Transformer insulating oil condition monitoring system and method based on cloud-edge-end collaboration

Abstract

The present invention belongs to the field of smart grids and provides a transformer insulating oil status monitoring system and method based on cloud-edge-end collaboration. The system includes: a device end, an edge end, and a cloud service end; the device end is used to regularly collect multiple operating status data of the transformer to be tested; the edge end is used to perform a preliminary diagnosis on the oil chromatography data in the target operating status data, obtain a preliminary diagnosis result, and upload it to the cloud service end; the cloud service end is used to input the oil chromatography data into a gas concentration prediction model to obtain a prediction result of the dissolved gas concentration in the transformer insulating oil, perform a secondary diagnosis on the fault data not defined in the preliminary diagnosis result, obtain a secondary diagnosis result, and determine the insulating oil status monitoring result of the transformer to be tested. The solution provided by the present invention can realize accurate monitoring of the insulating oil status of the transformer to be tested because the monitoring process is implemented based on the cloud-edge-end collaborative architecture, combined with the gas concentration prediction model and the secondary diagnosis link.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to a transformer insulating oil status monitoring system and method based on cloud-edge-end collaboration. Background Art

[0002] At present, transformer insulating oil condition monitoring technology mainly relies on online oil chromatography equipment to perform real-time detection and analysis of transformer insulating oil. When the online oil chromatography data continuously issues abnormal reminders for a period of time, the operation and maintenance personnel will arrange multiple insulating oil offline detection tests to review the abnormal reminders. The professional and technical personnel will then conduct a comprehensive analysis based on the test results and formulate corresponding inspection and maintenance plans based on the final manual diagnosis results.

[0003] Since the transformer insulating oil condition monitoring link relies too much on manual detection and analysis, the monitoring process is time-consuming and labor-intensive, and there are too many human interference factors, resulting in insufficient accuracy and reliability of the transformer insulating oil condition monitoring results.

[0004] It can be seen that the traditional transformer insulating oil condition monitoring solution has the problems of low monitoring efficiency, insufficient monitoring accuracy and reliability. Summary of the Invention

[0005] The present invention provides a transformer insulating oil condition monitoring system and method based on cloud-edge-end collaboration, which is used to solve the defects of traditional transformer insulating oil condition monitoring solutions, such as low monitoring efficiency, insufficient monitoring accuracy and reliability.

[0006] On the one hand, the present invention provides a transformer insulating oil condition monitoring system based on cloud-edge-end collaboration, comprising: a device end, an edge end, and a cloud service end;

[0007] The cloud service end and the device end are both connected to the edge end;

[0008] The device end is used to regularly collect multiple operating status data of the transformer under test according to the voltage level of the transformer under test; wherein the multiple operating status data include oil chromatogram data, oil withstand voltage value, dielectric loss factor and volume resistivity;

[0009] The edge end is used to obtain target operating status data from the device end according to the data detection instruction issued by the cloud service end, perform preliminary diagnosis on the oil chromatography data in the target operating status data, obtain preliminary diagnosis results, and upload the preliminary diagnosis results and target operating status data to the cloud service end;

[0010] The cloud service end is used to receive the preliminary diagnosis result and the target operating status data, input the oil chromatography data in the target operating status data into a pre-built gas concentration prediction model, obtain the prediction result of the dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model, perform a secondary diagnosis on the fault data not defined in the preliminary diagnosis result to obtain a secondary diagnosis result, and determine the insulating oil status monitoring result of the transformer to be tested based on the dissolved gas concentration prediction result, the secondary diagnosis result, the oil withstand voltage value, dielectric loss factor and volume resistivity in the target operating status data, and other key detection values ​​obtained in advance.

[0011] According to the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the present invention, the device side includes: oil chromatography detection equipment, insulating oil withstand voltage detection equipment, dielectric loss detection equipment and data sensing equipment;

[0012] The oil chromatography detection device, the insulating oil withstand voltage detection device, and the dielectric loss detection device are respectively connected to corresponding data sensing devices, and the data sensing devices are connected to the edge end;

[0013] The oil chromatography detection equipment is used to detect the oil chromatography data of the transformer insulating oil to be tested, the insulating oil withstand voltage detection equipment is used to detect the oil withstand voltage value of the transformer insulating oil to be tested, and the dielectric loss detection equipment is used to detect the dielectric loss factor and volume resistivity of the transformer insulating oil to be tested;

[0014] The data sensing device is used to receive data detected by a corresponding detection device and send the detected data to the edge end.

[0015] According to the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the present invention, the oil chromatography detection equipment includes: an online chromatography detector installed at the monitoring site and an offline chromatography detector installed in a standard laboratory environment;

[0016] The online chromatogram detector is used to detect online oil chromatogram data of the transformer insulating oil to be tested, and the offline chromatogram detector is used to perform offline testing on the transformer insulating oil to be tested obtained by regular sampling to obtain offline oil chromatogram data.

[0017] According to the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the present invention, the edge end includes: a first edge industrial computer, a second edge industrial computer, a third edge industrial computer and a fourth edge industrial computer;

[0018] The first edge industrial computer, the second edge industrial computer, and the third edge industrial computer are respectively connected to their corresponding data sensing devices, the first edge industrial computer, the second edge industrial computer, and the third edge industrial computer are all connected to the fourth edge industrial computer, and the fourth edge industrial computer is connected to the cloud server;

[0019] The first edge industrial computer is configured to obtain the oil chromatography data in the target operating status data according to the data detection instruction issued by the cloud server, perform a preliminary diagnosis on the oil chromatography data, and obtain a preliminary diagnosis result;

[0020] The second edge industrial computer is used to obtain the oil pressure resistance value in the target operating status data according to the data detection instruction issued by the cloud server;

[0021] The third edge industrial computer is used to obtain the dielectric loss factor and volume resistivity in the target operating status data according to the data detection instruction issued by the cloud service end;

[0022] The fourth edge industrial computer is used to use the preliminary diagnostic results, oil chromatography data, oil withstand voltage value, dielectric loss factor and volume resistivity as table content to generate a data collection report, and upload the data collection report to the cloud service end.

