Distribution transformer on-line health diagnosis system and method

Abstract

The invention discloses a distribution transformer on-line health diagnosis system and method. The distribution transformer on-line health diagnosis system comprises a low-voltage transformer area, a cloud platform and a big data application layer. The on-line health diagnosis of the distribution transformer comprises the steps of data acquisition, analysis, storage, statistical model establishment for prediction and artificial intelligence deep learning for prediction. Various data in the operation process of the transformer are collected in real time through the terminal data collector, meteorological indexes such as the temperature and the humidity of the operation environment of the transformer are combined, processing and analysis are carried out through combination of statistical model analysis and an artificial intelligence algorithm, transformer fault diagnosis is carried out, and the operation condition of the transformer is monitored. And once it is found that the operation state of the transformer has a deterioration trend, early warning is given out in advance, and early warning of the sub-health state of the transformer is achieved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of transformers, and in particular to an online health diagnosis system and method for distribution transformers. Background Art

[0002] At present, transformers are widely used in the power system and are one of the very important devices in the power system. The operating condition of the transformer is crucial for the stability of the power grid power supply. However, after long-term operation, due to factors such as changes in power load, changes in environmental temperature and humidity, insulation aging, overheating, mechanical vibration, and oil leakage, different types and locations of faults will gradually occur. For example, dry-type transformers are prone to electrical and mechanical faults, and oil-immersed power transformers are prone to problems such as abnormal oil level, oil leakage fault, and insulation moisture absorption fault.

[0003] The existing transformer monitoring methods are relatively single and have few monitoring dimensions. For example, the common noise analysis method judges the fault type by monitoring the sound of the transformer during operation, but the error is large, it is greatly affected by environmental noise, and the recognition result is too simplistic to identify different faults of the transformer.

[0004] Therefore, it is necessary to explore an online health diagnosis system and method for distribution transformers to monitor the operating state of the transformer from multiple dimensions to improve the accuracy of fault diagnosis, thereby solving the above problems. Summary of the Invention

[0005] The main object of the present invention is to provide an online health diagnosis system and method for distribution transformers to solve the problems of small monitoring dimensions of the operating state of the transformer and low accuracy of fault judgment and prediction in the prior art.

[0006] To achieve the above object, according to an online health diagnosis system and method for distribution transformers provided by the present invention, the online health diagnosis system for distribution transformers includes a low-voltage substation area, a cloud platform, and a big data application layer;

[0007] The low-voltage substation area includes a data acquisition intelligent terminal and a plurality of terminal data collectors; the terminal data collectors are used to collect the operating state data of the transformer and transmit the collected operating state data to the data acquisition intelligent terminal; the data acquisition intelligent terminal processes and analyzes the operating state data and transmits the processing result to the cloud platform;

[0008] The cloud platform includes a management and monitoring platform and a plurality of workstations; the management and monitoring platform receives the operating state data and updates the operating state information of the transformer in real time; the workstations maintain the management and monitoring platform;

[0009] The big data application layer includes a big data processing unit, an Internet of Things unit, and an artificial intelligence unit; the big data processing unit is connected to the management and monitoring platform, and processes and analyzes the real-time updated data provided by the management and monitoring platform; the Internet of Things unit is connected to the management and monitoring platform and manages the devices connected to it; the artificial intelligence unit is respectively connected to the big data processing unit, the Internet of Things unit, and the management and monitoring platform, and is used to analyze and predict whether there is a fault in the transformer by using the transformer knowledge graph and the artificial intelligence deep learning fault analysis and prediction algorithm.

[0010] Preferably, the terminal data collector is any one of a temperature collector, a humidity sensor, a dissolved gas sensor, a vibration collector, a voiceprint acquisition sensor, an infrared sensor, an ultrasonic sensor, a thermal imaging sensor, a transformer oil concentration sensor, a transformer oil temperature sensor, a transformer oil pressure sensor, or a transformer oil level sensor.

[0011] Preferably, the data acquisition intelligent terminal includes an STM32 embedded CPU, and a 4G communication module, a Bluetooth wireless communication module, an RS485 wired communication, a liquid crystal display module, a button, and 8-channel analog signal acquisition channels respectively connected to the STM32 embedded CPU.

[0012] Preferably, a wireless communication module is provided on the terminal data collector, and the terminal data collector is connected to an adjacent terminal data collector in a wireless communication manner through the wireless communication module; the terminal data collector is connected to the data acquisition intelligent terminal in a wireless communication manner.

