Method for evaluating insulation state of transformer based on multi-gas cooperative detection

Abstract

The present application relates to the technical field of transformer insulation state evaluation, and particularly discloses a transformer insulation state evaluation method based on multi-gas collaborative detection, which takes multi-gas collaborative online monitoring as the main line, realizes real gas dynamic collaborative linkage, accurately distinguishes fault types through multi-gas component matching and proportional correlation, dynamically tracks and analyzes the production rate and content change trend of characteristic gases, constructs a precise insulation state evaluation system, realizes early warning, accurate diagnosis and trend prediction of insulation faults, and guarantees the safe and stable operation of transformers and power systems.

Description

Technical Field

[0001] This application relates to the field of transformer insulation condition assessment technology, and specifically discloses a transformer insulation condition assessment method based on multi-gas collaborative detection. Background Technology

[0002] Power transformers are core and critical equipment in power systems, responsible for energy transmission and voltage transformation. Their operational reliability directly determines the safety, stability, and power supply quality of the entire power system. Insulation faults in transformers can lead to serious accidents such as winding burnout and tank explosions, causing widespread power outages, huge economic losses, and severe social impacts. The transformer's insulation system is crucial for its long-term stable operation. Currently, the vast majority of transformers in power systems use oil-paper insulation structures, employing mineral oil or synthetic oil as the insulating medium and cellulose-based insulating paper or paperboard as the solid insulating material. The aging and deterioration of the oil-paper insulation system is a key factor determining the transformer's service life and operating condition.

[0003] Practical research shows that during long-term operation, transformer oil-paper insulation systems are subjected to a combination of factors, including electrical stress, thermal stress, mechanical stress, and environmental humidity, leading to slow aging or sudden deterioration. During this process, the molecular chains of the oil paper break and decompose, producing various characteristic gases. These gases mainly include hydrogen, carbon monoxide, and hydrocarbon gases such as methane, ethane, ethylene, and acetylene. Different types and severity of insulation faults correspond to significant differences in the composition, content, and generation rate of these characteristic gases. This indicates a clear "fingerprint relationship" between insulation faults and characteristic gases, which forms the core theoretical basis for assessing the insulation condition of transformers through gas detection.

[0004] Currently, the main method for detecting gas in transformer insulation is offline dissolved gas analysis (DGA). This method involves periodically sampling the transformer insulating oil and sending it to the laboratory for testing of characteristic gas components and their contents. Combined with traditional criteria such as the three-ratio method, it enables qualitative judgment of insulation faults. However, this offline detection method has many inherent defects and cannot meet the power system's demand for real-time and accurate monitoring of transformer insulation status: First, the detection has a significant lag. The cycle of offline sampling and laboratory analysis is usually several days to several weeks, which cannot capture the rapid changes in characteristic gases in the early stage of insulation faults. This often leads to the fault being discovered when it has already developed to a more serious stage, missing the best time for treatment. Second, the sampling and detection process is susceptible to interference. Insulating oil contamination may occur during sampling, characteristic gases may be lost during oil sample transportation and storage, and human error in laboratory testing may also affect the accuracy of the data, leading to misjudgment or missed faults. Third, it cannot achieve multi-gas collaborative dynamic monitoring. Offline detection can only obtain gas composition and content data from a single detection, and cannot track the changing trend of characteristic gas production rate, making it difficult to distinguish between normal insulation aging and sudden faults, and also unable to accurately judge the development trend of faults.

[0005] As power systems develop towards higher voltage, larger capacity, and greater intelligence, the operating environment of transformers is becoming increasingly complex, and the requirements for the accuracy and real-time performance of insulation condition assessment are constantly increasing. The limitations of traditional offline detection methods are becoming increasingly apparent. Some existing technologies have online detection devices for single gases or a few gases, but these devices only monitor one characteristic gas and do not achieve coordinated detection of multiple characteristic gases. They cannot fully utilize the correspondence between different characteristic gases and insulation faults, and the changes in the content of a single gas cannot comprehensively reflect the true state of the insulation system. Incomplete gas detection can easily lead to misjudgment of fault types and biased assessment of fault severity. For example, detecting only CO cannot distinguish between normal aging of insulation paper and low-temperature overheating faults, and detecting only H2 cannot distinguish between partial discharge and arc discharge faults.

[0006] In addition, most existing online detection technologies only focus on the detection of characteristic gas content, lacking collaborative analysis of the ratio of multiple gas components and gas generation rate, making it difficult to achieve quantitative assessment of insulation status and unable to provide maintenance personnel with accurate fault location, severity judgment and maintenance decision-making basis.

