On-load tap-changer fault intelligent online diagnosis device based on multi-model fusion

Abstract

The invention discloses an on-load tap-changer fault intelligent online diagnosis device based on multi-model fusion, and relates to the technical field of electrical engineering. The core processing unit is used for receiving, processing and storing data acquired by the acquisition module; the fault diagnosis module is used for performing fault diagnosis on the data processed by the core processing unit; the multi-model fusion module is used for performing fusion processing on the diagnosis results to generate a final fault diagnosis report; according to the invention, a large amount of characteristic data is acquired through the acquisition module, and the relationship between each acquired characteristic data and each fault of the on-load tap-changer is established by using an artificial intelligence neural network model based on a large amount of test data. Therefore, the overall online diagnosis of electrical, mechanical and transmission faults of the on-load tap-changer is realized, and the reliability of equipment and the safe and stable operation of a power grid are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical engineering, and in particular to an intelligent online diagnosis device for on-load tap changer faults based on multi-model fusion. Background Art

[0002] The on-load tap changer (OLTC) is a core component of a voltage-regulating transformer. It achieves voltage regulation by gradually changing the high-voltage side windings. Its operational reliability directly impacts the safe and stable operation of the transformer and, ultimately, the entire power grid. Statistics show that failures caused by abnormal tap changers account for over 20% of all transformer failures. Therefore, abnormal OLTC operating conditions pose a significant safety hazard to the safe and stable operation of the transformer. Optimized and effective OLTC status identification and fault diagnosis are essential. Promptly and accurately identifying abnormal OLTC conditions provides crucial insights for maintenance personnel during condition-based maintenance, ultimately preventing accidents before they occur.

[0003] Existing transformer on-load tap-changer fault detection devices, based on the characteristics of electrical, mechanical, and drive faults in on-site OLTC equipment, mainly extract signal features from accompanying data samples and introduce artificial intelligence machine learning algorithms to improve the efficiency of identifying equipment status. However, due to the limitations of the status information carried by a single accompanying signal, many OLTC fault diagnosis methods are mostly applicable to specific faulty components or fault types, and cannot meet the overall monitoring needs of operating equipment. In addition, traditional transformer on-load tap-changer fault maintenance methods generally involve periodic power outages, which can easily lead to over- or under-maintenance of the OLTC, seriously reducing maintenance efficiency.

[0004] Therefore, it is necessary to design an intelligent online diagnosis device for on-load tap-changer faults based on multi-model fusion to solve the above technical problems. Summary of the Invention

[0005] In response to the above problems, the present application provides an intelligent online fault diagnosis device for on-load tap-changers based on multi-model fusion. A large amount of feature data is collected through ultra-high frequency sensors, voiceprint sensors, vibration sensors, oil chromatography sensors, and contactless magnetoresistive sensors. Based on a large amount of test data, an artificial intelligence neural network model is used to establish a relationship between the collected feature data and various types of faults of the on-load tap-changer. In this way, the overall online diagnosis of electrical, mechanical, transmission and other faults of the on-load tap-changer is realized, ensuring the reliability of the equipment and the safe and stable operation of the power grid.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion, comprising:

[0007] Acquisition module, used to collect real-time operating data of the on-load tap-changer;

[0008] The core processing unit is used to receive the data collected by the acquisition module, and perform data processing and data storage;

[0009] a fault diagnosis module, configured to receive abnormal data transmitted by the core processing unit, and further configured to perform fault diagnosis using at least two different types of fault diagnosis models based on the data processed by the core processing unit;

[0010] a multi-model fusion module, configured to fuse the multiple diagnostic results output by the fault diagnosis module to generate a final fault diagnosis report; and

[0011] Output module, used for data transmission, display and query of transformer related parameters.

[0012] Furthermore, preferably, the acquisition module includes but is not limited to an ultra-high frequency acquisition module, an acoustic ripple wave acquisition module, a vibration acquisition module, an oil chromatography acquisition module, and a magnetoresistive rotation angle acquisition module.

[0013] Furthermore, preferably, the core processing unit includes a main control chip and a data storage device, wherein the main control chip is used to receive and process the operation input information of the on-load tap changer and run the artificial neural network fault diagnosis model trained by the background terminal; the data storage device is used to store the trained artificial neural network fault diagnosis model and data information.

[0014] Furthermore, preferably, the at least two different types of fault diagnosis models adopted by the fault diagnosis module include but are not limited to a physical model-based method and a machine learning-based method.