[0013] Preferably, the terminal data collector and the data acquisition intelligent terminal are connected through an industrial field bus.

[0014] A method for online health diagnosis of a distribution transformer, the method comprising the following steps:

[0015] S1: The operating state data of each dimension of the transformer are respectively collected through a plurality of terminal data collectors, and the data acquisition intelligent terminal collects the data of each terminal data collector;

[0016] S2: The data acquisition intelligent terminal processes and analyzes the multi-dimensional data of the transformer, and transmits the processed data to the cloud platform;

[0017] S3: The cloud platform aggregates and stores the data transmitted by each data acquisition intelligent terminal, and transmits it to the big data processing unit and the artificial intelligence unit;

[0018] S4: For a smaller data set, the big data processing unit calls a statistical model to further predict data points based on historical time data at a later time and obtains a prediction result;

[0019] S5: For larger datasets, the artificial intelligence unit inputs the multi-dimensional data of the transformer into a selected deep learning model for artificial intelligence training, and analyzes and predicts whether the transformer has faults based on the transformer knowledge graph and the artificial intelligence deep learning fault analysis and prediction algorithm.

[0020] Preferably, in step S2, the process of processing and analyzing can be divided into the following steps:

[0021] S2-1: Data preprocessing: including data cleaning, outlier removal, and data normalization;

[0022] S2-2: Feature extraction: Extract meaningful feature data from the original data, including but not limited to RMS value, temperature value, liquid level value, or relative deviation;

[0023] S2-3: Data dimensionality reduction: Map multi-dimensional data to a low-dimensional space to reduce the amount of calculation and better display the characteristics of the data.

[0024] Preferably, the statistical model is any one of autoregressive model, moving average model, autoregressive moving average model, autoregressive integrated moving average model, seasonal autoregressive integrated moving average model, seasonal autoregressive integrated moving average model with exogenous regressors, vector autoregressive model, or vector error correction model.

[0025] Preferably, in step S5, the deep learning model is a multi-layer perceptron, recurrent neural network, long short-term memory network, autoregressive model, or convolutional neural network model.

[0026] Preferably, the deep learning model is a convolutional neural network model. The time series prediction based on the convolutional neural network model includes the following steps:

[0027] Data preparation: Divide the time series dataset into a training set and a test set. The training set is used to train the convolutional neural network model, and the test set is used to evaluate the prediction performance of the model;

[0028] Data transformation: Use the sliding window method, Fourier transform method, or other methods to transform the time series data into two-dimensional image data;

[0029] Convolutional neural network model construction: Construct a convolutional neural network model including a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer is used to extract features in the time series data, the pooling layer is used to reduce the feature dimension, and the fully connected layer is used to output the prediction result;

[0030] Model Training: Use the training set to train the convolutional neural network model, and update the weights and biases of the model through the backpropagation algorithm to minimize the prediction error;

[0031] Model Prediction: Use the trained convolutional neural network model to predict the test set, input the image data of the test set, and obtain the prediction result through the forward propagation algorithm;

[0032] Model Evaluation: Compare the prediction result with the true value of the test set, calculate indicators such as prediction error and root mean square error, and evaluate the prediction performance of the model.

[0033] As can be seen from the above technical solutions, the present invention has the following advantages: The present invention collects various data during the operation of the transformer in real time through the terminal data collector, combines meteorological indicators such as temperature and humidity of the transformer operation environment, and performs processing and analysis through the combination of statistical model analysis and artificial intelligence algorithms to conduct transformer fault diagnosis and monitor the operation of the transformer. Once it is found that the operation state of the transformer has a deteriorating trend, an early warning is issued in advance to realize the early warning of the sub-healthy state of the transformer. Brief Description of the Drawings

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0035] Figure 1 It is a schematic structural diagram of the on-line health diagnosis system of the distribution transformer of the present invention;

[0036] Figure 2 It is a schematic structural diagram of the data acquisition intelligent terminal of the present invention;

[0037] Among them, the arrow indicates the data transmission direction, and the marks in each drawing are as follows:

[0038] 1 - Data acquisition intelligent terminal, 11 - STM32 embedded CPU, 12 - 4G communication module, 13 - Bluetooth wireless communication module, 14 - RS485 wired communication, 15 - Liquid crystal display module, 16 - Button, 17 - 8-channel analog signal acquisition channel, 2 - Terminal data collector, 3 - Management and monitoring platform, 4 - Workstation, 5 - Big data processing unit, 6 - Internet of Things unit, 7 - Artificial intelligence unit. Detailed Embodiment

[0039] The main object of the present invention is to provide an on-line health diagnosis system and method for a distribution transformer, so as to solve the problems of small monitoring dimension of the operating state of the transformer and low accuracy of fault judgment and prediction in the prior art.

