain the noise data of the rotating machine equipment; inputting the rotating machine equipment noise data into the trained rotating machine equipment state detection model to obtain a state type corresponding to the rotating machine equipment noise data; therefore, the subsequent identification of the state type of the rotating machine equipment is more accurate, the corresponding abnormal state type is more thoroughly solved, and the repeated phenomenon after the abnormal state type is maintained is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of abnormality recognition of transfer equipment, and more particularly to an abnormality recognition system and method of transfer equipment based on machine learning. BACKGROUND

[0002] The prior art such as Chinese patent application with the authorized announcement number CN102520352B discloses an alternating current brushless generator fault diagnosis instrument, which solves the problem that the existing generator fault diagnosis device cannot uniformly monitor and centrally diagnose the early electrical faults of the generator; the invention comprises a sensor group, a signal conditioning circuit, a sampling circuit, a data processor and a man-machine interaction unit, the sensor group is composed of three voltage sensors and four current sensors, the data processor loads a stator winding inter-turn short circuit module, a rotor winding inter-turn short circuit module, an exciter fault diagnosis module and a rotating rectifier fault diagnosis module, performs fault detection and identification processing, and then outputs the results by the man-machine interaction unit. The invention is suitable for ship generators in small spaces, does not need to install various invasive sensors in the generator, can automatically calculate and judge various electrical faults, has good real-time performance and can meet the requirements of online detection, and has universality for different types of generators.

[0003] The prior art still has the following problems: 1. When collecting data, irrelevant or interfering environmental noise may be encountered, and no measures are taken to eliminate these noises, which may result in inaccurate or distorted data, leading to inaccurate defects identified according to the noise; 2. Data anomalies caused by equipment failure, misoperation or other reasons may lead to incorrect judgments of the true state due to the lack of identification and exclusion of these false images, so that the corresponding problems cannot be completely solved, leading to repeated occurrence of the problems after maintenance.

[0004] In view of this, the present application provides an abnormality recognition system and method of transfer equipment based on machine learning to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purposes, the present application provides the following technical solutions: The abnormality recognition method of transfer equipment based on machine learning comprises: Collecting first noise data of the operation of the transfer equipment; Collecting second noise data of the area where the transfer equipment is located; Respectively extracting the noise frequencies in the first noise data and the second noise data, obtaining the collection time corresponding to the noise frequencies, determining the environmental noise data according to the collection time, removing the noise data with the same noise frequency as the environmental noise data in the first noise data, and obtaining the transfer equipment noise data. inputting the transfer equipment noise data into the trained transfer equipment state detection model to obtain a state type corresponding to the transfer equipment noise data; wherein the state type includes a normal state type and an abnormal state type, and the abnormal state type includes a determined abnormal type and an undetermined abnormal type; selectively collecting a power supply network voltage frequency or a transfer equipment output power connected to the transfer equipment according to the undetermined abnormal type; defining the undetermined abnormal type as a self-recoverable abnormality and a non-self-recoverable abnormality based on the power supply network voltage frequency or the transfer equipment output power, and determining an abnormal reason.

[0006] Further, the method for collecting the second noise data comprises: installing the sound collecting device in a ring-shaped distribution around the transfer equipment, and above and below the transfer equipment.

[0007] Further, the method for obtaining the transfer equipment noise data comprises: extracting an i-th noise frequency in the first noise data, and obtaining a collection time T i1 of the i-th noise frequency, extracting an m-th noise frequency in the second noise data, and obtaining a collection time T m2 of the m-th noise frequency; i≥1, m≥1, T i1 is the collection time of the i-th noise frequency in the first noise data, and T m2 is the collection time of the m-th noise frequency in the second noise data; comparing the collection times of the same noise frequencies in the i-th noise frequency and the m-th noise frequency, if T m2 is greater than or equal to T i1 , then marking the noise data of the m-th noise frequency as environmental noise data; if T m2 is less than T i1 , then no marking is performed; removing the noise data having the same noise frequency in the first noise data according to the noise frequency corresponding to the environmental noise data, to obtain the transfer equipment noise data.

[0008] Further, the training method of the transfer equipment state detection model comprises: performing frequency spectrum analysis on the noise data to convert the time domain signal into a frequency domain feature containing three-dimensional information of frequency, amplitude and time; expanding the data set by using a data enhancement technology: obtaining an expanded data set by adding a random Gaussian white noise to the original noise data, performing frequency offset and time shift transformation; setting a corresponding digital label for each group of processed sound frequency features to form an initial training data pair; The data set containing the K-group initial training data pairs is divided into a training set and a test set according to a preset ratio, and E% of the training set is reserved as a validation set; stratified sampling training, testing and validation are performed on each state type; The sound frequencies in the training set are taken as inputs of the transfer device state detection model, the digital labels corresponding to the sound frequencies in the training set are taken as outputs, the transfer device state detection model is trained, an initial transfer device state detection model is obtained, the initial transfer device state detection model is tested by using the test set, and an initial transfer device state detection model meeting a preset accuracy is output as a test transfer device state detection model; when the loss value of the test transfer device state detection model is not decreased for L consecutive times, the current test transfer device state detection model is taken as the transfer device state detection model; the transfer device state detection model is a CNN-RNN hybrid neural network, the output layer adopts a softmax activation function, the prediction probability of each state type is output, and a confidence threshold is set; when the highest probability is not lower than the confidence threshold, the determined state type is output; when the highest probability is lower than the confidence threshold, the uncertain abnormal type is determined; the loss function of the model is calculated by using a weighted cross-entropy loss; Dropout is added in the full connection layer, and L2 regularization processing is used for the convolution layer.