[0023] According to the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the present invention, the cloud service end includes:

[0024] A concentration prediction module is used to input the oil chromatography data in the target operating state data into a pre-built gas concentration prediction model to obtain a prediction result of the dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model;

[0025] A secondary diagnosis module, configured to perform a secondary diagnosis on the fault data not defined in the primary diagnosis result to obtain a secondary diagnosis result;

[0026] The condition monitoring module is used to determine the insulating oil condition monitoring result of the transformer to be tested based on the dissolved gas concentration prediction result, the secondary diagnosis result, the oil withstand voltage value, dielectric loss factor and volume resistivity in the target operating status data, and other key detection values.

[0027] According to the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the present invention, the gas concentration prediction model includes:

[0028] Input layer, used to receive oil chromatography data;

[0029] an attention layer, configured to assign corresponding attention weights to different data elements in the oil chromatogram data;

[0030] a bidirectional gated recurrent layer, configured to perform bidirectional feature extraction on the oil chromatogram data according to the attention weights to obtain hidden state data;

[0031] The output layer is used to generate and output a prediction result of the dissolved gas concentration in the transformer insulating oil based on the hidden state data.

[0032] According to the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the present invention, the secondary diagnosis module performs secondary diagnosis on the undefined fault data in the preliminary diagnosis result to obtain a secondary diagnosis result, including:

[0033] Obtain oil chromatogram samples and corresponding fault code samples of the transformer to be tested;

[0034] Establishing a training sample set and a test sample set based on the oil chromatogram sample and the fault code sample;

[0035] The pre-built deep extreme learning machine model is trained using the training sample set, and the trained deep extreme learning machine model is tested using the test sample set to obtain a secondary diagnosis model;

[0036] The undefined fault data in the primary diagnosis result is input into the secondary diagnosis model to obtain a secondary diagnosis result output by the secondary diagnosis model.

[0037] According to the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the present invention, a training sample set and a test sample set are established based on the oil chromatogram sample and the fault coding sample, including:

[0038] constructing an input feature vector according to the oil chromatogram sample;

[0039] Inputting the input feature vector into a pre-built data enhancement model to obtain an oil chromatogram amplification sample output by the data enhancement model;

[0040] Establishing a sample data set based on the oil chromatogram amplification sample and the fault code sample;

[0041] The sample data set is divided according to a preset ratio to obtain a training sample set and a test sample set.

[0042] According to the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the present invention, an input feature vector is constructed based on the oil chromatogram sample, including:

[0043] performing normalization processing on the oil chromatogram sample to obtain oil chromatogram normalized data;

[0044] extracting characteristic gas content data from the oil chromatogram sample;

[0045] constructing a gas ratio graph based on the characteristic gas content data to obtain gas graphic data;

[0046] Performing three-ratio encoding on the characteristic gas content data to obtain three-ratio encoded data;

[0047] An input feature vector is constructed according to the oil chromatogram normalized data, the gas graph data, and the three-ratio encoded data.

[0048] On the other hand, the present invention also provides a transformer insulating oil condition monitoring method based on cloud-edge-end collaboration, which is executed by a cloud service end, the cloud service end is connected to an edge end, and the edge end is connected to a device end; the method includes:

[0049] Sending a data detection instruction to the edge end, so that the edge end obtains target operating status data from the device end according to the data detection instruction, and performs a preliminary diagnosis on the oil chromatogram data in the target operating status data to obtain a preliminary diagnosis result; wherein the target operating status data is obtained by regularly collecting multiple operating status data of the transformer to be tested from the device end according to the voltage level of the transformer to be tested; the multiple operating status data include oil chromatogram data, oil withstand voltage value, dielectric loss factor, and volume resistivity;

[0050] Receive preliminary diagnostic results and target operating status data uploaded by the edge;

[0051] Inputting the oil chromatogram data in the target operating state data into a pre-built gas concentration prediction model to obtain a prediction result of dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model;

[0052] Performing a secondary diagnosis on the fault data not defined in the preliminary diagnosis result to obtain a secondary diagnosis result;

[0053] The insulating oil status monitoring result of the transformer to be tested is determined based on the dissolved gas concentration prediction result, the secondary diagnosis result, the oil withstand voltage value, dielectric loss factor and volume resistivity in the target operating status data, and other key detection values ​​obtained in advance.

[0054] The present invention provides a transformer insulating oil condition monitoring system and method based on cloud-edge-end collaboration. By setting a device end, an edge end and a cloud service end, the device end regularly collects multiple operating status data such as oil chromatography data, oil withstand voltage value and dielectric loss factor of the transformer to be tested according to the voltage level of the transformer to be tested. The edge end obtains the target operating status data from the device end according to the data detection instruction issued by the cloud service end, performs a preliminary diagnosis on the oil chromatography data in the target operating status data, and obtains a preliminary diagnosis result. The cloud service end inputs the oil chromatography data in the target operating status data into a pre-constructed gas concentration prediction model to obtain a prediction result of the dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model, performs a secondary diagnosis on the fault data not defined in the preliminary diagnosis result, and obtains a secondary diagnosis result. The insulating oil condition monitoring result of the transformer to be tested is determined based on the dissolved gas concentration prediction result, the secondary diagnosis result, the oil withstand voltage value, dielectric loss factor and volume resistivity in the target operating status data, and other key detection values ​​obtained in advance. Since the monitoring process is implemented based on a cloud-edge-end collaborative architecture, combined with the gas concentration prediction model and secondary diagnosis link, accurate monitoring of the insulating oil status of the transformer under test can be achieved, the monitoring efficiency is effectively improved, and the monitoring results are more accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 Schematic diagram of the structure of a transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by an embodiment of the present invention;