[0040] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0041] Please refer to Figure 1 and Figure 2 , the present invention provides an on-line health diagnosis system and method for a distribution transformer. The on-line health diagnosis system for a distribution transformer includes a low-voltage area, a cloud platform and a big data application layer;

[0042] The low-voltage area includes a data acquisition intelligent terminal 1 and several terminal data collectors 2; the terminal data collector 2 is used to collect the operation state data of the transformer and transmit the collected operation state data to the data acquisition intelligent terminal 1; the data acquisition intelligent terminal 1 processes and analyzes the operation state data and transmits the processing result to the cloud platform;

[0043] The cloud platform includes a management and monitoring platform 3 and several workstations 4; the management and monitoring platform 3 receives the operation state data and updates the operation state information of the transformer in real time; the workstation 4 maintains the management and monitoring platform 3;

[0044] The big data application layer includes a big data processing unit 5, an Internet of Things unit 6 and an artificial intelligence unit 7; the big data processing unit 5 is connected to the management and monitoring platform 3 and processes and analyzes the real-time updated data provided by the management and monitoring platform; the Internet of Things unit 6 is connected to the management and monitoring platform 3 and manages the devices connected to it; the artificial intelligence unit 7 is respectively connected to the big data processing unit 5, the Internet of Things unit 6 and the management and monitoring platform 3, and is used to analyze and predict whether there is a fault in the transformer by using a transformer knowledge graph and an artificial intelligence deep learning fault analysis and prediction algorithm.

[0045] In the embodiments of the present invention, the fault prediction and diagnosis mainly include two types of objects: one is a dry-type transformer, and the other is a wet-type transformer.

[0046] The accidents of dry-type transformers are mainly insulation accidents, accounting for more than 81% of the total damage accidents. The traditional maintenance of the insulation of dry-type transformers not only requires the dry-type transformer to be powered off, but also regularly checks and replaces equipment that, in many cases, has not exceeded its service life, resulting in a waste of resources and economic losses. Therefore, the on-line detection and fault diagnosis of dry-type transformers are of great significance, which can not only reduce the maintenance cost of dry-type transformers, but also avoid various losses caused by faults.

[0047] The common faults of dry-type transformers are: electrical faults and mechanical faults; among them, electrical faults are divided into two situations: insulation failure faults and load overload faults. Insulation failure faults include partial discharge, overvoltage, chip winding, insulation aging, and poor insulation infiltration, etc. Load overload fault refers to the transformer load exceeding the rated capacity or working at the rated capacity load for a long time. Mechanical faults mainly include cooling system failure, settlement imbalance, abnormal external environment, and improper installation, etc. The data that needs to be on-line monitored includes the insulation degree of the transformer, temperature (analog quantity), transformer noise (analog quantity), ambient temperature and humidity (analog quantity).

[0048] The common faults of oil-immersed power transformers are: abnormal oil level, abnormal sound, transformer oil leakage fault, abnormal smell or color, insulation moisture absorption fault, and thermal fault;

[0049] Abnormal oil level is one of the most common faults of oil-immersed power transformers. First, the oil level of the capsule type oil conservator is abnormal, and there are various reasons for it, such as the aging of the pressure oil bag, the blockage of the breathing pipe by the breather resulting in poor pipeline flow, the poor sealing of parts such as the inspection window on the oil conservator, the residual gas in the oil conservator caused by improper operation during maintenance or installation, and the inappropriate amount of transformer oil added to the oil gauge, etc.; second, the oil level of the diaphragm type oil conservator is abnormal, which is caused by reasons such as the diaphragm rupture causing the oil level gauge to return to zero, the connecting rod bending causing the diaphragm movement not to correspond to the actual deflection of the pointer, or the float ball rupture resulting in the oil level gauge showing lower than the actual oil level.