[0015] Furthermore, preferably, the output module includes a wireless communication network module, a coordinator, a background terminal and a human-machine interface, wherein the wireless communication network module is based on ZigBee wireless communication network technology, and the wireless communication network module transmits each data to the coordinator through a router. The coordinator is responsible for establishing the entire ZigBee network, and the coordinator sends the data to the background terminal through the GPRS network. The background terminal is responsible for storing a large amount of real-time collected data for updating the artificial neural network model, and the human-machine interface is used for displaying and querying transformer-related parameters.

[0016] Furthermore, preferably, the core processing unit further includes a data cleaning submodule and a feature extraction submodule, wherein the data cleaning submodule is used to remove invalid data in the data sample, and the feature extraction submodule is used to extract key features from valid data in the data sample.

[0017] Furthermore, preferably, the fault diagnosis module further includes a model training submodule, and the model training submodule is used to train and update and optimize the at least two different types of fault diagnosis models based on historical fault data.

[0018] Furthermore, preferably, the voiceprint acquisition module includes but is not limited to a microphone, an analog-to-digital converter, an audio codec, a microprocessor and a USB controller. The microphone is used to convert the OLTC voiceprint signal into an analog signal, which is then converted into a pulse density modulation signal through the analog-to-digital converter. The audio decoder is used to convert the pulse density modulation signal into a pulse code modulation signal. The pulse code modulation signal data is stored in the direct memory access space of the microprocessor through the protocol and is transmitted to the computer through the USB audio protocol.

[0019] Furthermore, preferably, the magnetoresistive rotation angle acquisition module is connected to the driving motor rotating shaft through a soft rubber rod connector, and the magnetoresistive rotation angle acquisition module outputs the driving motor rotation angle signal as a series of voltage pulses for acquisition, and indirectly reflects the driving motor state of torque change through the rotation angle and speed of the driving motor rotating shaft.

[0020] An online diagnosis method for an intelligent online diagnosis device for on-load tap-changer faults based on multi-model fusion includes the following steps:

[0021] Step 1: After installing various sensors, the acquisition module collects voiceprint, vibration, ultra-high frequency, oil chromatography, rotation angle and other data in real time to obtain different types of OLTC data samples;

[0022] Step 2: The core processing unit preprocesses the collected data;

[0023] Step 3: The core processing unit pre-emphasizes the collected OLTC voiceprint. The signal obtained by short-time Fourier transform (SFT) is input into a Mel filter and logarithmized to correct for the nonlinearity of the auditory sense of frequency and sound intensity. Finally, the MFCC features of the OLTC voiceprint are obtained using discrete cosine transform (DCT).

[0024] Step 4: The core processing unit amplifies and filters the collected original vibration signal through the conditioner to obtain the time domain waveform signal under different states of the OLTC;

[0025] Step 5: The core processing unit converts the collected UHF, oil chromatography, and rotation angle data into a data set that can participate in algorithm training through feature engineering;

[0026] Step 6: Set the initial number of hidden layer nodes h0 and the maximum number of hidden layer nodes h max, initialize the IAO algorithm and LSTM neural network parameters, optimize the number of hidden layer nodes and learning rate of the LSTM neural network based on the IAO algorithm, and the optimization objective function is:

[0027]

[0028] Where: y sq , y′ sq are the actual value and predicted value of the sth state under the qth historical data respectively; N train is the amount of training set data, q is the number parameter of historical data, s is the number parameter of state, and f0 is the optimization objective function;

[0029] Step 7: Based on the optimized training set model, the optimal fitness values ​​optimized by the IAO algorithm with different numbers of hidden layer nodes are output. The LSTM neural network parameters at the maximum fitness are used as the final parameters to complete network training. The RF algorithm is then used to fuse the front-end multi-classifiers to finally obtain the fault result.

[0030] Step 8: Verify with test set data, output recognition results and evaluate the model.

[0031] Compared with the prior art, the present invention provides an intelligent online fault diagnosis device for on-load tap changers based on multi-model fusion, which has the following beneficial effects:

[0032] 1. The present invention collects a large amount of characteristic data through ultra-high frequency sensors, voiceprint sensors, vibration sensors, oil chromatography sensors, and contactless magnetoresistive sensors. Based on a large amount of test data, an artificial intelligence neural network model is used to establish the relationship between each collected characteristic data and various types of on-load tap-changer faults. A parallel integrated learning algorithm structure is adopted to achieve comprehensive online diagnosis of electrical, mechanical, and transmission faults of the on-load tap-changer, thereby ensuring the reliability of the equipment and the safe and stable operation of the power grid.