[0057] Figure 2 It is a schematic diagram of the process of conducting preliminary diagnosis on oil chromatography data;

[0058] Figure 3 It is a schematic diagram of the distribution of the number of samples under each fault type;

[0059] Figure 4 It is a line graph of dissolved gas content in insulating oil obtained by statistical analysis of offline oil chromatogram data;

[0060] Figure 5 It is a schematic diagram of the structure and construction process of the gas concentration prediction model;

[0061] Figure 6 It is a schematic diagram of the data enhancement principle;

[0062] Figure 7 This is a schematic diagram of the implementation principle of the secondary diagnosis link;

[0063] Figure 8 This is a flow chart of the transformer insulating oil status monitoring method based on cloud-edge-end collaboration provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0065] This embodiment relates to the field of smart grids and can be specifically applied to monitoring the status of transformer insulating oil. Currently, transformer insulating oil operating status monitoring technology primarily relies on online oil chromatography equipment for real-time testing and analysis of transformer oil. When the online oil chromatography continuously issues abnormality alerts over a period of time, operations and maintenance personnel will conduct multiple offline insulating oil testing tests to verify the abnormality alerts. Professional technicians will then conduct a comprehensive analysis of the test results and formulate a corresponding maintenance plan based on the final human diagnosis.

[0066] In recent years, demand for online oil chromatography equipment has gradually increased, and its functions have also improved to a certain extent. However, in actual applications, power accidents caused by online oil chromatography equipment still occur. During some holidays and special power supply guarantee periods, some substations or power plants will choose to temporarily shut down their equipment to prevent safety accidents. Therefore, the safety of online oil chromatography equipment needs to be improved, and the monitoring of transformer insulating oil status needs to be further improved on this basis.

[0067] In addition, when monitoring transformer insulating oil, operation and maintenance personnel generally focus on recent test data. However, the life of oil-immersed power transformers is generally up to 30 years. The single test data is only used to determine whether it meets the attention value specified in the standard. It is difficult to observe its future changes and cannot assist in judging the development trend of the equipment. To a certain extent, it has caused a waste of power data resources.

[0068] In view of the shortcomings of the above-mentioned traditional transformer insulating oil condition monitoring solution, this embodiment provides a corresponding improvement solution. Figures 1-8 Describe the detailed scheme of the transformer insulating oil condition monitoring system and method based on cloud-edge-end collaboration provided by an embodiment of the present invention.

[0069] Figure 1This is a structural diagram of a transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by an embodiment of the present invention.

[0070] like Figure 1 As shown, the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by an embodiment of the present invention includes: a device end 110, an edge end 120 and a cloud service end 130.

[0071] The cloud service end 130 and the device end 110 are both connected to the edge end 120 .

[0072] The device end 110 is used to regularly collect multiple operating status data of the transformer under test according to the voltage level of the transformer under test; wherein the multiple operating status data include oil chromatogram data, oil withstand voltage value, dielectric loss factor and volume resistivity.

[0073] The edge end 120 is used to obtain the target operating status data from the device end 110 according to the data detection instruction issued by the cloud service end 130, perform a preliminary diagnosis on the oil chromatography data in the target operating status data, obtain a preliminary diagnosis result, and upload the preliminary diagnosis result and the target operating status data to the cloud service end 130.

[0074] The cloud service terminal 130 is used to receive the preliminary diagnosis results and the target operating status data, input the oil chromatography data in the target operating status data into a pre-built gas concentration prediction model, obtain the prediction result of the dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model, perform a secondary diagnosis on the fault data not defined in the preliminary diagnosis results, obtain the secondary diagnosis result, and determine the insulating oil status monitoring result of the transformer to be tested based on the dissolved gas concentration prediction result, the secondary diagnosis result, the oil withstand voltage value, the dielectric loss factor and the volume resistivity in the target operating status data, and other key detection values ​​obtained in advance.

[0075] like Figure 1 As shown, the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration in this embodiment is mainly composed of three parts, namely the device end 110, the edge end 120 and the cloud service end 130. By distributing data preprocessing and preliminary diagnosis functions to the edge end 120 close to the device end 110, and utilizing the powerful data storage and analysis capabilities of the cloud service end 130, and coordinating with the application of intelligent models, the transformer insulating oil condition monitoring scheme is optimized, which can achieve accurate monitoring of the insulating oil condition of the transformer to be tested, effectively improve the monitoring efficiency, and make the monitoring results more accurate and reliable.

[0076] In one embodiment, the equipment side includes: oil chromatography detection equipment, insulating oil withstand voltage detection equipment, dielectric loss detection equipment and data sensing equipment.

[0077] The oil chromatography detection equipment, the insulating oil withstand voltage detection equipment and the dielectric loss detection equipment are respectively connected to their corresponding data sensing devices, and the data sensing devices are connected to the edge end.

[0078] The oil chromatography testing equipment is used to detect the oil chromatography data of the transformer insulating oil to be tested, the insulating oil withstand voltage testing equipment is used to detect the oil withstand voltage value of the transformer insulating oil to be tested, and the dielectric loss testing equipment is used to detect the dielectric loss factor and volume resistivity of the transformer insulating oil to be tested.

[0079] The data sensing device is used to receive the data detected by the corresponding detection device and send the detected data to the edge end.

[0080] In this embodiment, each detection device may be configured with a corresponding data sensing device, thereby ensuring synchronous and stable transmission of front-end data.

[0081] In one embodiment, the oil chromatography detection equipment specifically includes: an online chromatography detector installed at the monitoring site and an offline chromatography detector installed in a standard laboratory environment.