[0050] The abnormal sound of oil-immersed power transformers mainly occurs during the operation of the transformer itself. First, when the transformer is overloaded during operation, the load changes too much, and with the action of harmonics, it will emit "clucking discontinuous sounds or "wow" sounds; when the screws or parts inside the transformer are loose, it is also easy for the sound to become louder and accompanied by noise, etc.; when the transformer winding is short-circuited, abnormal sounds like boiling water may occur; when the surface enamel of the high-voltage bushing of the transformer falls off or cracks, it will emit a "hissing" sound; when the branch switch or drop fuse of the transformer is poorly contacted, it will produce a "click" or "hissing" sound; when there is electromagnetic resonance or single-phase grounding in the power grid, it will emit a sharp sound.

[0051] Transformer oil leakage refers to the phenomenon that oil stains or oil beads appear on the surface of the transformer. The parts prone to oil leakage mainly include the upper part of the bushing, the sealing joints of the tank rim, the welds of the tank body and the tank rim, the welds of the flange, the sealing parts of the throttle plug, and the sealing parts of the oil level gauge and thermometer, etc. From the parts with oil leakage, it can be analyzed that the reasons for the transformer oil leakage fault are the low welding quality at the welds and the low sealing level at the sealing parts. In addition, insufficient accuracy of the mating surface, thread disorder of the screws, and sand holes in the cast iron parts will also cause the transformer oil leakage fault.

[0052] The abnormal smell or color of the oil-immersed transformer is caused by problems such as internal part detachment, discharge, and damage. For example, when the fixed position of the bushing wiring end is loose, serious oxidation will occur on the contact surface, damaging the plating on the surface of the connection head and wire clamp, making it lose its luster and the color become darker; when the oil pump and cooling fan are burned out, a burnt smell will be emitted; when the bushing is damaged, corona and flashover will occur, and then an ozone smell will be generated.

[0053] Insulation dampness is caused by a relatively high humidity in the environment where the transformer operates or internal liquid evaporation during operation. First, the transformer being exposed to the air or insufficient vacuum treatment before oil injection will cause insulation dampness; second, the manufacturing cycle of the equipment is relatively short, and it is put into use before being fully dried, resulting in a reduction in the insulation of the transformer winding; third, the sealing of the parts where the transformer body or accessories are in contact with the air is insufficient, causing moisture to enter and resulting in insulation dampness.

[0054] The thermal faults of the transformer are mostly over-temperature phenomena, mainly referring to the situation where, under normal operating conditions of the transformer, the oil temperature is more than 10°C higher than the normal temperature usually, or the oil temperature continues to rise when the load has not changed. This phenomenon is mostly caused by abnormalities inside the transformer, such as short circuits occurring in the winding layers or turns of the coil, resulting in heat generation, and the core overheating due to multiple points of the core being connected to the ground. In addition, overloading operation, faults of the temperature gauge, cooler, etc. will also cause abnormal temperature of the transformer.

[0055] According to the above fault manifestations, the present invention collects the transformer operating state data related to the fault manifestations by setting a number of terminal data collectors 2, and the terminal data collectors 2 can select sensors with different functions according to the types of data to be collected.

[0056] For example, in the insulation test of the dry-type transformer, the partial discharge method, high-frequency voltage method or air diagnosis method can be used for diagnosis, and the terminal data collector 2 can use sensors with relative test methods to accurately detect the partial discharge conditions of parts such as the transformer winding, support, and insulation structure.

[0057] During the operation of dry-type transformers, the sound waves generated by mechanical vibrations will refract, reflect, and scatter. By using the terminal data collector 2 that can capture sound wave signals to analyze the sound waves emitted by the transformer, mechanical faults such as problems with bearings, oil pastes, flywheels, gears, etc. can be effectively detected and diagnosed.

[0058] Thermal diagnosis technology uses the terminal data collector 2 that can detect the surface temperature of the transformer, such as infrared thermometers, infrared thermal imagers, and infrared thermal televisions, etc., to monitor abnormal thermal conditions during the operation of the transformer. By recording the temperature changes during the operation of the transformer, the fault status of the transformer, such as winding aging, poor contact, etc., can be accurately and real-time judged.

[0059] For oil-immersed power transformers, the terminal data collector 2 can be set according to the corresponding detection characteristics, such as:

[0060] The generation rate of all specific hydrocarbon gases will change with temperature. Its working principle is that when the transformer is operating normally, the transformer oil and solid insulation will decompose to produce extremely small amounts of hydrocarbons and gases such as CO and CO2. When the transformer has a thermal fault, an internal insulation moisture fault, or a discharge fault, the content of these gases will change. And different gases change for different faults. Therefore, the terminal data collector 2 can be set as a sensor that can detect gases to specifically analyze and judge the type and location of the transformer fault through the gas change situation.