[0007] Currently, multi-gas collaborative online monitoring is the mainstream approach, but it still has shortcomings: First, the "synergy" of multi-gas collaborative monitoring is insufficient, often consisting of simple superposition of single-gas detections without achieving true dynamic collaborative linkage. Most multi-gas detection solutions merely integrate multiple single-gas sensors to collect data on the content of H2, CO, and various hydrocarbon gases, then analyze the data independently or simply piece them together, without establishing a correlation analysis mechanism between the multi-gas components. For example, while array-type gas sensors are used to construct multi-gas detection systems, this technology is not yet mature and does not fully utilize the "fingerprint correspondence" between different characteristic gases and insulation faults. It cannot accurately distinguish fault types through component matching and proportional correlation of multiple gases. Compared to single-gas detection, it does not demonstrate the core advantages of collaborative monitoring and still carries the risk of false or missed fault detection.

[0008] Second, the dynamic monitoring capability is weak, making it difficult to accurately capture the dynamic evolution of insulation faults. Existing research focuses on the static content detection of multiple gases, lacking dynamic tracking and analysis of the gas generation rate and content change trends of characteristic gases, thus failing to achieve a closed loop of real-time monitoring, dynamic analysis, and trend prediction. Although some solutions attempt to introduce the concept of dynamic monitoring, they simply set a fixed time interval sampling period without designing an adaptive sampling mechanism based on the evolution law of insulation faults. Either the sampling frequency is too low to capture the rapid changes in characteristic gases in the early stage of insulation faults, or the sampling frequency is too high, leading to data redundancy and increased energy consumption. Furthermore, without establishing a quantitative correlation model between gas generation rate and fault development degree, it is difficult to distinguish the gas generation differences between normal insulation aging and sudden faults, thus failing to provide reliable support for fault trend prediction.

[0009] Third, the detection accuracy and environmental adaptability are insufficient, making it difficult to meet the needs of complex field operations. Existing multi-gas collaborative detection schemes still have significant limitations in their detection technologies: schemes using gas chromatography are limited by carrier gas, resulting in insufficient real-time performance; schemes using photoacoustic spectroscopy have high requirements for the operating environment and cannot adapt to the complex field operating conditions of transformers; schemes using array-type gas sensors suffer from poor measurement sensitivity, accuracy, and data repeatability, and are easily affected by environmental humidity, temperature, and other interference factors, leading to distorted detection data and affecting the accuracy of insulation condition assessment. Furthermore, existing technologies mostly focus on laboratory simulations or small-scale experimental verification, lacking long-term field testing and verification under the complex operating conditions of actual substations, resulting in weak engineering practicality and difficulty in direct promotion and application.

[0010] Fourth, the ability to fuse and analyze multi-gas collaborative data is lacking, resulting in insufficient accuracy of the assessment model. Most existing patents and papers only perform simple threshold judgments on multi-gas content data, lacking multi-dimensional fusion analysis of multi-gas component ratios, gas generation rates, and historical data. They also fail to incorporate artificial intelligence algorithms and big data analysis to construct accurate insulation condition assessment models. While some studies have attempted to introduce pattern recognition technologies such as BP neural networks and grey theory, these are mostly in the offline training and verification stage, failing to achieve dynamic linkage with real-time multi-gas monitoring data. This prevents the model from adaptively optimizing assessment parameters based on real-time monitoring data, leading to low accuracy in assessment results. Consequently, it is difficult to achieve early and accurate warnings of insulation faults and quantitative determination of fault severity, and it cannot provide maintenance personnel with targeted maintenance decision-making support, falling far short of the needs of intelligent power system development.

[0011] In summary, existing patents and papers on multi-gas collaborative dynamic monitoring of transformers have not effectively solved core problems such as insufficient synergy, weak dynamic tracking capability, poor detection accuracy and environmental adaptability, and insufficient data fusion analysis. They cannot achieve efficient collaborative real-time monitoring of multiple characteristic gases, nor can they build an accurate and reliable transformer insulation condition assessment system.

[0012] In view of the above-mentioned problems, the inventors have provided a method for evaluating the insulation status of transformers based on multi-gas collaborative detection in order to solve the above problems. Summary of the Invention

[0013] The purpose of this invention is to solve one or more problems raised in the background art.