[0082] The online chromatogram detector is used to detect the online oil chromatogram data of the transformer insulating oil to be tested, and the offline chromatogram detector is used to perform offline testing on the transformer insulating oil to be tested obtained by regular sampling to obtain offline oil chromatogram data.

[0083] It should be noted that the oil chromatogram data in this embodiment includes not only real-time online oil chromatogram data, but also offline oil chromatogram data obtained by an offline chromatogram detector, which can ensure the integrity and reliability of the insulating oil chromatogram data.

[0084] In some embodiments, the device side may also be equipped with more types of detection equipment such as a copper ion detector and an antioxidant detector to obtain more comprehensive front-end data.

[0085] In one embodiment, the edge end specifically includes: a first edge industrial computer, a second edge industrial computer, a third edge industrial computer, and a fourth edge industrial computer.

[0086] The first edge industrial computer, the second edge industrial computer and the third edge industrial computer are respectively connected to their corresponding data sensing devices, the first edge industrial computer, the second edge industrial computer and the third edge industrial computer are all connected to the fourth edge industrial computer, and the fourth edge industrial computer is connected to the cloud service end.

[0087] The first edge industrial computer is used to obtain the oil chromatography data in the target operating status data according to the data detection instructions issued by the cloud server, perform preliminary diagnosis on the oil chromatography data, and obtain preliminary diagnosis results.

[0088] The second edge industrial computer is used to obtain the oil pressure value in the target operating status data based on the data detection instructions issued by the cloud server.

[0089] The third edge industrial computer is used to obtain the dielectric loss factor and volume resistivity in the target operating status data based on the data detection instructions issued by the cloud server.

[0090] The fourth edge industrial computer is used to generate data collection reports using preliminary diagnostic results, oil chromatography data, oil withstand voltage value, dielectric loss factor and volume resistivity as table contents, and upload the data collection reports to the cloud server.

[0091] In this embodiment, each detection device is equipped with an edge industrial computer, enabling simultaneous acquisition of multiple operational status data points and providing edge computing services for all detection devices. In some embodiments, the first edge industrial computer can also perform data preprocessing operations on the oil chromatogram data, such as removing abnormal data and filling missing data.

[0092] It can be understood that oil chromatography data refers to the data obtained by analyzing the content of dissolved gas components in insulating oil through gas chromatography, mainly including the content data of various gases such as hydrogen, oxygen, nitrogen, methane, ethane, ethylene and acetylene.

[0093] The data collection report can summarize the preliminary diagnosis results, oil chromatography data, oil pressure resistance value, dielectric loss factor, volume resistivity and other data in the form of a table, and display them clearly and intuitively.

[0094] In practical applications, the three-ratio method can be used to make preliminary diagnosis of oil chromatography data. Figure 2 The following is an example of a process for preliminary diagnosis of oil chromatogram data, specifically including:

[0095] Step 210: Input oil chromatogram data, which includes content data of various gases. At the same time, it is necessary to pre-set the standard attention value corresponding to each gas content data.

[0096] Step 220: Compare the content data of each gas in the oil chromatogram data with the corresponding standard caution value to determine whether the content data of each gas exceeds the standard caution value.

[0097] Step 230: If the judgment result of step 220 is no, it means that the data is normal.

[0098] Step 240: If the result of step 220 is yes, the data is defective upon verification. Subsequently, it is necessary to determine the abnormal gas content data exceeding the standard warning value. The fault type is then preliminarily determined using the three-ratio method based on the abnormal gas content data. Specifically, the fault type can be determined by the coding combination of three pairs of abnormal gas content ratios in the abnormal gas content data. For example, coding combination 001 corresponds to low-temperature overheating, and coding combination 012 corresponds to medium-temperature overheating, thereby obtaining a preliminary diagnosis result.

[0099] Figure 2 Various fault types are shown, including low-temperature overheating (below 150°C), low-temperature overheating (150°C-300°C), medium-temperature overheating (300°C-700°C), high-temperature overheating (above 700°C), partial discharge, low-energy discharge and overheating, low-energy discharge, arc discharge, and arc discharge and overheating. There are also undefined fault types. Figure 3 The figure shows, by way of example, the distribution of the number of samples under each fault type after the abnormal gas content data are classified according to different fault types.

[0100] In this embodiment, an on-site inspection terminal can also be installed at the edge. This terminal can be a smartphone, tablet, or other mobile device, making it easy for staff to carry during on-site inspections. A diagnostic app can be configured within the on-site inspection terminal. The app provides functions such as abnormal status determination of the transformer under test, trend prediction, data consolidation, data browsing, and inspection prompts. It also has a built-in hardware security encryption unit to improve the security and reliability of information transmission.

[0101] The on-site inspection terminal is mainly used to complete the regular offline experimental testing tasks of the insulating oil of the transformer to be tested. By taking oil samples on-site and sending them to the laboratory for testing, the dissolved gas content data in the insulating oil can be tested offline. After uploading the offline oil chromatography data to the cloud server, the cloud server can perform statistical analysis on the offline oil chromatography data, thereby realizing regular inspections of the operating status of the transformer to be tested. Figure 4 The following is an exemplary diagram showing the dissolved gas content in insulating oil obtained by statistically analyzing the offline oil chromatogram data.

[0102] In one embodiment, the cloud service terminal specifically includes:

[0103] The concentration prediction module is used to receive the preliminary diagnosis results and target operating status data, input the oil chromatography data in the target operating status data into a pre-built gas concentration prediction model, and obtain the prediction result of the dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model.

[0104] The secondary diagnosis module is used to perform secondary diagnosis on the fault data not defined in the primary diagnosis result to obtain the secondary diagnosis result.

[0105] The condition monitoring module is used to determine the insulating oil condition monitoring results of the transformer to be tested based on the dissolved gas concentration prediction results, the secondary diagnosis results, the oil withstand voltage value, dielectric loss factor, dielectric loss factor, volume resistivity in the target operating status data, and other key detection values ​​obtained in advance.