[0061] Using ultrasonic technology to collect the sound waves during partial discharge and fault discharge of the transformer, convert the sound waves into electrical signals, and then analyze the electrical signals to analyze the fault can quickly locate the fault location and effectively avoid the influence of electromagnetic interference. Therefore, the terminal data collector 2 can adopt a sensor with ultrasonic detection function to collect sound wave data and be used for fault auxiliary diagnosis.

[0062] The terminal data collector 2 can be set with a sensor to detect the resistance of the transformer, which can be used to detect faults such as inter-turn short circuit of windings, contact of tap changers, broken strands of windings, and poor contact of joints. It can also judge the balance of the DC resistance of each phase winding and the correctness of the tap changer position. It can also sensitively respond to overall moisture, deterioration, and insulation penetration defects of the transformer insulation.

[0063] The terminal data collector 2 can be set as a radio voltage interferometer to judge the fault situation by detecting the radio control or radio communication interference generated by partial discharge. In addition, the radio waves emitted by partial discharge can be received through a receiving coil, and the center frequency of the frequency selection amplifier can be determined according to different test object situations.

[0064] In addition, the terminal data collector 2 can also be connected to the measurement circuit to detect the pulse current signal caused by partial discharge, obtain basic information such as the number of partial discharges, discharge phase, and discharge amount, so as to effectively judge the partial discharge fault situation. It is mainly used to detect partial discharge faults in transformer bushings, shell ground wires, core ground wires, end screen ground wires, windings, etc. Since the detection impedance and amplifier affect the measurement sensitivity, resolution, and accuracy, etc., when the capacitance in the test is large, it will be limited by the coupling impedance, reducing the measurement sensitivity of the test instrument.

[0065] Through the above structural design, the terminal data collector 2 is used to collect data such as transformer operation noise, vibration, winding temperature, core temperature, low-voltage terminal temperature, transformer oil top temperature, oil pressure, oil level, ambient temperature, zero-sequence current, etc. in real time. The data acquisition intelligent terminal 1 performs preliminary processing on the data, and stores the operation data and fault data in the cloud platform, and establishes a data analysis model in the big data application layer for data analysis and fault diagnosis. Once it is found that the operation state of the transformer has a deteriorating trend, an early warning will be issued in advance, and the device will be controlled through the Internet of Things unit 6 to realize early warning and control of the sub-healthy state of the transformer.

[0066] Preferably, the terminal data collector 2 is any one of a temperature collector, a humidity sensor, a dissolved gas sensor, a vibration collector, a voiceprint acquisition sensor, an infrared sensor, an ultrasonic sensor, a thermal imaging sensor, a transformer oil concentration sensor, a transformer oil temperature sensor, a transformer oil pressure sensor, or a transformer oil level sensor.

[0067] Preferably, the data acquisition intelligent terminal 1 includes an STM32 embedded CPU 11, and a 4G communication module 12, a Bluetooth wireless communication module 13, an RS485 wired communication 14, a liquid crystal display module 15, a key 16, and an 8-channel analog signal acquisition channel 17 that are respectively connected to the STM32 embedded CPU 11.

[0068] In this embodiment, the data collected by the data acquisition intelligent terminal 1 or the data uploaded by other terminal data collectors 2 are classified and analyzed and then uploaded to the cloud platform via the Internet (GPRS / 4G) wireless network. The data acquisition intelligent terminal 1 in this embodiment uses an STM32 embedded CPU 11, which is characterized by its powerful functions and high speed, which can meet the needs of the system. The data acquisition intelligent terminal 1 has 8 analog signal acquisition channels 17, which can sample analog signals such as transformer temperature, noise, oil concentration, and ambient temperature and humidity. In addition, it also has a 4G communication module 12, a Bluetooth wireless communication module 13, and an RS485 wired communication module 14, which can perform wired or wireless communication with other terminal devices in the substation. To facilitate on-site management, the data acquisition intelligent terminal 1 has a liquid crystal display function.

[0069] The data acquisition intelligent terminal 1 communicates with the cloud platform via an Internet (4G) wireless communication network, forming an Internet network based on the TCP / IP protocol. The core technology of the data acquisition intelligent terminal 1 is to enable interoperability among a heterogeneous control network composed of intelligent terminal devices supporting different communication protocols. This means that information can be exchanged between terminal devices using different transmission media, different speeds, and different communication protocols. Data is aggregated to a gateway and then transmitted to the cloud platform via the Internet. Data can also be sent to the administrator's mobile terminal via the mobile Internet.