[0009] Further, the determined abnormal type includes a short circuit and an insulation fault of the transfer device; the uncertain abnormal type includes core abnormal vibration, winding abnormal vibration and cooling oil leakage fault; According to the uncertain abnormal type, the voltage frequency of the power supply grid connected with the transfer device or the output power of the transfer device is selectively collected, and the selective collection method includes: When the uncertain abnormal type is core abnormal vibration or winding abnormal vibration, the voltage frequency of the power supply grid connected with the transfer device is collected. When the uncertain abnormal type is a cooling oil leakage fault, the output power of the transfer device is collected.

[0010] Further, the method for defining the uncertain abnormal type as a self-recoverable abnormality and a non-self-recoverable abnormality includes: When the uncertain abnormal type is core abnormal vibration or winding abnormal vibration, if the absolute value of the difference between the voltage frequency of the power supply grid and the inherent frequency of the transfer device is less than or equal to a frequency threshold, the uncertain abnormal type is defined as a self-recoverable abnormality. If the absolute value of the difference between the voltage frequency of the power supply grid and the inherent frequency of the transfer device is greater than the frequency threshold, the uncertain abnormal type is defined as a non-self-recoverable abnormality. When the uncertain abnormal type is a cooling oil leakage fault, the cooling oil leakage fault is defined as a non-self-recoverable abnormality.

[0011] Further, the method for determining the abnormal reason includes: When the type of abnormality is uncertain, such as abnormal vibration of the iron core or abnormal vibration of the winding: If the absolute value of the difference between the power grid voltage frequency and the inherent frequency of the switching equipment is less than or equal to the frequency threshold, the cause of the abnormality is determined to be the power grid voltage frequency, and a power grid voltage frequency adjustment command is generated and sent to the power grid management terminal. If the absolute value of the difference between the power grid voltage frequency and the inherent frequency of the rotating equipment is greater than the frequency threshold, the cause of the abnormality is determined to be material aging, insulation aging or loose fasteners of the iron core winding, and a material maintenance command is generated and sent to the rotating equipment maintenance terminal. When the type of abnormality is uncertain and is classified as a cooling oil leak: When the output power of the switching equipment is greater than or equal to the rated output power of the switching equipment, the cause of the abnormality is determined to be excessive power demand in the power consumption area. A voltage divider command is generated and sent to the power supply network management terminal. When the output power of the rotating equipment is less than the rated output power of the rotating equipment, the cause of the abnormality is determined to be a cooling system abnormality, a cooling system abnormality command is generated and sent to the rotating equipment maintenance terminal.

[0012] The machine learning-based transfer equipment anomaly identification system implements the aforementioned machine learning-based transfer equipment anomaly identification method, including: The first data acquisition module is used to collect the first noise data of the rotating equipment during operation; The second data acquisition module collects the second noise data of the area where the transfer equipment is located; The environmental noise processing module extracts the noise frequencies from the first noise data and the second noise data respectively, obtains the acquisition time of the corresponding noise frequencies, determines the environmental noise data based on the acquisition time, removes the noise data in the first noise data that has the same noise frequency as the environmental noise data, and obtains the noise data of the rotating equipment. The noise recognition module inputs the noise data of the transfer equipment into the trained transfer equipment status detection model to obtain the status type corresponding to the noise data of the transfer equipment; among which, the status type includes normal status type and abnormal status type, and the abnormal status type includes definite abnormal type and uncertain abnormal type. The third data acquisition module selectively acquires the voltage and frequency of the power grid connected to the transfer equipment or the output power of the transfer equipment based on the type of uncertain anomaly. The deep analysis module, based on the power grid voltage and frequency or the output power of the switching equipment, defines uncertain anomaly types as self-recoverable anomalies and non-self-recoverable anomalies, and determines the cause of the anomaly.

[0013] An electronic device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the machine learning-based transfer equipment abnormality identification method when executing the computer program.

[0014] A computer-readable storage medium having a computer program stored thereon, the computer program being executable by a processor to implement the machine learning-based transfer equipment abnormality identification method.

[0015] The technical effects and advantages of the machine learning-based transfer equipment abnormality identification system and method of the present application are as follows: The environmental noise data is determined through the receiving time of the first and second noise data, and the noise frequency corresponding to the environmental noise data is extracted, and then the noise data with the same noise frequency in the first noise data is removed according to the noise frequency corresponding to the environmental noise data, so as to obtain the real transfer equipment noise data, thereby making the subsequent identification of the transfer equipment state type more accurate.