[0106] In this embodiment, the concentration prediction module mainly uses the gas concentration prediction model to realize the prediction function of the dissolved gas concentration in the transformer insulating oil. The concentration prediction module is implemented using a machine learning model architecture. The input of the concentration prediction module is oil chromatography data, and the output is the prediction result of the dissolved gas concentration in the transformer insulating oil.

[0107] In one embodiment, the gas concentration prediction model specifically includes:

[0108] Input layer, used to receive oil chromatogram data.

[0109] The attention layer is used to assign corresponding attention weights to different data elements in the oil chromatogram data.

[0110] The bidirectional gated recurrent layer is used to perform bidirectional feature extraction on the oil chromatography data according to the attention weights to obtain hidden state data.

[0111] The output layer is used to generate and output the prediction results of dissolved gas concentration in transformer insulating oil based on the hidden state data.

[0112] like Figure 5 As shown, the input layer 310 is mainly used to input oil chromatogram data, specifically receiving the data sequence corresponding to the oil chromatogram data. Figure 5 X1, X2, X3, ..., X t Represents different data elements in a data sequence.

[0113] The attention layer 320 uses an attention mechanism to enable the prediction model to focus on a small number of key information, thereby improving the overall performance of the model. Figure 5 The middle attention layer 320 can convert the attention weights e1, e2, e3, ..., e t Assigned to the corresponding data element.

[0114] The Bidirectional Gated Recurrent Unit (Bi-GRU) layer 330 is a variation of the Long Short Term Memory (LSTM) network. It retains the functions of the LSTM network, but has a simpler structure and is less prone to overfitting. The GRU part changes the gate structure, replacing the three-gate structure of the LSTM network with an update gate and a reset gate. While ensuring the same prediction accuracy as the LSTM network, it reduces memory usage and network training time. The Bidirectional Gate Recurrent Unit (Bi-GRU) layer uses bidirectional gate recurrent units to capture the complex relationships of time series, and enhances the ability of neural networks to extract spatiotemporal features of data through a combined model. In practical applications, the bidirectional gated recurrent layer 330 uses forward and reverse GRU units to extract and process data features, capturing contextual information. Figure 5 In the formula, h1', h2', h3', and ht' represent the hidden states of the forward process, h1", h2", h3", and ht' represent the hidden states of the reverse process, and h1, h2, h3, ..., and ht represent the outputs of the bidirectional gated recurrent layer 330. The final output layer 340 outputs the predicted results of the dissolved gas concentration in the transformer insulating oil {y1, y2, y3, y4, ..., y t}.

[0115] It is understandable that, after determining the above model architecture, the gas concentration prediction model needs to be trained and tested using sample data to obtain a gas concentration prediction model that can be applied to actual prediction scenarios.

[0116] like Figure 5 As shown, in the sample generation stage 350, previously acquired transformer monitoring time series data, i.e., historical online oil chromatogram data, can be obtained. After subsequent data standardization processing, a sample data set is obtained, and then the sample data set is divided according to a preset ratio to obtain a training set and a test set.

[0117] During the model training and testing phase 360, to ensure the continued effectiveness of the model, this embodiment dynamically monitors the error of the gas fitting curve generated based on the prediction results output by the gas concentration prediction model. The obtained error is compared with the dynamic threshold value of the current state to determine whether the obtained error is within the threshold range. If the error is within the threshold range, the model state can be determined to be normal. If the error is not within the threshold range, the trend similarity between the actual value and the predicted value can be further analyzed to determine whether the trend similarity continues to decrease. If the trend similarity continues to decrease, the model mismatch can be determined. If the trend similarity does not continue to decrease, a fault warning can be output. In addition, by inputting test data from the test set into the gas concentration prediction model, the prediction results can be evaluated multidimensionally using multidimensional evaluation indicators. In this embodiment, the multidimensional evaluation indicators include the cross-correlation coefficient, mean absolute percentage error, and MSE (mean squared error). The effectiveness of the model is determined by analyzing the prediction sequence.

[0118] In one embodiment, the secondary diagnosis module performs secondary diagnosis on the fault data not defined in the primary diagnosis result to obtain a secondary diagnosis result, specifically including:

[0119] The first step is to obtain the oil chromatogram sample and the corresponding fault code sample of the transformer to be tested.

[0120] In this embodiment, the oil chromatogram sample refers to offline oil chromatogram data of historical detection, and the offline oil chromatogram data can be detected by an offline oil chromatogram detector installed in a standard laboratory environment.

[0121] In the second step, based on the oil chromatography samples and fault coding samples, a training sample set and a test sample set are established.

[0122] In a specific implementation, a training sample set and a test sample set are established based on the oil chromatogram samples and the fault code samples, specifically including:

[0123] First, the input feature vector is constructed based on the oil chromatogram sample.

[0124] In this embodiment, an input feature vector is constructed based on the oil chromatogram sample, specifically including:

[0125] On the one hand, the oil chromatography samples are normalized to obtain oil chromatography normalized data.

[0126] On the other hand, characteristic gas content data in the oil chromatogram sample is extracted; a gas ratio graph is constructed based on the characteristic gas content data to obtain gas graphic data.

[0127] On the other hand, the characteristic gas content data is triple-ratio encoded to obtain triple-ratio encoded data.

[0128] Finally, the input feature vector is constructed based on the oil chromatogram normalized data, gas graph data and three-ratio encoding data.

[0129] Then, the input feature vector is input into the pre-built data augmentation model to obtain the oil chromatogram amplification sample output by the data augmentation model.