[0033] 2. The present invention can directly monitor the faults of the transformer on-load tap-changer online, automatically complete data collection and preliminary analysis, timely discover potential faults, reduce unplanned downtime caused by sudden faults, and reduce the workload of operation and maintenance personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application.

[0035] Figure 1 The figure is a schematic diagram of the structure of an intelligent online diagnosis device for on-load tap-changer faults based on multi-model fusion;

[0036] Figure 2 Schematic diagram of the OLTC voiceprint collection process in the intelligent online diagnosis device for on-load tap-changer faults based on multi-model fusion;

[0037] Figure 3 This is a schematic diagram of the MFCC feature extraction process in the intelligent online diagnosis device for on-load tap-changer faults based on multi-model fusion;

[0038] Figure 4 Schematic diagram of the online diagnosis process of the intelligent online diagnosis device for on-load tap-changer faults based on multi-model fusion; DETAILED DESCRIPTION

[0039] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0040] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the technical features being referred to. Thus, a feature specified as "first" or "second" may explicitly or implicitly include one or more of such features; and in the description of this application, unless otherwise specified, "plurality" means two or more.

[0041] See also Figures 1-4 The embodiment of the present invention provides an intelligent online diagnosis device for on-load tap changer faults based on multi-model fusion, which includes:

[0042] Acquisition module, used to collect real-time operating data of the on-load tap-changer;

[0043] The core processing unit is used to receive the data collected by the acquisition module, and perform data processing and data storage;

[0044] a fault diagnosis module, configured to receive abnormal data transmitted by the core processing unit, and further configured to perform fault diagnosis using at least two different types of fault diagnosis models based on the data processed by the core processing unit;

[0045] a multi-model fusion module, configured to fuse the multiple diagnostic results output by the fault diagnosis module to generate a final fault diagnosis report; and

[0046] Output module, used for data transmission, display and query of transformer related parameters.

[0047] Based on the above technical solution, acquisition modules are first installed on the key components of the on-load tap changer to ensure comprehensive coverage of possible fault sources and obtain different types of OLTC data samples. The core processing unit preprocesses the received data and extracts features from each data. The fault diagnosis module diagnoses abnormal data and classifies and predicts the data using the trained model. The diagnostic results of the two models are output to the multi-model fusion module. The multi-model fusion module then fuses the diagnostic results of the two models. The output module generates the final fault diagnosis results, displays them in graphical form on the operation interface, and sends them to relevant maintenance personnel.

[0048] On the basis of the above technical solution, the acquisition module can collect a large amount of characteristic data. Based on a large amount of test data, the artificial intelligence neural network model is used to establish the relationship between each collected characteristic data and various types of on-load tap-changer faults. The parallel integrated learning algorithm structure is adopted to realize the overall online diagnosis of electrical, mechanical, transmission and other faults of the on-load tap-changer, ensuring the reliability of the equipment and the safe and stable operation of the power grid. In addition, the intelligent online diagnosis device for on-load tap-changer faults based on multi-model fusion can directly monitor the faults of the transformer on-load tap-changer online, automatically complete data collection and preliminary analysis, and can timely discover potential faults, reduce unplanned downtime caused by sudden faults, and reduce the workload of operation and maintenance personnel.

[0049] In a possible implementation, the acquisition module includes but is not limited to an ultra-high frequency acquisition module, an acoustic ripple wave acquisition module, a vibration acquisition module, an oil chromatography acquisition module, and a magnetoresistive rotation angle acquisition module.

[0050] Using the above technical solution, the ultra-high frequency acquisition module is mainly used to detect the electromagnetic wave signal generated by partial discharge inside the on-load tap changer. The electromagnetic wave signal generated by the partial discharge resonates and its frequency can reach 3GHz. Therefore, the ultra-high frequency sensor can well acquire the electromagnetic wave signal generated by the partial discharge; the vibration acquisition module is mainly used to collect the mechanical vibration signal of the on-load tap changer to identify whether there is any sticking, loosening, deformation, wear, lubrication abnormality, etc. during the operation of the switching mechanism; the oil chromatography acquisition module is mainly used to online detect the gas component-related data in the insulating oil of the on-load tap changer. Based on the gas chromatography analysis of the content of dissolved gas components in the insulating oil through the component ratio and gas production rate of the characteristic gas, the traditional three-ratio method can be used. The traditional three-ratio method uses the three ratios of C2H2 / C2H4, CH4 / H2, and C2H4 / C2H6 to form an oil and gas feature code to describe the fault type and severity, thereby realizing the operation status monitoring and electrical fault diagnosis of the OLTC.