[0016] Secondly, the abnormal state types included in the transfer equipment state type are further analyzed and determined to obtain self-recoverable abnormality and non-self-recoverable abnormality, and the abnormal reasons are determined according to the self-recoverable abnormality and the non-self-recoverable abnormality, so as to more targetedly provide a maintenance scheme, thereby more completely solving the corresponding abnormal state type and reducing the phenomenon of repeated occurrence after maintenance of the abnormal state type. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The figure is a schematic diagram of the system module principle of the machine learning-based transfer equipment abnormality identification system of embodiment 1 of the present application; Figure 2 The figure is a schematic diagram of the installation position of the second data acquisition module in embodiment 1 of the present application; Figure 3 The figure is a schematic diagram of the machine learning-based transfer equipment abnormality identification method flow of embodiment 2 of the present application.

[0018] REFERENCE NUMERALS 1, sound acquisition device. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] Embodiment 1 Please refer to Figure 1As shown, the machine learning-based abnormality identification system for transfer equipment in the embodiment comprises a first data acquisition module, a second data acquisition module, an environmental noise processing module, a third data acquisition module, a noise identification module, and a deep analysis module. Each module is connected through wired and / or wireless means to realize data transmission.

[0021] The first data acquisition module is used to acquire first noise data of the transfer equipment in operation. The first data acquisition module is arranged close to the transfer equipment. The first noise data can be acquired by a patch-type noise acquisition device arranged on the surface of the transfer equipment.

[0022] The second data acquisition module is used to acquire second noise data of the area where the transfer equipment is located. The method for acquiring the second noise data comprises: Please refer to Figure 2 As shown, the sound acquisition device 1 is arranged in a ring shape around the transfer equipment, above and below the transfer equipment, so as to capture the second noise data generated around the transfer equipment. The sound acquisition device 1 is, for example, a sound sampler, a microphone, or the like.

[0023] The environmental noise processing module is used to extract noise frequencies in the first noise data and the second noise data, respectively, and acquire the acquisition time of the corresponding noise frequencies. The environmental noise data is determined according to the acquisition time. The noise data having the same noise frequency as the environmental noise data in the first noise data is removed, and the transfer equipment noise data is obtained.

[0024] The method for determining the transfer equipment noise data comprises: The i-th noise frequency in the first noise data is extracted, and the acquisition time T i1 of the i-th noise frequency is acquired. m2

[0025] The acquisition times of the same noise frequencies in the i-th noise frequency and the m-th noise frequency are compared. If T m2 is greater than or equal to T i1 , the noise data of the m-th noise frequency is marked as environmental noise data. If T m2 is less than T i1 , no marking is performed. i≥1, m≥1, T i1 is the acquisition time of the i-th noise frequency in the first noise data, and T m2 is the acquisition time of the m-th noise frequency in the second noise data.

[0026] The noise data having the same noise frequency as the environmental noise data in the first noise data is removed according to the noise frequency corresponding to the environmental noise data, and the transfer equipment noise data is obtained. ​

[0027] Since the sound collection device 1 is installed in a ring-shaped distribution around the transfer device and above and below the transfer device, the sound collection device 1 can obtain environmental noise data faster than a patch noise acquisition device on the surface of the transfer device.

[0028] Secondly, it needs to be explained that when T m2 is equal to T i1 , it means that the noise is generated between the first data collection module and the second data collection module, so this noise is not the noise generated by the operation of the transfer device, that is, the environmental noise data.

[0029] The method for removing noise data with the same noise frequency in the first noise data according to the noise frequency corresponding to the environmental noise data comprises: S1, extracting the noise frequency corresponding to the environmental noise data.

[0030] S2, using a multi-track audio editor, removing noise data with the same noise frequency in the first noise data according to the noise frequency corresponding to the environmental noise data.

[0031] The specific method of using a multi-track audio editor (Audacity) is: find the noise reduction function through the path of

Effects

Noise Removal and Repair

[0032] The noise recognition module inputs the transfer device noise data into the trained transfer device state detection model to obtain the state type corresponding to the transfer device noise data; wherein the state type includes a normal state type and an abnormal state type, the abnormal state type includes a determined abnormal type and an undetermined abnormal type, the determined abnormal type includes a short circuit and an insulation fault of the transfer device, and the undetermined abnormal type includes core abnormal vibration, winding abnormal vibration and cooling oil leakage fault.

[0033] Wherein, the short circuit is the noise caused by the abnormal increase of current when the short circuit occurs between the windings or between the windings and the ground of the transfer device, arc discharge, current surge or winding vibration; the insulation fault is the noise caused by discharge when the insulation strength decreases due to damage or aging of the insulation material of the transfer device.