[0130] like Figure 6 As shown, the data augmentation model includes a generator 410 and a judge 420. Before data augmentation, a series of data preprocessing operations are first performed on the oil chromatogram sample containing the content data of gases such as H2, CH4, C2H6, C2H4, and C2H2 to obtain the input feature vector corresponding to the real sample. At the same time, random noise is input into the generator 410. After the sample is generated, the generated sample and the input feature vector are both input into the judge 420. The judge 420 determines whether Nash equilibrium is reached. If the judgment result is no, the Wasserstein distance and "true, false" judgment loss regression are introduced to continue generating samples through the generator 410 until the judgment result is yes. The quality of the generated sample is evaluated by FID (Fréchet Inception Distance) and MMD (Maximum Mean Discrepancy), and finally the oil chromatogram amplified sample is obtained.

[0131] It's understandable that before training the secondary diagnosis model, since the distribution of oil chromatogram samples is uneven, it's necessary to amplify those with fewer occurrences to make the sample distribution more even. Therefore, data augmentation can improve sample imbalance. In practical applications, a WGAN (Wasserstein Generative Adversarial Network) model can be used as the data augmentation model. This model can generate samples with fewer fault types, making the sample dataset more balanced.

[0132] The graphical method used in constructing the gas ratio diagram serves as a supplementary method to the three-ratio method, allowing for further differentiation between similar fault types. The five-gas graphical method uses H2, CH4, C2H6, C2H4, and C2H2 as the horizontal axis, with the highest concentration as 1, and the ratio of each component's concentration to this highest concentration as the vertical axis. The gas ratio diagram is a complementary diagram to the five-gas three-ratio method, facilitating the identification and recording of characteristics.

[0133] Subsequently, a sample data set is established based on the oil chromatogram amplification samples and fault coding samples.

[0134] Finally, the sample data set is divided according to the preset ratio to obtain the training sample set and the test sample set.

[0135] In the third step, the pre-built deep extreme learning machine model is trained through the training sample set, and the trained deep extreme learning machine model is tested through the test sample set to obtain a secondary diagnostic model.

[0136] The fourth step is to input the undefined fault data in the preliminary diagnosis result into the secondary diagnosis model to obtain the secondary diagnosis result output by the secondary diagnosis model.

[0137] In this embodiment, the secondary diagnosis link is mainly aimed at the situation where there is an undefined fault type after the initial diagnosis. Figure 7 The implementation principle of the secondary diagnosis link is shown as an example. Figure 7 As shown in the figure, the implementation process of the secondary diagnosis link is as follows:

[0138] Step 510: establishing a sample data set based on the oil chromatogram amplified samples and fault code samples obtained after the oil chromatogram sample data is enhanced, and dividing the sample data set into a training sample set and a test sample set.

[0139] Step 520: Establish a Deep Extreme Learning Machine (DELM) model and optimize the model parameters using a dynamic particle swarm optimization (PSO) algorithm.

[0140] Step 530: Initialize the particle swarm and establish a mixed swarm, a disadvantaged swarm, and a dominant swarm.

[0141] Step 540: Update the particle velocity of the mixed group and add sine and cosine learning factors.

[0142] Step 550: Update the particle positions of the dominant group, determine the step size through Lévy flight, and evaluate through the greedy algorithm.

[0143] Step 560: Update the particle velocity of the inferior group, add mixed particles and mixed learning factors during the update process, and perform adaptive mutation judgment.

[0144] Step 570: Merge populations.

[0145] Step 580: Determine whether the maximum number of iterations has been reached. If not, return to the particle swarm initialization step to continue generating the population.

[0146] Step 590: If the judgment result is yes, then after merging the populations, the characteristic parameter quantity is determined based on the undefined samples (ie, undefined fault data), and the model parameters of the DELM model are optimized by a dynamic multi-population algorithm.

[0147] Step 5100: Compare the predicted value output by the DELM model with the actual fault label of the transformer.

[0148] Step 5110: Based on the comparison results, the secondary diagnosis results output by the DELM model are evaluated.

[0149] This embodiment classifies faults by training a deep extreme learning machine model, and the model parameters are optimized by a dynamic multi-population algorithm, and then the secondary diagnosis model is used to effectively analyze the fault data that is not defined in the preliminary diagnosis results.

[0150] In some embodiments, a quantitative analysis strategy may be employed to determine the insulating oil condition monitoring results of the transformer under test based on the dissolved gas concentration prediction results, the secondary diagnosis results, and the oil withstand voltage value, dielectric loss factor, volume resistivity, and other key test values ​​obtained in advance in the target operating status data. Specifically, the strategy includes:

[0151] Firstly, according to the prediction results of dissolved gas concentration, the three-ratio method is used to perform fault diagnosis and obtain the fault prediction results.

[0152] Then, corresponding evaluation values ​​are assigned to the fault prediction results, secondary diagnosis results, oil withstand voltage value, dielectric loss factor and volume resistivity in the target operating status data, and other key test values ​​including copper ion content, moisture content, furfural content in the oil, and oil oxidation test result values.

[0153] Then, the individual evaluation values ​​are added together to obtain a comprehensive score.

[0154] Finally, based on the comprehensive score, the status level of the insulating oil is determined. The status levels include good, caution, abnormal, severe, etc., thus obtaining the insulating oil status monitoring results.

[0155] In other embodiments, a qualitative analysis strategy can also be employed to determine the insulating oil condition monitoring results of the transformer under test based on the dissolved gas concentration prediction results, the secondary diagnostic results, the oil withstand voltage and dielectric loss factor in the target operating status data, and other key test values. Specifically, a comprehensive judgment can be made based on the specific conditions of each indicator. For example, if the dissolved gas concentration prediction results indicate a potential fault, the oil withstand voltage and dielectric loss factor both exceed the standard values, and the secondary diagnostic results also show an abnormality, then the insulating oil condition can be determined to be serious and immediate action is required.