[0014] To achieve the above objectives, the present invention provides the following basic solution: The transformer insulation condition assessment method based on multi-gas collaborative detection includes the following steps: Step S1: Build a multi-gas collaborative online acquisition system. This system integrates an adaptive characteristic gas sensor array to obtain insulating oil samples from the transformer body and separate the characteristic gases dissolved in the oil. It synchronously collects real-time content data of each gas, and the acquisition frequency is adaptively adjusted according to the insulation status. During the acquisition process, environmental interference is eliminated, and the raw data is processed by the data preprocessing unit, and abnormal data is removed. Step S2: Establish a fault gas production fingerprint rule base. Using the real-time data preprocessed in step S1, match the content data of each gas with the preset fault gas production fingerprint rule base to obtain a preliminary fault type determination result. Introduce the ratio method to correct the preliminary determination result through the ratio range, and then use the weighted voting method to obtain the final fault type determination result. Step S3: Calculate the instantaneous gas production rate and average gas production rate of each characteristic gas in real time, then perform trend fitting on the gas content change data to obtain the content change trend curve. Based on the change trend curve and gas production rate data, predict the change trend of gas content and gas production rate, as well as the direction and severity of fault development in the future. When the preset warning threshold is reached, trigger the corresponding level of warning. Step S4: Integrate the fault type determination results from Step S2 and the gas production rate and trend analysis results from Step S3, and combine them with historical transformer monitoring data and operating conditions to construct a multi-dimensional evaluation index system; construct a fusion evaluation model, input real-time data of each evaluation index into the fusion evaluation model, output quantitative evaluation results of insulation status, and then output targeted operation and maintenance decision suggestions based on the evaluation results.

[0015] Furthermore, it also includes step S12, which is as follows: The characteristic gases are divided into an electrical fault-related group, a thermal fault-related group, and an insulation aging-related group. Through real-time data linkage between the acquisition module and the analysis module, the gases within each group are automatically increased when the content of any gas in a certain group shows abnormal fluctuations. In step S12, the electrical fault-related group includes... , and The thermal fault association group includes , and The insulation aging association group includes , , The feature gas sensor array in step S1 is a high-precision sensor array, which includes a hydrogen gas sensor, a carbon monoxide gas sensor, and a hydrocarbon gas sensor.

[0016] Furthermore, in step S2, the fault gas generation fingerprint rule base includes the characteristic gas component range of partial discharge, arc discharge, low temperature overheating, medium and high temperature overheating, insulation aging, and insulation moisture. The characteristic gas component range needs to be dynamically updated in combination with previous experimental data and on-site operation and maintenance experience. The previous experimental data refers to the simulated fault gas generation experimental data.

[0017] Furthermore, in step S2, the ratio method is specifically implemented as follows: The ratio method is introduced, including... / , / , / and Using the gas content data collected in step S1, the above four ratios are calculated respectively. Each ratio corresponds to a fault judgment result at the ratio level. The judgment result at the ratio level is compared with the preliminary judgment result of the fault type. If the two are consistent, the preliminary judgment result is reliable; if the two are inconsistent, the judgment result at the ratio level takes precedence and the preliminary judgment result is corrected. The specific operation of the weighted voting method is as follows: the four ratios correspond to four voting subjects. The weights are set according to the correlation between each gas and the corresponding fault type. Each voting subject votes for the corresponding fault type according to its own results. The total number of votes is calculated according to the weights. The fault type with the most votes is the final judgment result of the fault type.

[0018] Furthermore, in step S3, based on the severity of the predicted gas production rate and content change trends, corresponding warning levels are triggered: Mild warning: The average gas production rate exceeds the normal threshold, but the trend coefficient is small, and it is predicted that it will not reach the moderate abnormality threshold within the next T time period; Moderate warning: The average gas production rate reaches the upper limit of mild abnormality, and the trend coefficient shows an upward trend, and it is predicted that it will approach the severe abnormality threshold within the next T time period; Severe warning: The average gas production rate exceeds the moderate abnormality threshold, and the trend coefficient is large, and it is predicted that it will reach the fault danger threshold within the next T time period. When any group of gases triggers a warning, the monitoring frequency of the associated group of gases is simultaneously linked to ensure full tracking of the fault development.

[0019] Furthermore, in step S4, each indicator in the multi-dimensional evaluation index system is set with four levels of thresholds: normal, slightly abnormal, moderately abnormal, and severely abnormal; the fusion evaluation model uses real-time monitoring data to dynamically optimize parameters, thereby improving its adaptability and evaluation accuracy.