[0034] Wherein, in the use process of the rotating machine equipment (such as motor, generator), the causes of abnormal vibration of the core and abnormal vibration of the winding are aging of the core winding material, aging of the insulation and loosening of the fastener, etc., and due to the fact that the power supply network voltage frequency is close to or matches the inherent frequency of the rotating machine equipment, resonance phenomenon occurs, which also causes abnormal vibration of the core and abnormal vibration of the winding to produce noise, therefore, this type of noise may be abnormal noise caused by the fault of the rotating machine equipment itself (i.e. aging of the core winding material, aging of the insulation and loosening of the fastener, etc.), or abnormal noise caused by the rotating machine equipment affected by the power supply network voltage frequency; if it is abnormal noise caused by the power supply network voltage frequency, it is self-recoverable abnormality, if not, it is not self-recoverable abnormality, and maintenance or replacement of the core winding is required.

[0035] Cooling oil is an important component of the cooling system of the rotating machine equipment, when the pipeline is broken, the cooling effect is reduced or completely fails, the cooling effect is reduced, when the pipeline is broken, the fluid (such as gas or liquid) will be ejected from the broken port at a high speed, forming a jet sound, this noise is usually sharp and high frequency, similar to the sound of water or air jet; there are many reasons for the pipeline breakage, such as natural damage of the cooler, in this case, the damaged parts can be replaced, it is also possible that the load of the power supply area of the rotating machine equipment increases, exceeds the rated load of the rotating machine equipment, the cooling effect cannot keep up, too much heat is generated inside the rotating machine equipment, steam or gas is generated, if the cooling system cannot effectively discharge these gases, the pipeline is expanded and broken to cause cooling oil leakage, in this case, only the damaged parts are replaced, then the problem cannot be completely solved, the problem still occurs after replacement, the reason is that the cause of the problem is not accurately grasped.

[0036] The training method of the rotating machine equipment state detection model comprises: Performing frequency spectrum analysis on the noise data to convert the time domain signal into frequency domain features containing three-dimensional information of frequency, amplitude and time; Expanding the data set by using data enhancement technology: by adding random Gaussian white noise to the original noise data, frequency offset and time shift transformation are performed to obtain an expanded data set; Setting corresponding digital labels for each group of processed sound frequency features to form an initial training data pair; Dividing the data set containing K initial training data pairs into a training set and a test set according to a predetermined proportion, and reserving E% of the training set as a validation set; stratified sampling training, testing and validation are performed on each state type; The sound frequencies in the training set are taken as inputs of the transfer equipment state detection model, and the digital labels corresponding to the sound frequencies in the training set are taken as outputs, the transfer equipment state detection model is trained, an initial transfer equipment state detection model is obtained, the initial transfer equipment state detection model is tested by using the test set, and the initial transfer equipment state detection model meeting the preset accuracy is output as a test transfer equipment state detection model; when the continuous L times loss values of the test transfer equipment state detection model verified by the verification set do not decrease, the current test transfer equipment state detection model is taken as the transfer equipment state detection model, and the transfer equipment state detection model is a CNN-RNN hybrid neural network, wherein a softmax activation function is used in an output layer to output prediction probabilities of each state type, and a confidence threshold is set; when the highest probability is not lower than the confidence threshold, a determined state type is output; when the highest probability is lower than the confidence threshold, an undetermined abnormal type is determined; a weighted cross-entropy loss is used to calculate the loss function of the model; Dropout is added in a fully connected layer, and L2 regularization processing is used for the convolution layer.

[0037] The training method of the transfer equipment state detection model extracts frequency domain features containing three-dimensional information of frequency, amplitude and time by performing frequency spectrum analysis on noise data, thereby providing an effective input basis for the model; data augmentation techniques such as adding random Gaussian white noise, frequency offset and time shift transformation are used to expand the data set and enhance the adaptability of the model to complex environments and working conditions; the data set is divided into a training set, a test set and a verification set through stratified sampling, so as to ensure the rationality of data distribution and avoid model bias; the CNN-RNN hybrid neural network is used to combine the advantages of the convolution layer in extracting local key features and the recurrent layer in capturing time dynamic changes, so as to adapt to the complex characteristics of the noise features of the transfer equipment; the weighted cross-entropy loss function is used to handle the sample imbalance problem to improve the recognition ability of rare abnormal types, Dropout is added in the fully connected layer and L2 regularization is used for the convolution layer to suppress overfitting, the confidence threshold is set to distinguish between the determined state type and the undetermined abnormal type to solve the problem of ambiguous abnormal type recognition, and the early stopping mechanism is used to obtain the optimal model, thereby improving the recognition accuracy, generalization ability and stability of the model for various states of the transfer equipment, effectively solving the technical problems of large environmental interference, inaccurate abnormal type determination and weak model adaptability in traditional detection, and achieving accurate and reliable detection of the state of the transfer equipment.

[0038] The third data acquisition module is configured to selectively acquire the power supply network voltage frequency connected to the transfer equipment or the output power of the transfer equipment according to the undetermined abnormal type.

[0039] The power supply network voltage frequency can be measured by a frequency meter, and the output power of the transfer equipment can be obtained by multiplying the voltage and current at the output end of the transfer equipment.