[0156] This embodiment can more accurately determine the insulating oil condition monitoring results of the transformer under test through a comprehensive analysis of the dissolved gas concentration prediction results, secondary diagnosis results, oil withstand voltage value, dielectric loss factor and volume resistivity, and other key detection values, providing an important basis for transformer maintenance and management.

[0157] In actual applications, the cloud service end can also visualize the monitoring results of the insulating oil status of the transformer under test, allowing management and operation and maintenance personnel at all levels to intuitively view the various test and detection information of the transformer under test, the completion status of the experiment, the change trend of the dissolved gas in the insulating oil, the existing fault types, and inspection reminders through the on-site inspection terminal, thereby realizing real-time monitoring of the operating status of each transformer.

[0158] At the same time, after analyzing the abnormal results of the insulating oil status monitoring, the cloud server can generate alarm prompt information and on-site inspection task work orders. The operation and maintenance personnel can obtain the work order information through the on-site inspection terminal and arrange for oil collection and inspection in a timely manner. After the offline inspection is completed, the inspection results can be sent back to the cloud server.

[0159] In summary, the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration provided by the embodiments of the present invention has at least the following advantages:

[0160] By integrating various experimental data on the transformer's insulating oil under test and leveraging offline oil chromatography data, an effective fault sample dataset can be formed, improving model generalization from the training data. Training a secondary diagnostic model addresses issues such as fuzzy boundary values, missing fault codes, and repeated judgments of undefined samples that plague traditional methods. Appropriate data preprocessing methods improve the imbalance of the original dataset and effectively compensate for the loss of online oil chromatography data. Using a gas concentration prediction model to predict the changing trends of dissolved gases in the insulating oil in stages provides more comprehensive and accurate information for monitoring the transformer's insulating oil condition, improving the operational stability and reliability of the transformer under test.

[0161] Based on the same general inventive concept, the present invention also protects a transformer insulating oil condition monitoring method based on cloud-edge-end collaboration. The transformer insulating oil condition monitoring method based on cloud-edge-end collaboration provided by the present invention is described below. The transformer insulating oil condition monitoring method based on cloud-edge-end collaboration described below and the transformer insulating oil condition monitoring system based on cloud-edge-end collaboration described above can be referenced to each other.

[0162] like Figure 8 As shown, the transformer insulating oil condition monitoring method based on cloud-edge-end collaboration provided by an embodiment of the present invention can be executed by a cloud service end, the cloud service end is connected to the edge end, and the edge end is connected to the device end; the above method mainly includes the following steps:

[0163] Step 610: Send a data detection instruction to the edge end, so that the edge end, in accordance with the data detection instruction, obtains target operating status data from the device end and performs a preliminary diagnosis on the oil chromatogram data in the target operating status data to obtain a preliminary diagnosis result. The target operating status data is obtained from multiple operating status data of the transformer under test, which is collected periodically by the device end based on the voltage level of the transformer under test. The multiple operating status data include oil chromatogram data, oil withstand voltage value, dielectric loss factor, and volume resistivity.

[0164] Step 620: Receive the preliminary diagnosis results and target operating status data uploaded by the edge end.

[0165] Step 630: Input the oil chromatography data in the target operating state data into the pre-built gas concentration prediction model to obtain the prediction result of the dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model.

[0166] Step 640: Perform secondary diagnosis on the fault data not defined in the preliminary diagnosis result to obtain a secondary diagnosis result.

[0167] Step 650: Determine the insulating oil condition monitoring result of the transformer to be tested based on the dissolved gas concentration prediction result, the secondary diagnosis result, the oil withstand voltage value, dielectric loss factor and volume resistivity in the target operating status data, and other key detection values ​​obtained in advance.

[0168] Regarding the method in the above embodiment, the specific manner of each step has been described in detail in the embodiment of the relevant system and will not be elaborated again here.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A transformer insulating oil condition monitoring system based on cloud-edge-end collaboration, characterized in that: include: Device side, edge side, and cloud service side; The cloud service end and the device end are both connected to the edge end; The device end is used to regularly collect multiple operating status data of the transformer under test according to the voltage level of the transformer under test; wherein the multiple operating status data include oil chromatogram data, oil withstand voltage value, dielectric loss factor and volume resistivity; The edge end is used to obtain target operating status data from the device end according to the data detection instruction issued by the cloud service end, perform preliminary diagnosis on the oil chromatography data in the target operating status data, obtain preliminary diagnosis results, and upload the preliminary diagnosis results and target operating status data to the cloud service end; The cloud service end includes: A concentration prediction module is configured to input the oil chromatogram data in the target operating state data into a pre-built gas concentration prediction model to obtain a prediction result of the dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model; wherein the gas concentration prediction model includes: an input layer for receiving the oil chromatogram data; an attention layer for assigning corresponding attention weights to different data elements in the oil chromatogram data; a bidirectional gated recurrent layer for performing bidirectional feature extraction on the oil chromatogram data based on the attention weights to obtain hidden state data; and an output layer for generating and outputting a prediction result of the dissolved gas concentration in the transformer insulating oil based on the hidden state data; A secondary diagnosis module is configured to obtain oil chromatogram samples and corresponding fault code samples of the transformer to be tested; establish a training sample set and a test sample set based on the oil chromatogram samples and the fault code samples; train a pre-built deep extreme learning machine model using the training sample set, and test the trained deep extreme learning machine model using the test sample set to obtain a secondary diagnosis model; input undefined fault data in the preliminary diagnosis result into the secondary diagnosis model to obtain a secondary diagnosis result output by the secondary diagnosis model; The condition monitoring module is used to determine the insulating oil condition monitoring result of the transformer to be tested based on the dissolved gas concentration prediction result, the secondary diagnosis result, the oil withstand voltage value, dielectric loss factor and volume resistivity in the target operating status data, and other key detection values.