[0051] Based on the above technical solution, due to the complex actual operating environment of OLTC equipment, the collected original vibration signal is accompanied by a large amount of noise interference, which may cause the effective signal to be submerged. Therefore, the original signal is amplified and filtered by the conditioner to obtain the time domain waveform signal under different OLTC states.

[0052] In one possible implementation, the core processing unit includes a main control chip and a data storage device, wherein the main control chip is used to receive and process operating input information of the on-load tap changer and run the artificial neural network fault diagnosis model trained by the background terminal; the data storage device is used to store the trained artificial neural network fault diagnosis model and data information.

[0053] In a possible implementation, the at least two different types of fault diagnosis models used by the fault diagnosis module include but are not limited to a physical model-based method and a machine learning-based method.

[0054] In one possible implementation, the output module includes a wireless communication network module, a coordinator, a background terminal, and a human-machine interface, wherein the wireless communication network module is based on ZigBee wireless communication network technology, and the wireless communication network module transmits each data to the coordinator through a router. The coordinator is responsible for establishing the entire ZigBee network, and the coordinator sends the data to the background terminal through the GPRS network. The background terminal is responsible for storing a large amount of real-time collected data for updating the artificial neural network model, and the human-machine interface is used to display and query transformer-related parameters.

[0055] In a possible implementation, the core processing unit further includes a data cleaning submodule and a feature extraction submodule. The data cleaning submodule is used to remove invalid data in the data sample, and the feature extraction submodule is used to extract key features from valid data in the data sample.

[0056] In a possible implementation, the fault diagnosis module further includes a model training submodule, and the model training submodule is used to train and update and optimize the at least two different types of fault diagnosis models based on historical fault data.

[0057] In one possible implementation, the voiceprint acquisition module includes but is not limited to a microphone, an analog-to-digital converter, an audio codec, a microprocessor, and a USB controller. The microphone is used to convert the OLTC voiceprint signal into an analog signal, which is then converted into a pulse density modulation signal through the analog-to-digital converter. The audio decoder is used to convert the pulse density modulation signal into a pulse code modulation signal. The pulse code modulation signal data is stored in the direct memory access space of the microprocessor through the protocol and transmitted to the computer through the USB audio protocol.

[0058] Through the above technical solution, the microphone is used to convert the OLTC voiceprint signal into an analog signal, and then convert it into a pulse density modulation signal (PDM) through an analog-to-digital converter. It can use the SPH0655 chip that integrates a microphone and a digital-to-analog converter; the audio decoder can use the audio decoder ADAU7118 chip. The audio decoder ADAU7118 chip converts PDM into a pulse code modulation signal (PCM), which can better process and analyze the voiceprint data. The ADAU7118 chip has an I2S protocol interface and supports a sampling rate of 4kHz-92kHz, enabling it to achieve high-quality sound collection of OLTC voiceprints; the microprocessor can use the STM32F407 microprocessor. The PCM data can be stored in the direct memory access space of the STM32F407 microprocessor through the I2S protocol, and finally transmitted to the computer through the USB audio protocol. At the same time, in order to improve the sample quality, the voiceprint collection device can be placed in the middle position between the electric mechanism and the switch, and the sampling rate for collecting OLTC voiceprint samples can be set to 192kHz.

[0059] It should be noted that the core processing unit uses a short-time Fourier transform (SFT) to pre-emphasize the collected OLTC voiceprint. The SFT includes framing, windowing, and fast Fourier transform (FFT). Since OLTC voiceprints are non-periodic audio signals, they require framing to extract local features. Within the short time span of a single frame, the signal can be roughly considered stationary. After framing, the signal's start and end points are highly discontinuous. To avoid spectral leakage and the appearance of high-frequency components at discontinuities after direct FFT transformation, Hamming windowing is applied to each frame.

[0060] In one possible embodiment, the magnetoresistive rotation angle acquisition module is connected to the driving motor rotating shaft through a soft rubber rod connector. The magnetoresistive rotation angle acquisition module outputs the driving motor rotation angle signal as a series of voltage pulses for acquisition, and indirectly reflects the driving motor state of torque change through the rotation angle and speed of the driving motor rotating shaft.