[0040] According to the uncertain abnormal type, the power supply network voltage frequency connected with the rotating machine device or the rotating machine device output power is selectively collected, and the selective collection method comprises: When the uncertain abnormal type is the core abnormal vibration or the winding abnormal vibration, the power supply network voltage frequency connected with the rotating machine device is collected.

[0041] When the uncertain abnormal type is the cooling oil leakage fault, the rotating machine device output power is collected.

[0042] The deep analysis module defines the uncertain abnormal type as the self-recovery abnormality and the non-self-recovery abnormality based on the power supply network voltage frequency or the rotating machine device output power, and determines the abnormal reason.

[0043] The method for defining the uncertain abnormal type as the self-recovery abnormality and the non-self-recovery abnormality is as follows: When the uncertain abnormal type is the core abnormal vibration or the winding abnormal vibration, if the absolute value of the difference between the power supply network voltage frequency and the inherent frequency of the rotating machine device is less than or equal to the frequency threshold value, the uncertain abnormal type is defined as the self-recovery abnormality, and the rotating machine device can be restored to normal by changing the power supply network voltage frequency; the frequency threshold value is a value causing the resonance of the rotating machine device to cause noise, which is determined by a person skilled in the art according to a large number of experiments.

[0044] If the absolute value of the difference between the power supply network voltage frequency and the inherent frequency of the rotating machine device is greater than the frequency threshold value, the uncertain abnormal type is defined as the non-self-recovery abnormality, which indicates that the core abnormal vibration and the winding abnormal vibration at this time are not caused by the power supply network voltage frequency, but are caused by the aging of the core winding material, the aging of the insulation, the loosening of the fastener and the like.

[0045] When the uncertain abnormal type is the cooling oil leakage fault, the cooling oil leakage fault is defined as the non-self-recovery abnormality, and the leakage point needs to be repaired or the corresponding part needs to be replaced, and further analysis and determination of the cause of the cooling oil leakage fault are required.

[0046] The method for determining the abnormal reason is as follows: When the uncertain abnormal type is the core abnormal vibration or the winding abnormal vibration: If the absolute value of the difference between the power supply network voltage frequency and the inherent frequency of the rotating machine device is less than or equal to the frequency threshold value, the abnormal reason is determined to be the power supply network voltage frequency, and a power supply network voltage frequency adjustment instruction is generated and sent to the power supply network management end, and the power supply network management end adjusts the power supply network voltage frequency of the power supply network input to the rotating machine device according to the power supply network voltage frequency adjustment instruction, so that the power supply network voltage frequency matches the inherent frequency of the rotating machine device.

[0047] If the absolute value of the difference between the power supply network voltage frequency and the inherent frequency of the rotating machine device is greater than the frequency threshold, it is determined that the abnormal reason is the aging of the material of the core winding, the aging of the insulation or the loosening of the fastener, a material maintenance instruction is generated and sent to the rotating machine device maintenance end, and the maintenance personnel go to the rotating machine device for maintenance or replace the corresponding material.

[0048] When the abnormal type is uncertain to be a cooling oil leakage fault: When the output power of the rotating machine device is greater than or equal to the rated output power of the rotating machine device, it is determined that the abnormal reason is that the power demand of the power consumption area is too large, a voltage division instruction is generated and sent to the power supply network management end, and the power supply network management end calls other rotating machine devices to supply power to the power consumption area according to the voltage division instruction to avoid the cooling oil leakage fault from occurring again.

[0049] When the output power of the rotating machine device is less than the rated output power of the rotating machine device, it is determined that the abnormal reason is a cooling system abnormal instruction, which is sent to the rotating machine device maintenance end, and the maintenance personnel go to the rotating machine device for plugging or replacing the corresponding parts.

[0050] The embodiment determines the environmental noise data through the receiving time of the first and second noise data, extracts the noise frequency corresponding to the environmental noise data, and removes the noise data with the same noise frequency in the first noise data according to the noise frequency corresponding to the environmental noise data, to obtain the real rotating machine device noise data, so that the subsequent identification of the rotating machine device state type is more accurate.

[0051] Secondly, the abnormal state types included in the rotating machine device state type are further analyzed and determined to obtain self-recoverable abnormalities and non-self-recoverable abnormalities, and the abnormal reasons are determined according to the self-recoverable abnormalities and the non-self-recoverable abnormalities, so that the maintenance scheme is more targeted, thereby more thoroughly solving the corresponding abnormal state type and reducing the phenomenon of repeated occurrence after maintenance of the abnormal state type.