2. The transformer insulating oil condition monitoring system based on cloud-edge-end collaboration according to claim 1 is characterized in that: The equipment side includes: oil chromatography detection equipment, insulating oil withstand voltage detection equipment, dielectric loss detection equipment and data sensing equipment; The oil chromatography detection device, the insulating oil withstand voltage detection device, and the dielectric loss detection device are respectively connected to corresponding data sensing devices, and the data sensing devices are connected to the edge end; The oil chromatography detection equipment is used to detect the oil chromatography data of the transformer insulating oil to be tested, the insulating oil withstand voltage detection equipment is used to detect the oil withstand voltage value of the transformer insulating oil to be tested, and the dielectric loss detection equipment is used to detect the dielectric loss factor and volume resistivity of the transformer insulating oil to be tested; The data sensing device is used to receive data detected by a corresponding detection device and send the detected data to the edge end.

3. The transformer insulating oil condition monitoring system based on cloud-edge-end collaboration according to claim 2 is characterized in that: The oil chromatography detection equipment includes: an online chromatography detector installed at the monitoring site and an offline chromatography detector installed in a standard laboratory environment; The online chromatogram detector is used to detect online oil chromatogram data of the transformer insulating oil to be tested, and the offline chromatogram detector is used to perform offline testing on the transformer insulating oil to be tested obtained by regular sampling to obtain offline oil chromatogram data.

4. The transformer insulating oil condition monitoring system based on cloud-edge-end collaboration according to claim 2 is characterized in that: The edge end includes: a first edge industrial computer, a second edge industrial computer, a third edge industrial computer and a fourth edge industrial computer; The first edge industrial computer, the second edge industrial computer, and the third edge industrial computer are respectively connected to their corresponding data sensing devices, the first edge industrial computer, the second edge industrial computer, and the third edge industrial computer are all connected to the fourth edge industrial computer, and the fourth edge industrial computer is connected to the cloud server; The first edge industrial computer is configured to obtain the oil chromatography data in the target operating status data according to the data detection instruction issued by the cloud server, perform a preliminary diagnosis on the oil chromatography data, and obtain a preliminary diagnosis result; The second edge industrial computer is used to obtain the oil pressure resistance value in the target operating status data according to the data detection instruction issued by the cloud server; The third edge industrial computer is used to obtain the dielectric loss factor and volume resistivity in the target operating status data according to the data detection instruction issued by the cloud server; The fourth edge industrial computer is used to use the preliminary diagnostic results, oil chromatography data, oil withstand voltage value, dielectric loss factor and volume resistivity as table content to generate a data collection report, and upload the data collection report to the cloud service end.

5. The transformer insulating oil condition monitoring system based on cloud-edge-end collaboration according to claim 1 is characterized in that: Based on the oil chromatogram sample and the fault code sample, a training sample set and a test sample set are established, including: constructing an input feature vector according to the oil chromatogram sample; Inputting the input feature vector into a pre-built data enhancement model to obtain an oil chromatogram amplification sample output by the data enhancement model; Establishing a sample data set based on the oil chromatogram amplification sample and the fault code sample; The sample data set is divided according to a preset ratio to obtain a training sample set and a test sample set.

6. The transformer insulating oil condition monitoring system based on cloud-edge-end collaboration according to claim 5 is characterized in that: According to the oil chromatogram sample, an input feature vector is constructed, including: performing normalization processing on the oil chromatogram sample to obtain oil chromatogram normalized data; extracting characteristic gas content data from the oil chromatogram sample; constructing a gas ratio graph based on the characteristic gas content data to obtain gas graphic data; Performing three-ratio encoding on the characteristic gas content data to obtain three-ratio encoded data; An input feature vector is constructed according to the oil chromatogram normalized data, the gas graph data, and the three-ratio encoded data.

7. A transformer insulating oil condition monitoring method based on cloud-edge-end collaboration, characterized in that: The method is executed by a cloud service end, the cloud service end is connected to an edge end, and the edge end is connected to a device end; the method includes: Sending a data detection instruction to the edge end, so that the edge end obtains target operating status data from the device end according to the data detection instruction, and performs a preliminary diagnosis on the oil chromatogram data in the target operating status data to obtain a preliminary diagnosis result; wherein the target operating status data is obtained by regularly collecting multiple operating status data of the transformer to be tested from the device end according to the voltage level of the transformer to be tested; the multiple operating status data include oil chromatogram data, oil withstand voltage value, dielectric loss factor, and volume resistivity; Receive preliminary diagnostic results and target operating status data uploaded by the edge; The oil chromatogram data in the target operating state data is input into a pre-built gas concentration prediction model to obtain a prediction result of the dissolved gas concentration in the transformer insulating oil output by the gas concentration prediction model; wherein the gas concentration prediction model includes: an input layer for receiving the oil chromatogram data; an attention layer for assigning corresponding attention weights to different data elements in the oil chromatogram data; a bidirectional gated recurrent layer for performing bidirectional feature extraction on the oil chromatogram data according to the attention weights to obtain hidden state data; and an output layer for generating and outputting a prediction result of the dissolved gas concentration in the transformer insulating oil according to the hidden state data; Obtaining an oil chromatogram sample and a corresponding fault code sample of the transformer to be tested; establishing a training sample set and a test sample set based on the oil chromatogram sample and the fault code sample; training a pre-built deep extreme learning machine model using the training sample set, and testing the trained deep extreme learning machine model using the test sample set to obtain a secondary diagnosis model; inputting undefined fault data in the preliminary diagnosis result into the secondary diagnosis model to obtain a secondary diagnosis result output by the secondary diagnosis model; The insulating oil status monitoring result of the transformer to be tested is determined based on the dissolved gas concentration prediction result, the secondary diagnosis result, the oil withstand voltage value, dielectric loss factor and volume resistivity in the target operating status data, and other key detection values ​​obtained in advance.

Images