[0020] Furthermore, the operating conditions include the transformer's load, voltage, and ambient temperature; the operation and maintenance decision recommendations include normal operation and maintenance, enhanced monitoring, and shutdown for maintenance.

[0021] The principle and effect of this solution are as follows: 1. Compared with existing technologies, it achieves true dynamic collaborative linkage of multiple gases, completely breaking through the limitations of simple superposition of existing single gas detection: it realizes synchronous gas acquisition and linkage analysis within each group, as well as mutual verification and supplementation of data between groups. At the same time, it builds a real-time data linkage mechanism between the acquisition module and the analysis module. When a gas in a certain group shows abnormal fluctuations, it automatically triggers the increase of sampling frequency and detection accuracy of that group and related groups, truly giving full play to the core advantages of multi-gas collaborative monitoring, solving the pain point of insufficient collaboration in existing technologies, ensuring that the gas change characteristics in the early stage of faults can be quickly captured under abnormal operating conditions, and reducing the risk of missed fault detection.

[0022] 2. Compared with existing technologies, the multi-dimensional integrated assessment is accurate and reliable, enabling quantitative assessment of insulation status and targeted operation and maintenance decision output: It achieves quantitative assessment of insulation status and combines the insulation status assessment results, fault types and development trends to output targeted operation and maintenance decisions such as normal operation and maintenance, enhanced monitoring, and shutdown for maintenance, providing accurate decision support for operation and maintenance personnel, reducing operation and maintenance costs and improving operation and maintenance efficiency.

[0023] 3. Compared with existing technologies, the accuracy of fault type differentiation is significantly improved, effectively avoiding misjudgment and omission: By correcting the component matching deviation through the core gas ratio, and then achieving comprehensive judgment through the weighted voting method, it can accurately distinguish different types of faults such as electrical faults, thermal faults, insulation aging / dampness, etc., and can also distinguish different degrees of severity of the same fault.

[0024] 4. Compared with existing technologies, this invention, through the above-mentioned technical improvements, effectively solves the core technical problems of insufficient multi-gas detection coordination, inaccurate fault judgment, weak dynamic tracking, low evaluation accuracy, and poor environmental adaptability in existing transformers. It realizes a closed loop of multi-gas collaborative acquisition, dynamic linkage analysis, accurate fault judgment, trend prediction, quantitative evaluation, and operation and maintenance decision-making, significantly improving the accuracy, real-time performance, and reliability of transformer insulation condition monitoring and evaluation, extending the service life of transformers, reducing the failure rate, and providing strong protection for the safe and stable operation of the power system. It has good economic and social benefits and broad prospects for promotion and application. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A schematic flowchart of the transformer insulation condition assessment method based on multi-gas collaborative detection proposed in this application is shown. Detailed Implementation

[0027] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0028] Implementation, for example Figure 1 As shown: The transformer insulation condition assessment method based on multi-gas collaborative detection includes the following steps: Step S1: Build a multi-gas collaborative online acquisition system. This system integrates an adaptive characteristic gas sensor array to obtain insulating oil samples from the transformer body and separate the characteristic gases dissolved in the oil. It synchronously collects real-time content data of each gas, and the acquisition frequency is adaptively adjusted according to the insulation status. During the acquisition process, environmental interference is eliminated, and the raw data is processed by the data preprocessing unit, and abnormal data is removed. Regarding obtaining insulating oil samples from the transformer body and separating the characteristic gases dissolved in the oil, the specifics are as follows: The extracted insulating oil sample is fed into a specially designed gas extractor built into the acquisition system. An oil-resistant and high-temperature-resistant polymer separation membrane is laid on one side of the oil chamber. This membrane only allows characteristic gases such as hydrogen, carbon monoxide, and hydrocarbons to permeate through, while blocking insulating oil molecules from passing through. The dual circulation design of oil and gas paths increases the oil-gas contact area and accelerates the gas permeation rate, enabling the dissolved gas in the oil to reach permeation equilibrium with the oil phase in a short time, ensuring that the separation efficiency meets the requirements of real-time monitoring. The characteristic gas that permeates through the membrane enters the gas buffer chamber to complete the separation from the insulating oil sample. The separated insulating oil is pumped back to the transformer body through a closed-loop pipeline, avoiding oil sample waste and environmental pollution. In this way, the characteristic gas can be obtained through the characteristic gas sensor array.