[0070] Preferably, the terminal data collector 2 is provided with a wireless communication module, and is wirelessly connected to adjacent terminal data collectors 2 via the wireless communication module; the terminal data collector 2 is also wirelessly connected to the data acquisition intelligent terminal 1. In this embodiment, the terminal data collector 2 uses wireless transmission. If it cannot reach the data acquisition intelligent terminal 1 via a single hop, the wireless transmission module of another terminal data collector 2 is used as a relay. A path optimization algorithm is employed to transmit the data in a multi-hop manner to the transformer data acquisition intelligent terminal 1, and then the data is transmitted to the cloud platform via the Internet to ensure the timeliness and stability of data transmission.

[0071] Preferably, the terminal data collector 2 is connected to the data acquisition intelligent terminal 1 via an industrial field bus. In this embodiment, for a centralized terminal data collector 2, data can be collected to the data acquisition intelligent terminal 1 via a wired industrial field bus and then transmitted to the cloud platform via the Internet.

[0072] A method for online health diagnosis of a distribution transformer, comprising the following steps:

[0073] S1: The operating status data of each dimension of the transformer are collected by multiple terminal data collectors respectively, and the data acquisition intelligent terminal collects the data of each terminal data collector. In this step, the multiple terminal data collectors respectively and real-time detect the data of each dimension of the transformer, including but not limited to the operating noise, vibration, winding temperature, core temperature, low-voltage terminal temperature, transformer oil top temperature, oil pressure, oil level, ambient temperature, zero-sequence current and other data of the transformer.

[0074] S2: The data acquisition intelligent terminal processes and analyzes the multi-dimensional data of the transformer, and transmits the processed data to the cloud platform. In this step, after the data acquisition is completed, the data acquisition intelligent terminal processes and analyzes the data transmitted by each terminal data collector to obtain valuable information.

[0075] S3: The cloud platform aggregates and stores the data transmitted by each data acquisition intelligent terminal, and transmits it to the big data processing unit and the artificial intelligence unit. In this step, the cloud platform can store and process the data. Users can detect real-time data and view historical data through the cloud platform. The cloud platform can also provide users with a remote viewing function through the Internet.

[0076] S4: For a smaller data set, the big data processing unit calls a statistical model to further predict data points at future times based on historical time data and obtains a prediction result.

[0077] S5: For a larger data set, the artificial intelligence unit inputs the multi-dimensional data of the transformer into a selected deep learning model for artificial intelligence training, and analyzes and predicts whether the transformer has a fault according to the transformer knowledge graph and the artificial intelligence deep learning fault analysis and prediction algorithm.

[0078] After the above steps, by real-time collecting data such as the operating noise, vibration, winding temperature, core temperature, low-voltage terminal temperature, transformer oil top temperature, oil pressure, oil level, ambient temperature, zero-sequence current of the transformer, combining meteorological indicators such as the temperature and humidity of the transformer operating environment, the transformer operating data and fault data will be saved in the cloud platform, and a data analysis model will be established in the big data application layer. Through the combination of traditional and artificial intelligence algorithms for processing and analysis, transformer fault diagnosis is carried out, the operating conditions of the transformer are monitored. Once it is found that the operating status of the transformer has a deteriorating trend, an early warning will be issued in advance, realizing the early warning of the sub-healthy state of the transformer, and providing a basis for subsequent maintenance and other decisions.

[0079] Preferably, in step S2, after the data acquisition is completed, it is necessary to process and analyze the data to obtain valuable information. The process of processing and analyzing can be divided into the following steps:

[0080] S2-1: Data preprocessing: including data cleaning, outlier removal, and data normalization;

[0081] S2-2: Feature extraction: extracting meaningful feature data from the original data, including but not limited to RMS value, temperature value, liquid level value, or relative deviation;

[0082] S2-3: Data dimensionality reduction: mapping multi-dimensional data to a low-dimensional space to reduce the computational amount and better display the characteristics of the data.

[0083] Preferably, the statistical model is any one of autoregressive model, moving average model, autoregressive moving average model, autoregressive integrated moving average model, seasonal autoregressive integrated moving average model, seasonal autoregressive integrated moving average model with exogenous regressors, vector autoregressive model, or vector error correction model. In this step, the online state prediction method of the transformer is a process of further predicting data points at future times based on historical time data. This can be accomplished using statistical models, such as: autoregressive (AR) model, moving average (MA) model, autoregressive moving average (ARMA) model, autoregressive integrated moving average (ARIMA) model, seasonal autoregressive integrated moving average (SARIMA) model, seasonal autoregressive integrated moving average with exogenous regressors (SARIMAX) model, vector autoregressive (VAR) model, and vector error correction (VECM) model, etc.