[0052] Embodiment 2 Please refer to Figure 3 The embodiment provides a rotating machine device abnormality identification method based on machine learning, which comprises: Collecting first noise data of the rotating machine device; Collecting second noise data of the area where the rotating machine device is located; Respectively extracting the noise frequencies in the first noise data and the second noise data, and obtaining the collection time corresponding to the noise frequencies, determining the environmental noise data according to the collection time, removing the noise data in the first noise data with the same noise frequency as the environmental noise data, and obtaining the rotating machine device noise data; The transfer equipment noise data is input into the trained transfer equipment state detection model to obtain a state type corresponding to the transfer equipment noise data; wherein, the state type includes a normal state type and an abnormal state type, and the abnormal state type includes a determined abnormal type and an undetermined abnormal type; According to the undetermined abnormal type, the power supply network voltage frequency connected to the transfer equipment or the output power of the transfer equipment is selectively collected; Based on the power supply network voltage frequency or the output power of the transfer equipment, the undetermined abnormal type is defined as a self-recoverable abnormality and a non-self-recoverable abnormality, and the abnormal reason is determined.

[0053] Further, the method for collecting the second noise data comprises: The sound collecting device 1 is installed in a ring-shaped distribution around the transfer equipment, and above and below the transfer equipment.

[0054] Further, the method for obtaining the transfer equipment noise data comprises: The i-th noise frequency in the first noise data is extracted, and the collection time T i1 of the i-th noise frequency is obtained m2 ; i≥1, m≥1, T i1 is the collection time of the i-th noise frequency in the first noise data, and T m2 is the collection time of the m-th noise frequency in the second noise data; The collection times of the same noise frequencies in the i-th noise frequency and the m-th noise frequency are compared, if T m2 is greater than or equal to T i1 , the noise data of the m-th noise frequency is marked as environmental noise data; if T m2 is less than T i1 , no marking is performed; According to the noise frequency corresponding to the environmental noise data, the noise data with the same noise frequency in the first noise data is removed to obtain the transfer equipment noise data.

[0055] Further, the training method of the transfer equipment state detection model comprises: The noise data is subjected to frequency spectrum analysis to convert the time domain signal into a frequency domain feature containing three-dimensional information of frequency, amplitude and time; A data enhancement technique is used to expand the data set: by adding random Gaussian white noise to the original noise data, frequency offset and time shift transformation are performed to obtain an expanded data set; A corresponding digital label is set for each group of processed sound frequency features to form an initial training data pair; The data set containing the K-group initial training data pairs is divided into a training set and a test set according to a preset ratio, and E% of the training set is reserved as a validation set; stratified sampling training, testing and validation are performed on each state type; The sound frequencies in the training set are taken as inputs of the transfer device state detection model, and the digital labels corresponding to the sound frequencies in the training set are taken as outputs, the transfer device state detection model is trained, an initial transfer device state detection model is obtained, the initial transfer device state detection model is tested by using the test set, and an initial transfer device state detection model meeting a preset accuracy is output as a test transfer device state detection model; when the loss value of the test transfer device state detection model is not decreased for L consecutive times, the current test transfer device state detection model is taken as the transfer device state detection model; the transfer device state detection model is a CNN-RNN hybrid neural network, wherein a softmax activation function is used in the output layer to output the prediction probability of each state type, and a confidence threshold is set; when the highest probability is not lower than the confidence threshold, the determined state type is output; when the highest probability is lower than the confidence threshold, the uncertain abnormal type is determined; the loss function of the model is calculated by using a weighted cross-entropy loss; Dropout is added in the full connection layer, and L2 regularization processing is used for the convolution layer.

[0056] Further, the determined abnormal type includes a transfer device short circuit and an insulation fault; the uncertain abnormal type includes core abnormal vibration, winding abnormal vibration and cooling oil leakage fault; According to the uncertain abnormal type, the power supply network voltage frequency connected to the transfer device or the output power of the transfer device is selectively collected, and the selective collection method includes: When the uncertain abnormal type is core abnormal vibration or winding abnormal vibration, the power supply network voltage frequency connected to the transfer device is collected. When the uncertain abnormal type is a cooling oil leakage fault, the output power of the transfer device is collected.

[0057] Further, the method for defining the uncertain abnormal type as self-recoverable abnormality and non-self-recoverable abnormality includes: When the uncertain abnormal type is core abnormal vibration or winding abnormal vibration, if the absolute value of the difference between the power supply network voltage frequency and the inherent frequency of the transfer device is less than or equal to a frequency threshold, the uncertain abnormal type is defined as self-recoverable abnormality. If the absolute value of the difference between the power supply network voltage frequency and the inherent frequency of the transfer device is greater than the frequency threshold, the uncertain abnormal type is defined as non-self-recoverable abnormality. When the uncertain abnormal type is a cooling oil leakage fault, the cooling oil leakage fault is defined as non-self-recoverable abnormality.