[0029] Regarding the characteristic gas sensor array: it also includes step S12, which is as follows: the characteristic gases are divided into electrical fault associated group, thermal fault associated group and insulation aging associated group. The gases in each group are linked by real-time data from the acquisition module and the analysis module. When the content of any gas in a certain group fluctuates abnormally, the sampling frequency of the gas in that group and the associated group is automatically increased, so as to realize the adaptive adjustment of the acquisition frequency according to the insulation status.

[0030] Specifically: Normal insulation status: When the content of each characteristic gas collected in step S1 is within the preset normal threshold, for example... ≤200μL / L When the concentration of gas is 0 μL / L and there are no abnormal fluctuations in gas content, a conventional sampling frequency should be used, such as once every 2 hours. This can be adjusted freely by the staff to meet the daily monitoring needs, thereby reducing system energy consumption and data redundancy.

[0031] Suspected abnormal insulation condition: When the content of any gas in a certain group approaches the normal threshold, such as... =180μL / L, close to the normal threshold of 200μL / L, or with slight fluctuations, automatically triggers an increase in the sampling frequency of the gas in this group and related groups, automatically increases the sampling frequency, strengthens monitoring, captures subtle changes in the gas, and avoids missing minor anomalies in the early stages.

[0032] Clearly abnormal insulation condition: When the content of any gas within a group exceeds the normal threshold, such as... When the concentration is ≥500μL / L, and an insulation abnormality or fault is clearly determined, the sampling frequency of the gas in this group and related groups is automatically increased, which further significantly increases the acquisition frequency and enhances the detection accuracy. It can quickly track the dynamic changes in gas content and gas production rate, providing high-frequency and accurate data support for fault type determination and trend analysis.

[0033] In step S12, the electrical fault association group includes , and The thermal fault association group includes , and The insulation aging association group includes , , The feature gas sensor array in step S1 is a high-precision sensor array, which includes a hydrogen gas sensor, a carbon monoxide gas sensor, and a hydrocarbon gas sensor.

[0034] Step S2: Establish a fault gas generation fingerprint rule base. Using the real-time data preprocessed in step S1, match the content data of each gas with the preset fault gas generation fingerprint rule base to obtain a preliminary fault type determination result. Introduce the ratio method to correct the preliminary determination result through the ratio range, and then use the weighted voting method to obtain the final fault type determination result.

[0035] In step S2, the fault gas generation fingerprint rule base includes the characteristic gas component range of partial discharge, arc discharge, low temperature overheating, medium and high temperature overheating, insulation aging and insulation moisture. The characteristic gas component range needs to be dynamically updated in combination with previous experimental data and on-site operation and maintenance experience. The previous experimental data refers to the simulated fault gas generation experimental data.

[0036] In step S2, the ratio method is specifically implemented as follows: The ratio method is introduced, including... / , / , / and Using the gas content data collected in step S1, the above four ratios are calculated respectively. Each ratio corresponds to a fault judgment result at the ratio level. The judgment result at the ratio level is compared with the preliminary judgment result of the fault type. If the two are consistent, the preliminary judgment result is reliable; if the two are inconsistent, the judgment result at the ratio level is taken as the main factor, and the preliminary judgment result is corrected.

[0037] Details are as follows: / , / , / and The four ratios correspond to the energy level of electrical faults, the temperature level of thermal faults, and the aging and moisture absorption of insulating paper, respectively. Each ratio has a preset fault range, for example... / >2.5 corresponds to high-intensity arc discharge; <0.2 corresponds to normal aging of the insulating paper. The four calculated ratios are compared with the preset fault range to obtain a fault judgment result at the ratio level.

[0038] Then, the results of the ratio-level judgment are compared with the preliminary judgment results of the component matching. If they are consistent, the preliminary judgment result is reliable; if they are inconsistent, for example, the preliminary judgment of component matching indicates insulation aging, but... If the ratio indicates that the insulation is damp, then the ratio analysis result should be taken as the main factor to correct the preliminary judgment result and correct it to damp insulation. The ratio method is used to eliminate the bias of a single content judgment.

[0039] The weighted voting method operates as follows: four ratios correspond to four voting entities, and weights are assigned based on the strength of the correlation between each gas and its corresponding fault type. For example... It is a hallmark gas of electric arc discharge, therefore / The ratio corresponding to the highest weight in the judgment result is CO / CO2, which only corresponds to insulation aging / dampness and has a lower weight than the ratios related to electrical faults and thermal faults. Each voting entity votes for the corresponding fault type based on its own results. The total number of votes is calculated according to the weights, and the fault type with the highest number of votes is the final judgment result of the fault type. That is, after the four ratios are divided according to their weights, a ratio with the largest weight will be obtained, and the result corresponding to this ratio is the final judgment result of the fault type.