[0084] Preferably, in step S5, the deep learning model is a multi-layer perceptron, recurrent neural network, long short-term memory network, autoregressive model, or convolutional neural network model.

[0085] Preferably, the deep learning model is a convolutional neural network model. In this embodiment, Convolutional Neural Networks (CNN) is a class of deep feedforward neural networks with convolution and pooling operations at the core. At the beginning of its design, it was used to solve the problem of image recognition in the field of computer vision. The principle of using a convolutional neural network for time series prediction is to utilize the ability of the convolutional kernel to sense the changes in a period of historical data and make predictions based on the changes in this historical data. The pooling operation can retain key information and reduce information redundancy. The convolutional neural network can effectively reduce the human resource consumption of feature extraction by previous algorithms and avoid the generation of human errors. Time series prediction based on Convolutional Neural Networks (CNN) is a method of using a CNN model to process time series data and make predictions. Compared with traditional methods based on Recurrent Neural Networks (RNN), CNN has some unique advantages when dealing with time series data.

[0086] Time series prediction based on a convolutional neural network model includes the following steps:

[0087] Data preparation: Divide the time series data set into a training set and a test set. The training set is used to train the convolutional neural network model, and the test set is used to evaluate the prediction performance of the model.

[0088] Data transformation: Use the sliding window method, Fourier transform method or other methods to transform the time series data into two-dimensional image data.

[0089] Construction of the convolutional neural network model: Construct a convolutional neural network model including convolutional layers, pooling layers and fully connected layers. The convolutional layers are used to extract features from the time series data, the pooling layers are used to reduce the feature dimension, and the fully connected layers are used to output the prediction results.

[0090] Model training: Use the training set to train the convolutional neural network model, and update the weights and biases of the model through the backpropagation algorithm to minimize the prediction error.

[0091] Model prediction: Use the trained convolutional neural network model to predict the test set. Input the image data of the test set and obtain the prediction results through the forward propagation algorithm.

[0092] Model evaluation: Compare the prediction results with the true values of the test set, calculate indexes such as prediction error and root mean square error, and evaluate the prediction performance of the model.

[0093] As can be seen from the above technical solutions, the present invention has the following advantages: The present invention collects various data during the operation of the transformer in real time through the terminal data collector, combines meteorological indexes such as temperature and humidity of the transformer operation environment, and performs processing and analysis through the combination of statistical model analysis and artificial intelligence algorithms to carry out transformer fault diagnosis and monitor the operation of the transformer. Once it is found that the operation state of the transformer has a deteriorating trend, an early warning is issued in advance to realize the early warning of the sub-healthy state of the transformer.

[0094] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the embodiments of the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An on-line health diagnosis system for a distribution transformer, characterized in that, It includes a low-voltage power distribution area, a cloud platform, and a big data application layer; The low-voltage power distribution area includes a data acquisition intelligent terminal (1) and several terminal data collectors (2); the terminal data collectors are used to collect the operation status data of the transformer and transmit the collected operation status data to the data acquisition intelligent terminal; the data acquisition intelligent terminal processes and analyzes the operation status data and transmits the processing results to the cloud platform; The cloud platform includes a management and monitoring platform (3) and several workstations (4); the management and monitoring platform receives the operation status data and updates the operation status information of the transformer in real time; the workstations maintain the management and monitoring platform; The big data application layer includes a big data processing unit (5), an Internet of Things unit (6), and an artificial intelligence unit (7); the big data processing unit is connected to the management and monitoring platform and processes and analyzes the real-time updated data provided by the management and monitoring platform; the Internet of Things unit is connected to the management and monitoring platform and manages the devices connected to it; the artificial intelligence unit is respectively connected to the big data processing unit, the Internet of Things unit, and the management and monitoring platform, and is used to analyze and predict whether there is a fault in the transformer by using the transformer knowledge graph and the artificial intelligence deep learning fault analysis and prediction algorithm.