[0058] Further, the method for determining the abnormal reason includes: When the abnormal type is uncertain to be core abnormal vibration or winding abnormal vibration: If the absolute value of the difference between the power supply network voltage frequency and the inherent frequency of the rotating machine device is less than or equal to the frequency threshold, it is determined that the abnormal cause is the power supply network voltage frequency, and a power supply network voltage frequency adjustment instruction is generated and sent to the power supply network management end; If the absolute value of the difference between the power supply network voltage frequency and the inherent frequency of the rotating machine device is greater than the frequency threshold, it is determined that the abnormal cause is the material aging, insulation aging or fastener loosening of the core winding, and a material maintenance instruction is generated and sent to the rotating machine device maintenance end; When the abnormal type is uncertain to be cooling oil leakage fault: When the output power of the rotating machine device is greater than or equal to the rated output power of the rotating machine device, it is determined that the abnormal cause is the excessive power demand of the power consumption area, and a voltage division instruction is generated and sent to the power supply network management end; When the output power of the rotating machine device is less than the rated output power of the rotating machine device, it is determined that the abnormal cause is the cooling system abnormality, and a cooling system abnormality instruction is generated and sent to the rotating machine device maintenance end.

[0059] Embodiment 3 The present application also provides an electronic device. The electronic device can include one or more processors and one or more memories. The memory stores computer readable code which, when executed by the one or more processors, can perform the machine learning-based rotating machine device abnormality identification method as described above.

[0060] The electronic device can include a bus, one or more CPUs, read-only memory (ROM), random access memory (RAM), a communication port connected to a network, input / output, a hard disk, etc. The storage device in the electronic device, such as ROM or hard disk, can store the machine learning-based rotating machine device abnormality identification method provided by the present application. Further, the electronic device can also include a user interface. Of course, the architecture disclosed in the present application is only exemplary, and when implementing different devices, one or more components of the electronic device disclosed in the present application can be omitted according to actual needs.

[0061] Embodiment 4 One embodiment of the present application discloses a computer readable storage medium. The computer readable storage medium stores computer readable instructions. When the computer readable instructions are executed by a processor, one of the embodiments of the present application can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory can include random access memory (RAM) and cache memory, etc. The non-volatile memory can include read-only memory (ROM), hard disk, flash memory, etc.

[0062] In addition, according to the embodiments of the present application, the above disclosed machine learning based transfer equipment anomaly identification method can be implemented as a computer software program. For example, the present application provides a non-transitory machine readable storage medium storing machine readable instructions executable by a processor to perform instructions corresponding to the method steps provided by the present application, for example: a machine learning based transfer equipment anomaly identification method.

[0063] The above describes only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0064] Finally: the above described only a preferred embodiment of the present application, and is not intended to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included within the protection scope of the present application.