[0040] Step S3: Calculate the instantaneous gas production rate and average gas production rate of each characteristic gas in real time, and then perform trend fitting on the gas content change data to obtain the content change trend curve. Based on the change trend curve and gas production rate data, predict the change trend of gas content and gas production rate, as well as the direction and severity of fault development in the future. When the preset warning threshold is reached, trigger the corresponding level of warning.

[0041] Specifically: Regarding the instantaneous and average gas production rates: The sliding window method is used to calculate the instantaneous and average gas production rates of each characteristic gas in real time, eliminating the impact of fluctuations at a single data point on the calculation results, balancing real-time performance and accuracy. The specific algorithm is as follows: First, set up a sliding window: Let the size of the sliding window be N, where N is a positive integer that is adaptively adjusted according to the sampling frequency. The higher the sampling frequency, the larger the value of N. Set it to N∈[3,10]. The window sliding step size is 1, that is, for each new set of gas content data collected, the window slides forward 1 unit, discarding the earliest set of data and keeping the latest N sets of data for calculation.

[0042] Secondly, for any characteristic gas i, i represents , , , , Let the gas content collected at time k be any one of the following: The gas content collected at time k-1 is The time interval between the two data collections is Then the instantaneous gas production rate at time k is as follows: .

[0043] Next, the average gas production rate is calculated. Based on N sets of instantaneous gas production rate data within the sliding window, the average gas production rate within the window is calculated. , .

[0044] A combination of linear regression and exponential fitting was used to fit the trend of the content change data of characteristic gas i, resulting in a content change trend curve. Based on the coefficients of the content change trend curve and the average gas production rate... The coordinated variation trend of gas production rates in each group was analyzed.

[0045] If the trend coefficient > 0 and ≤0.1μL / (L·d) is considered normal insulation aging, with a stable fault development trend. If the trend coefficient is >0 and 0.1μL / (L·d) < ≤1μL / (L·d) is considered a mild anomaly, indicating a slow fault development trend. If the trend coefficient is >0 and A concentration >1 μL / (L·d) is considered a severe anomaly, indicating a rapidly deteriorating fault. If the trend coefficient is ≤0 and the coordinated changes in each gas group are stable, it is considered that there is no obvious fault and the gas content is in a stable state. Then... Based on the predicted trend coefficient and gas production rate data, and referring to the fault development pattern, the direction and severity of the fault are predicted. According to the severity of the predicted gas production rate and content change trend, the corresponding level of warning is triggered. Mild warning: The average gas production rate exceeds the normal threshold, but the trend coefficient is small, and it is predicted that it will not reach the moderate abnormal threshold within the next T time period.

[0046] Moderate warning: The average gas production rate has reached the upper limit of mild anomaly, and the trend coefficient is showing an upward trend. It is predicted that it will approach the threshold of severe anomaly within the next T time period.

[0047] Severe warning: The average gas production rate exceeds the moderate anomaly threshold and the trend coefficient is large. It is predicted that the fault danger threshold will be reached within the next T time period. When any group of gases triggers the warning, the monitoring frequency of the associated gas group will be linked simultaneously to ensure full tracking of the fault development.

[0048] Step S4: Integrate the fault type determination results from Step S2 and the gas production rate and trend analysis results from Step S3, and combine them with historical transformer monitoring data and operating conditions to construct a multi-dimensional evaluation index system; construct a fusion evaluation model, input real-time data of each evaluation index into the fusion evaluation model, output quantitative evaluation results of insulation status, and then output targeted operation and maintenance decision suggestions based on the evaluation results.

[0049] Specifically: In step S4, each indicator in the multi-dimensional evaluation index system is set with four levels of thresholds: normal, slightly abnormal, moderately abnormal, and severely abnormal; the fusion evaluation model uses real-time monitoring data to dynamically optimize parameters, improving adaptability and evaluation accuracy; the operating conditions include transformer load, voltage, and ambient temperature; the operation and maintenance decision suggestions include normal operation and maintenance, enhanced monitoring, and shutdown for maintenance.

[0050] Details are as follows: First, the fault type determination results from step S2 and the gas production rate and trend analysis results from step S3 are integrated through normalization processing, such as average rate, instantaneous rate, content change trend coefficient, historical insulation status assessment results, historical fault records and handling status.