2. The on-line health diagnosis system for a distribution transformer according to claim 1, characterized in that, The terminal data collector is any one of a temperature collector, a humidity sensor, a dissolved gas sensor, a vibration collector, a voiceprint acquisition sensor, an infrared sensor, an ultrasonic sensor, a thermal imaging sensor, a transformer oil concentration sensor, a transformer oil temperature sensor, a transformer oil pressure sensor, or a transformer oil level sensor.

3. The on-line health diagnosis system for distribution transformers according to claim 1, characterized in that The data acquisition intelligent terminal includes an STM32 embedded CPU (11), and a 4G communication module (12), a Bluetooth wireless communication module (13), an RS485 wired communication (14), a liquid crystal display module (15), a key (16), and an 8-channel analog signal acquisition channel (17) respectively connected to the STM32 embedded CPU.

4. The on-line health diagnosis system for a distribution transformer according to claim 3, characterized in that, A wireless communication module is provided on the terminal data collector, and the terminal data collector is connected to an adjacent terminal data collector in a wireless communication manner through the wireless communication module; the terminal data collector is connected to the data acquisition intelligent terminal in a wireless communication manner.

5. The on-line health diagnosis system for a distribution transformer according to claim 3, characterized in that The terminal data collector and the data acquisition intelligent terminal are connected through an industrial field bus.

6. An on-line health diagnosis method for a distribution transformer, which is used in the on-line health diagnosis system of the distribution transformer according to any one of claims 1 to 5, characterized in that, It includes the following steps: S1: The operation status data of each dimension of the transformer are respectively collected by multiple terminal data collectors, and the data acquisition intelligent terminal collects the data of each terminal data collector; S2: The data acquisition intelligent terminal processes and analyzes the multi-dimensional data of the transformer and transmits the processed data to the cloud platform; S3: The cloud platform aggregates and stores the data transmitted by each data acquisition intelligent terminal and transmits them to the big data processing unit and the artificial intelligence unit; S4: For a relatively small data set, the big data processing unit calls a statistical model to further predict data points at a later time based on historical time data and obtains a prediction result; S5: For a relatively large dataset, the artificial intelligence unit inputs the multi-dimensional data of the transformer into a selected deep learning model for artificial intelligence training, and analyzes and predicts whether there are faults in the transformer according to the transformer knowledge graph and the artificial intelligence deep learning fault analysis and prediction algorithm.

7. The on-line health diagnosis method for a distribution transformer according to claim 6, characterized in that, In step S2, the process of processing and analyzing can be divided into the following steps: S2-1: Data preprocessing: including data cleaning, outlier removal, and data normalization processing; S2-2: Feature extraction: extracting meaningful feature data from the original data, including but not limited to RMS value, temperature value, liquid level value, or relative deviation; S2-3: Data dimensionality reduction: mapping multi-dimensional data to a low-dimensional space to reduce the computational amount and better display the characteristics of the data.

8. The on-line health diagnosis method for a distribution transformer according to claim 6, characterized in that, The statistical model is any one of autoregressive model, moving average model, autoregressive moving average model, autoregressive integrated moving average model, seasonal autoregressive integrated moving average model, seasonal autoregressive integrated moving average model with exogenous regressors, vector autoregressive model, or vector error correction model.

9. The on-line health diagnosis method of a distribution transformer according to claim 6, characterized in that, In step S5, the deep learning model is a multi-layer perceptron, recurrent neural network, long short-term memory network, autoregressive model, or convolutional neural network model.

10. The on-line health diagnosis method of a distribution transformer according to claim 9, characterized in that, The deep learning model is a convolutional neural network model. The time series prediction based on the convolutional neural network model includes the following steps: Data preparation: dividing the time series dataset into a training set and a test set. The training set is used to train the convolutional neural network model, and the test set is used to evaluate the prediction performance of the model; Data conversion: using the sliding window method, Fourier transform method, or other methods to convert the time series data into two-dimensional image data; Convolutional neural network model construction: constructing a convolutional neural network model including convolutional layers, pooling layers, and fully connected layers, where the convolutional layers are used to extract features in the time series data, the pooling layers are used to reduce the feature dimension, and the fully connected layers are used to output the prediction results; Model training: using the training set to train the convolutional neural network model, and updating the weights and biases of the model through the backpropagation algorithm to minimize the prediction error; Model prediction: using the trained convolutional neural network model to predict the test set, inputting the image data of the test set, and obtaining the prediction results through the forward propagation algorithm; Model evaluation: comparing the prediction results with the true values of the test set, calculating indexes such as prediction error and root mean square error, and evaluating the prediction performance of the model.

Images