Claims

1. A method for identifying abnormality of a transfer device based on machine learning, characterized by, The method comprises the following steps: Collecting first noise data of the operation of the transfer equipment; Collecting second noise data of the area where the transfer equipment is located; Respectively extracting noise frequencies in the first noise data and the second noise data, and obtaining the collection time corresponding to the noise frequencies, determining the environmental noise data according to the collection time, removing the noise data in the first noise data with the same noise frequency as the environmental noise data, and obtaining the transfer equipment noise data; Inputting the transfer equipment noise data into the trained transfer equipment state detection model to obtain the state type corresponding to the transfer equipment noise data; wherein the state type includes a normal state type and an abnormal state type, and the abnormal state type includes a determined abnormal type and an undetermined abnormal type; According to the undetermined abnormal type, the power supply network voltage frequency or the output power of the transfer equipment connected to the transfer equipment is selectively collected; Based on the power supply network voltage frequency or the output power of the transfer equipment, the undetermined abnormal type is defined as a self-recoverable abnormality and a non-self-recoverable abnormality, and the abnormal reason is determined. 2.The machine learning-based transfer equipment abnormality recognition method of claim 1, wherein, The method for collecting the second noise data comprises: The sound collecting device is installed in a ring-shaped distribution around the transfer equipment, and above and below the transfer equipment. 3.The machine learning-based transfer equipment abnormality recognition method of claim 2, wherein, The method for obtaining the transfer equipment noise data comprises: extracting the i-th noise frequency in the first noise data, and obtaining the collection time T of the i-th noise frequency i1 extracting the m-th noise frequency in the second noise data, and obtaining the collection time T of the m-th noise frequency m2 ; i≥1, m≥1, T i1 is the collection time of the i-th noise frequency in the first noise data, T m2 is the collection time of the m-th noise frequency in the second noise data; If T m2 is greater than or equal to T i1 , the noise data of the mth noise frequency is marked as environmental noise data; if T m2 is less than T i1 , no marking is performed. According to the noise frequency corresponding to the environmental noise data, the noise data with the same noise frequency in the first noise data is removed to obtain the transfer equipment noise data. 4.The machine learning-based transfer equipment abnormality recognition method of claim 3, wherein, The training method of the transfer equipment state detection model comprises: Performing frequency spectrum analysis on the noise data to convert the time domain signal into a frequency domain feature containing frequency, amplitude and time three-dimensional information; Using data enhancement technology to expand the data set: by adding random Gaussian white noise to the original noise data, frequency offset and time shift transformation are performed to obtain an expanded data set; Setting a corresponding digital label for each set of processed sound frequency features to form an initial training data pair; Dividing the data set containing K initial training data pairs into a training set and a test set according to a predetermined proportion, while reserving E% of the training set as a validation set; stratified sampling training, testing and validation are performed on each state type; The sound frequencies in the training set are taken as inputs of the transfer equipment state detection model, and the digital labels corresponding to the sound frequencies in the training set are taken as outputs, the transfer equipment state detection model is trained, an initial transfer equipment state detection model is obtained, the initial transfer equipment state detection model is tested by using the test set, and the initial transfer equipment state detection model meeting a preset accuracy is output as a test transfer equipment state detection model; when the continuous L loss values of the test transfer equipment state detection model verified by the verification set do not decrease, the current test transfer equipment state detection model is taken as the transfer equipment state detection model; the transfer equipment state detection model is a CNN-RNN hybrid neural network, wherein, a softmax activation function is used in an output layer to output prediction probabilities of each state type, and a confidence threshold is set; when the highest probability is not lower than the confidence threshold, a determined state type is output; when the highest probability is lower than the confidence threshold, an undetermined abnormal type is determined; a weighted cross-entropy loss is used to calculate the loss function of the model; Dropout is added in a fully connected layer, and L2 regularization is used to process the convolutional layer. 5.The machine learning-based transfer equipment abnormality recognition method of claim 4, wherein, The determined abnormal type includes a short circuit and an insulation fault of the transfer equipment; and the undetermined abnormal type includes core abnormal vibration, winding abnormal vibration and cooling oil leakage fault. The selective collection method includes the following steps: When the undetermined abnormal type is core abnormal vibration or winding abnormal vibration, the power grid voltage frequency connected with the transfer equipment is collected; When the undetermined abnormal type is cooling oil leakage fault, the output power of the transfer equipment is collected. 6.The machine learning based transfer equipment abnormality recognition method of claim 5, wherein, The method for defining the undetermined abnormal type as self-recoverable abnormality and non-self-recoverable abnormality includes the following steps: When the undetermined abnormal type is core abnormal vibration or winding abnormal vibration, if the absolute value of the difference between the power grid voltage frequency and the inherent frequency of the transfer equipment is less than or equal to a frequency threshold, the undetermined abnormal type is defined as self-recoverable abnormality; If the absolute value of the difference between the power grid voltage frequency and the inherent frequency of the transfer equipment is greater than the frequency threshold, the undetermined abnormal type is defined as non-self-recoverable abnormality; When the undetermined abnormal type is cooling oil leakage fault, the cooling oil leakage fault is defined as non-self-recoverable abnormality. 7.The machine learning-based transfer equipment abnormality recognition method of claim 6, wherein, The method for determining the abnormal cause includes the following steps: When the undetermined abnormal type is core abnormal vibration or winding abnormal vibration: If the absolute value of the difference between the power grid voltage frequency and the inherent frequency of the transfer equipment is less than or equal to a frequency threshold, the abnormal cause is determined to be the power grid voltage frequency, and a power grid voltage frequency adjustment instruction is generated and sent to a power grid management end; If the absolute value of the difference between the power grid voltage frequency and the inherent frequency of the transfer equipment is greater than the frequency threshold, the abnormal cause is determined to be material aging, insulation aging or fastener loosening of the core winding, and a material maintenance instruction is generated and sent to a transfer equipment maintenance end; When the undetermined abnormal type is cooling oil leakage fault: When the output power of the transfer equipment is greater than or equal to the rated output power of the transfer equipment, the abnormal cause is determined to be excessive power demand of a power consumption area, and a voltage division instruction is generated and sent to the power grid management end; When the output power of the transfer device is less than the rated output power of the transfer device, it is determined that the abnormal cause is a cooling system abnormality, a cooling system abnormality instruction is generated, and the instruction is sent to a transfer device maintenance end.

8. A system for identifying abnormality of a transfer equipment based on machine learning, implementing the method for identifying abnormality of a transfer equipment based on machine learning according to any one of claims 1 to 7, characterized in that, The method comprises the following steps: A first data acquisition module is configured to acquire first noise data of the transfer device in operation. A second data acquisition module is configured to acquire second noise data of an area in which the transfer device is located. An environmental noise processing module is configured to extract noise frequencies from the first noise data and the second noise data, respectively, and to obtain collection times corresponding to the noise frequencies, to determine environmental noise data based on the collection times, and to remove noise data having noise frequencies consistent with the environmental noise data from the first noise data to obtain transfer device noise data. A noise recognition module is configured to input the transfer device noise data into a trained transfer device state detection model to obtain a state type corresponding to the transfer device noise data, wherein the state type comprises a normal state type and an abnormal state type, and the abnormal state type comprises a determined abnormal type and an undetermined abnormal type. A third data acquisition module is configured to selectively acquire a power supply network voltage frequency or a transfer device output power based on the undetermined abnormal type. A deep analysis module is configured to define the undetermined abnormal type as a self-recoverable abnormality and a non-self-recoverable abnormality based on the power supply network voltage frequency or the transfer device output power, and to determine an abnormal cause.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the machine learning-based transfer device abnormality recognition method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the machine learning-based transfer device abnormality recognition method of any one of claims 1 to 7.

Images