[0051] Then, based on the normalized and integrated data mentioned above, a multi-dimensional evaluation index system is constructed. Each index is assigned a four-level threshold: normal, mildly abnormal, moderately abnormal, and severely abnormal. The scoring criteria for each index are then defined, meaning that each data point will receive a scoring index.

[0052] Next, a fusion evaluation model is constructed. The essence of the fusion evaluation model is a neural network algorithm. Real-time data of each evaluation index is input into the fusion evaluation model, and the quantitative evaluation result of the insulation state is output.

[0053] This result will fall into one of the four levels of classification thresholds: normal, slightly abnormal, moderately abnormal, and severely abnormal. At this point, the insulation status of the transformer will be classified into four levels: normal, slightly abnormal, moderately abnormal, and severely abnormal.

[0054] Reasonable suggestions will be given based on the range in which the transformer insulation condition falls.

[0055] For example, if the insulation status is normal: the corresponding fault type is "no obvious fault" or "normal insulation aging", the fault development trend is stable, and the gas production rate and content change trend is stable; operation and maintenance decision recommendation: normal operation and maintenance, maintain the regular sampling frequency, periodically review and evaluate the results, no additional shutdown maintenance is required, just keep daily inspection records.

[0056] The insulation condition is slightly abnormal: the corresponding fault types are "slight partial discharge", "low temperature overheating" and "slight insulation dampness". The fault development is slow and the gas production rate is within the slightly abnormal range. There is no severe warning. Operation and maintenance decision recommendation: strengthen monitoring, increase the sampling frequency of the corresponding fault-related group to once every 30 minutes, closely track the gas production rate and content change trend, and output an assessment report once a week. No shutdown is required. Auxiliary protection measures can be taken under normal transformer operation.

[0057] The insulation condition is moderately abnormal: the corresponding fault types are "moderate partial discharge", "medium-high temperature overheating" and "moderate insulation dampness". The fault development trend is upward, and the gas production rate exceeds the range of mild abnormality, triggering a moderate warning. Operation and maintenance decision recommendation: Plan to shut down for maintenance, and develop a targeted maintenance plan based on the fault type. For example, for partial discharge faults, check for winding insulation damage; for overheating faults, check for poor contact; for dampness faults, dry the insulating oil. Before maintenance, collect multi-gas data every 10 minutes to track the fault development. After maintenance, restart the system and monitor it continuously for 24 hours. Only after confirming that the insulation condition has recovered to "healthy" or "mild aging" can it be put back into operation.

[0058] The insulation condition is severely abnormal: the corresponding fault types are "high-intensity arc discharge," "high-temperature overheating," and "severe insulation dampness." The fault is rapidly deteriorating, with the gas production rate significantly exceeding the normal range, triggering a severe warning. It is predicted that the fault will further intensify within the next 24 hours. Maintenance recommendations: Immediately shut down for repairs; continued operation is strictly prohibited to avoid serious accidents such as winding burnout or tank explosion. During repairs, focus on identifying the fault source; for example, for arc discharge, check for inter-turn short circuits; for high-temperature overheating, check for core overheating. After repairs, conduct insulation tests and multi-gas collaborative detection verification to ensure the insulation condition returns to normal. Only after passing the review and evaluation can the system be put back into operation. For the following month, maintain the sampling frequency at once every 10 minutes to strengthen monitoring.

[0059] Compared with existing transformer insulation condition gas detection and evaluation technologies, this invention can effectively solve the core defects of existing technologies such as insufficient synergy, inaccurate fault judgment, weak dynamic tracking, and low evaluation accuracy, thereby improving the intelligence and accuracy of transformer insulation monitoring and ensuring the safe and stable operation of the power system.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for assessing the insulation condition of transformers based on multi-gas synergistic detection, characterized in that, Includes the following steps: Step S1: Build a multi-gas collaborative online acquisition system. This system integrates an adaptive characteristic gas sensor array to obtain insulating oil samples from the transformer body and separate the characteristic gases dissolved in the oil. It synchronously collects real-time content data of each gas, and the acquisition frequency is adaptively adjusted according to the insulation status. During the acquisition process, environmental interference is eliminated, and the raw data is processed by the data preprocessing unit, and abnormal data is removed. Step S2: Establish a fault gas production fingerprint rule base. Using the real-time data preprocessed in step S1, match the content data of each gas with the preset fault gas production fingerprint rule base to obtain a preliminary fault type determination result. Introduce the ratio method to correct the preliminary determination result through the ratio range, and then use the weighted voting method to obtain the final fault type determination result. Step S3: Calculate the instantaneous gas production rate and average gas production rate of each characteristic gas in real time, then perform trend fitting on the gas content change data to obtain the content change trend curve. Based on the change trend curve and gas production rate data, predict the change trend of gas content and gas production rate, as well as the direction and severity of fault development in the future. When the preset warning threshold is reached, trigger the corresponding level of warning. Step S4: Integrate the fault type determination results from Step S2 and the gas production rate and trend analysis results from Step S3, and combine them with historical transformer monitoring data and operating conditions to construct a multi-dimensional evaluation index system. A fusion assessment model is constructed, and real-time data of each assessment indicator is input into the fusion assessment model to output quantitative assessment results of insulation status. Based on the assessment results, targeted operation and maintenance decision suggestions are then output.