[0061] By adopting the above technical solution, the magnetoresistive rotation angle acquisition module can adopt the WYT-3 contactless angle sensor, which converts the mechanical rotational motion of the OLTC drive motor spindle into an electrical signal output. The core component of the WYT-3 contactless angle sensor is a magnetoresistive element, which works based on the magnetoresistive effect.

[0062] It's important to explain that the magnetoresistance effect occurs when an external magnetic field is applied perpendicular to the surface of a current-carrying semiconductor. Not only does a Hall potential appear, but the resistance across the semiconductor also changes. Generally, semiconductor resistance increases with increasing magnetic field strength. Under the influence of an external magnetic field, the Lorentz force causes linearly moving charge carriers in the semiconductor to deviate from their original direction, extending their travel distance and time. This increases their chances of collision with the crystal lattice, impurity atoms, and lattice defects, thereby reducing electron mobility and manifesting itself macroscopically as an increase in component resistance. Utilizing this principle, when the rotating shaft of an angle sensor drives a fixed permanent magnet to rotate, the changing magnetic field applied to the magnetic sensor causes its resistance to change, resulting in a sinusoidal output voltage that varies periodically with the rotation angle.

[0063] Based on the above technical solutions, UHF, oil chromatography, and rotation angle data can be transformed into a form suitable for algorithm training through feature engineering. Feature engineering is a key step in improving the overall performance of a model. It includes feature mapping, feature selection, feature reconstruction, feature normalization, and feature extraction. Through this process, data is transformed into a dataset suitable for model training.

[0064] An online diagnosis method for an intelligent online diagnosis device for on-load tap-changer faults based on multi-model fusion includes the following steps:

[0065] Step 1: After installing various sensors, the acquisition module collects voiceprint, vibration, ultra-high frequency, oil chromatography, rotation angle and other data in real time to obtain different types of OLTC data samples;

[0066] Step 2: The core processing unit preprocesses the collected data;

[0067] Step 3: The core processing unit pre-emphasizes the collected OLTC voiceprint. The signal obtained by short-time Fourier transform (SFT) is input into a Mel filter and logarithmized to correct for the nonlinearity of the auditory sense of frequency and sound intensity. Finally, the MFCC features of the OLTC voiceprint are obtained using discrete cosine transform (DCT).

[0068] Step 4: The core processing unit amplifies and filters the collected original vibration signal through the conditioner to obtain the time domain waveform signal under different states of the OLTC;

[0069] Step 5: The core processing unit converts the collected UHF, oil chromatography, and rotation angle data into a data set that can participate in algorithm training through feature engineering;

[0070] Step 6: Set the initial number of hidden layer nodes h0 and the maximum number of hidden layer nodes h max , initialize the IAO algorithm and LSTM neural network parameters, optimize the number of hidden layer nodes and learning rate of the LSTM neural network based on the IAO algorithm, and the optimization objective function is:

[0071]

[0072] Where: y sq , y′ sq are the actual value and predicted value of the sth state under the qth historical data respectively; N train is the amount of training set data, q is the number parameter of historical data, s is the number parameter of state, and f0 is the optimization objective function;

[0073] Step 7: Based on the optimized training set model, the optimal fitness values ​​optimized by the IAO algorithm with different numbers of hidden layer nodes are output. The LSTM neural network parameters at the maximum fitness are used as the final parameters to complete network training. The RF algorithm is then used to fuse the front-end multi-classifiers to finally obtain the fault result.

[0074] Step 8: Verify with test set data, output recognition results and evaluate the model.

[0075] The above is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, can make equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. An intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion, characterized in that: It includes: Collection module, used to collect the operation data of the on-load tap changer in real time; A core processing unit, used to receive the data collected by the collection module, and perform data processing and data storage; A fault diagnosis module, used for receiving abnormal data transmitted by the core processing unit, and the fault diagnosis module is further used for performing fault diagnosis using at least two different types of fault diagnosis models according to the data processed by the core processing unit; A multi-model fusion module, used to fuse multiple diagnosis results output by the fault diagnosis module to generate a final fault diagnosis report; as well as Output module, used for data transmission, display and query of transformer related parameters.

2. The intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion according to claim 1 is characterized in that: The acquisition modules include but are not limited to an ultra-high frequency acquisition module, an acoustic ripple wave acquisition module, a vibration acquisition module, an oil chromatography acquisition module, and a magnetic resistance rotation angle acquisition module.