2. The transformer insulation condition assessment method based on multi-gas synergistic detection according to claim 1, characterized in that, The process also includes step S12, which is as follows: The characteristic gases are divided into an electrical fault association group, a thermal fault association group, and an insulation aging association group. Real-time data from the acquisition module and the analysis module are linked to the gases within each group. When the content of any gas in a group shows abnormal fluctuations, the sampling frequency of that group and its associated groups is automatically increased. In step S12, the electrical fault association group includes... , and The thermal fault association group includes , and The insulation aging association group includes , , The feature gas sensor array in step S1 is a high-precision sensor array, which includes a hydrogen gas sensor, a carbon monoxide gas sensor, and a hydrocarbon gas sensor.

3. The transformer insulation condition assessment method based on multi-gas synergistic detection according to claim 2, characterized in that, In step S2, the fault gas generation fingerprint rule base includes the characteristic gas component range of partial discharge, arc discharge, low temperature overheating, medium and high temperature overheating, insulation aging and insulation moisture. The characteristic gas component range needs to be dynamically updated in combination with previous experimental data and on-site operation and maintenance experience. The previous experimental data refers to the simulated fault gas generation experimental data.

4. The transformer insulation condition assessment method based on multi-gas synergistic detection according to claim 3, characterized in that, In step S2, the ratio method is specifically implemented as follows: The ratio method is introduced, including... / , / , / and Using the gas content data collected in step S1, the above four ratios are calculated respectively. Each ratio corresponds to a fault judgment result at the ratio level. The judgment result at the ratio level is compared with the preliminary judgment result of the fault type. If the two are consistent, the preliminary judgment result is reliable. If the two are inconsistent, the judgment result at the ratio level shall prevail, and the preliminary judgment result shall be corrected. The specific operation of the weighted voting method is as follows: four ratios correspond to four voting subjects. The weights are set according to the correlation between each gas and the corresponding fault type. Each voting subject votes for the corresponding fault type according to its own results. The total number of votes is calculated according to the weights. The fault type with the most votes is the final judgment result of the fault type.

5. The transformer insulation condition assessment method based on multi-gas synergistic detection according to claim 4, characterized in that, In step S3, based on the severity of the predicted gas production rate and content change trend, a corresponding level of warning is triggered: Mild warning: The average gas production rate exceeds the normal threshold, but the trend coefficient is small, and it is predicted that it will not reach the moderate abnormal threshold within the next T time period. Moderate warning: The average gas production rate has reached the upper limit of mild anomaly and the trend coefficient is showing an upward trend. It is predicted that it will approach the threshold of severe anomaly in the next T time period. Severe warning: The average gas production rate exceeds the moderate anomaly threshold and the trend coefficient is large. It is predicted that the fault danger threshold will be reached within the next T time period. When any group of gases triggers the warning, the monitoring frequency of the associated gas group will be linked simultaneously to ensure full tracking of the fault development.

6. The transformer insulation condition assessment method based on multi-gas synergistic detection according to claim 5, characterized in that, In step S4, each indicator in the multi-dimensional evaluation index system is set with four levels of thresholds: normal, mildly abnormal, moderately abnormal, and severely abnormal. The fusion evaluation model uses real-time monitoring data to dynamically optimize parameters, thereby improving its adaptability and evaluation accuracy.

7. The transformer insulation condition assessment method based on multi-gas synergistic detection according to claim 6, characterized in that, The operating conditions include the transformer's load, voltage, and ambient temperature; the operation and maintenance decision recommendations include normal operation and maintenance, enhanced monitoring, and shutdown for maintenance.

Images