3. The intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion according to claim 1 is characterized in that: The core processing unit includes a main control chip and a data storage device, wherein the main control chip is used to receive and process the operation input information of the on-load tap changer and run the artificial neural network fault diagnosis model trained by the background terminal; the data storage device is used to store the trained artificial neural network fault diagnosis model and data information.

4. The intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion according to claim 1 is characterized in that: The at least two different types of fault diagnosis models adopted by the fault diagnosis module include but are not limited to a method based on a physical model and a method based on machine learning.

5. The intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion according to claim 1 is characterized in that: The output module includes a wireless communication network module, a coordinator, a background terminal and a human-machine interface, wherein the wireless communication network module is based on ZigBee wireless communication network technology, the wireless communication network module transmits each data to the coordinator through a router, the coordinator is responsible for establishing the entire ZigBee network, and the coordinator sends the data to the background terminal through the GPRS network, the background terminal is responsible for storing a large amount of real-time collected data for updating the artificial neural network model, and the human-machine interface is used for displaying and querying transformer-related parameters.

6. The intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion according to claim 1 is characterized in that: The core processing unit also includes a data cleaning submodule and a feature extraction submodule. The data cleaning submodule is used to remove invalid data in the data sample, and the feature extraction submodule is used to extract key features from valid data in the data sample.

7. The intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion according to claim 1 is characterized in that: The fault diagnosis module also includes a model training submodule, which is used to train and update and optimize the at least two different types of fault diagnosis models according to historical fault data.

8. The intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion according to claim 2 is characterized in that: The voice ripple wave acquisition module includes but is not limited to a microphone, an analog-to-digital converter, an audio codec, a microprocessor and a USB controller. The microphone is used to convert the OLTC voiceprint signal into an analog signal, and then convert it into a pulse density modulation signal through the analog-to-digital converter. The audio decoder is used to convert the pulse density modulation signal into a pulse code modulation signal. The pulse code modulation signal data is stored in the direct memory access space of the microprocessor through the protocol and transmitted to the computer through the USB audio protocol.

9. The intelligent online fault diagnosis device for on-load tap changer based on multi-model fusion according to claim 2 is characterized in that: The reluctance type rotation angle acquisition module is connected to the driving motor rotating shaft through a soft rubber rod connector. The reluctance type rotation angle acquisition module outputs the driving motor rotation angle signal as a series of voltage pulses for acquisition, and indirectly reflects the driving motor state of torque change through the rotation angle and speed of the driving motor rotating shaft.

10. An online diagnosis method for an on-load tap changer fault intelligent online diagnosis device based on multi-model fusion according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: Install various sensors, and the acquisition module collects voiceprint, vibration, ultra-high frequency, oil chromatography, rotation angle and other data in real time to obtain different types of OLTC data samples; Step 2, the core processing unit preprocesses the collected data; Step 3: The core processing unit performs pre-emphasis processing on the collected OLTC voiceprint, inputs the signal obtained by short-time Fourier transform of the OLTC voiceprint into the Mel filter and takes the logarithm to correct the nonlinearity of the auditory sense to frequency and sound intensity, and finally obtains the MFCC features of the OLTC voiceprint using discrete cosine transform; Step 4: The core processing unit amplifies and filters the collected original vibration signal through a conditioner, and obtains the time domain waveform signal under different states of OLTC after processing; Step 5: The core processing unit converts the collected UHF, oil chromatography, and rotation angle data into a data set that can participate in algorithm training through feature engineering; Step 6: Set the initial number of nodes h0 and the maximum number of nodes h in the hidden layer. max , initialize the IAO algorithm and LSTM neural network parameters, optimize the number of hidden layer nodes and learning rate of the LSTM neural network based on the IAO algorithm, and the optimization objective function is: Where:y sq , y′ sq are the actual value and predicted value of the sth state under the qth historical data respectively; N train is the amount of training set data, q is the number parameter of historical data, s is the number parameter of state, and f0 is the optimization objective function; Step 7: Based on the optimized training set model, the different optimal fitness values ​​optimized by the IAO algorithm under different numbers of hidden layer nodes are output, and the LSTM neural network parameters at the maximum fitness are used as the final parameters to complete the network training, and the front-end multi-classifiers are fused using the RF algorithm to finally obtain the fault result; Step 8: Verify with test set data, output recognition results and evaluate the model.

Images